]>
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 VB |
121 | #ifdef CONFIG_SLUB_DEBUG |
122 | #ifdef CONFIG_SLUB_DEBUG_ON | |
123 | DEFINE_STATIC_KEY_TRUE(slub_debug_enabled); | |
124 | #else | |
125 | DEFINE_STATIC_KEY_FALSE(slub_debug_enabled); | |
126 | #endif | |
79270291 | 127 | #endif /* CONFIG_SLUB_DEBUG */ |
ca0cab65 | 128 | |
59052e89 VB |
129 | static inline bool kmem_cache_debug(struct kmem_cache *s) |
130 | { | |
131 | return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS); | |
af537b0a | 132 | } |
5577bd8a | 133 | |
117d54df | 134 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be | 135 | { |
59052e89 | 136 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) |
d86bd1be JK |
137 | p += s->red_left_pad; |
138 | ||
139 | return p; | |
140 | } | |
141 | ||
345c905d JK |
142 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
143 | { | |
144 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
145 | return !kmem_cache_debug(s); | |
146 | #else | |
147 | return false; | |
148 | #endif | |
149 | } | |
150 | ||
81819f0f CL |
151 | /* |
152 | * Issues still to be resolved: | |
153 | * | |
81819f0f CL |
154 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
155 | * | |
81819f0f CL |
156 | * - Variable sizing of the per node arrays |
157 | */ | |
158 | ||
b789ef51 CL |
159 | /* Enable to log cmpxchg failures */ |
160 | #undef SLUB_DEBUG_CMPXCHG | |
161 | ||
2086d26a | 162 | /* |
dc84207d | 163 | * Minimum number of partial slabs. These will be left on the partial |
2086d26a CL |
164 | * lists even if they are empty. kmem_cache_shrink may reclaim them. |
165 | */ | |
76be8950 | 166 | #define MIN_PARTIAL 5 |
e95eed57 | 167 | |
2086d26a CL |
168 | /* |
169 | * Maximum number of desirable partial slabs. | |
170 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 171 | * sort the partial list by the number of objects in use. |
2086d26a CL |
172 | */ |
173 | #define MAX_PARTIAL 10 | |
174 | ||
becfda68 | 175 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 176 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 177 | |
149daaf3 LA |
178 | /* |
179 | * These debug flags cannot use CMPXCHG because there might be consistency | |
180 | * issues when checking or reading debug information | |
181 | */ | |
182 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
183 | SLAB_TRACE) | |
184 | ||
185 | ||
fa5ec8a1 | 186 | /* |
3de47213 DR |
187 | * Debugging flags that require metadata to be stored in the slab. These get |
188 | * disabled when slub_debug=O is used and a cache's min order increases with | |
189 | * metadata. | |
fa5ec8a1 | 190 | */ |
3de47213 | 191 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 192 | |
210b5c06 CG |
193 | #define OO_SHIFT 16 |
194 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 195 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 196 | |
81819f0f | 197 | /* Internal SLUB flags */ |
d50112ed | 198 | /* Poison object */ |
4fd0b46e | 199 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 200 | /* Use cmpxchg_double */ |
4fd0b46e | 201 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 202 | |
02cbc874 CL |
203 | /* |
204 | * Tracking user of a slab. | |
205 | */ | |
d6543e39 | 206 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 207 | struct track { |
ce71e27c | 208 | unsigned long addr; /* Called from address */ |
ae14c63a LT |
209 | #ifdef CONFIG_STACKTRACE |
210 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
d6543e39 | 211 | #endif |
02cbc874 CL |
212 | int cpu; /* Was running on cpu */ |
213 | int pid; /* Pid context */ | |
214 | unsigned long when; /* When did the operation occur */ | |
215 | }; | |
216 | ||
217 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
218 | ||
ab4d5ed5 | 219 | #ifdef CONFIG_SYSFS |
81819f0f CL |
220 | static int sysfs_slab_add(struct kmem_cache *); |
221 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
81819f0f | 222 | #else |
0c710013 CL |
223 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
224 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
225 | { return 0; } | |
81819f0f CL |
226 | #endif |
227 | ||
64dd6849 FM |
228 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG) |
229 | static void debugfs_slab_add(struct kmem_cache *); | |
230 | #else | |
231 | static inline void debugfs_slab_add(struct kmem_cache *s) { } | |
232 | #endif | |
233 | ||
4fdccdfb | 234 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
235 | { |
236 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
237 | /* |
238 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
239 | * avoid this_cpu_add()'s irq-disable overhead. | |
240 | */ | |
241 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
242 | #endif |
243 | } | |
244 | ||
7e1fa93d VB |
245 | /* |
246 | * Tracks for which NUMA nodes we have kmem_cache_nodes allocated. | |
247 | * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily | |
248 | * differ during memory hotplug/hotremove operations. | |
249 | * Protected by slab_mutex. | |
250 | */ | |
251 | static nodemask_t slab_nodes; | |
252 | ||
81819f0f CL |
253 | /******************************************************************** |
254 | * Core slab cache functions | |
255 | *******************************************************************/ | |
256 | ||
2482ddec KC |
257 | /* |
258 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
259 | * with an XOR of the address where the pointer is held and a per-cache | |
260 | * random number. | |
261 | */ | |
262 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
263 | unsigned long ptr_addr) | |
264 | { | |
265 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
d36a63a9 | 266 | /* |
aa1ef4d7 | 267 | * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged. |
d36a63a9 AK |
268 | * Normally, this doesn't cause any issues, as both set_freepointer() |
269 | * and get_freepointer() are called with a pointer with the same tag. | |
270 | * However, there are some issues with CONFIG_SLUB_DEBUG code. For | |
271 | * example, when __free_slub() iterates over objects in a cache, it | |
272 | * passes untagged pointers to check_object(). check_object() in turns | |
273 | * calls get_freepointer() with an untagged pointer, which causes the | |
274 | * freepointer to be restored incorrectly. | |
275 | */ | |
276 | return (void *)((unsigned long)ptr ^ s->random ^ | |
1ad53d9f | 277 | swab((unsigned long)kasan_reset_tag((void *)ptr_addr))); |
2482ddec KC |
278 | #else |
279 | return ptr; | |
280 | #endif | |
281 | } | |
282 | ||
283 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
284 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
285 | void *ptr_addr) | |
286 | { | |
287 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
288 | (unsigned long)ptr_addr); | |
289 | } | |
290 | ||
7656c72b CL |
291 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
292 | { | |
aa1ef4d7 | 293 | object = kasan_reset_tag(object); |
2482ddec | 294 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
295 | } |
296 | ||
0ad9500e ED |
297 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
298 | { | |
0882ff91 | 299 | prefetch(object + s->offset); |
0ad9500e ED |
300 | } |
301 | ||
1393d9a1 CL |
302 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
303 | { | |
2482ddec | 304 | unsigned long freepointer_addr; |
1393d9a1 CL |
305 | void *p; |
306 | ||
8e57f8ac | 307 | if (!debug_pagealloc_enabled_static()) |
922d566c JK |
308 | return get_freepointer(s, object); |
309 | ||
f70b0049 | 310 | object = kasan_reset_tag(object); |
2482ddec | 311 | freepointer_addr = (unsigned long)object + s->offset; |
fe557319 | 312 | copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p)); |
2482ddec | 313 | return freelist_ptr(s, p, freepointer_addr); |
1393d9a1 CL |
314 | } |
315 | ||
7656c72b CL |
316 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
317 | { | |
2482ddec KC |
318 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
319 | ||
ce6fa91b AP |
320 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
321 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
322 | #endif | |
323 | ||
aa1ef4d7 | 324 | freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr); |
2482ddec | 325 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
326 | } |
327 | ||
328 | /* Loop over all objects in a slab */ | |
224a88be | 329 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
330 | for (__p = fixup_red_left(__s, __addr); \ |
331 | __p < (__addr) + (__objects) * (__s)->size; \ | |
332 | __p += (__s)->size) | |
7656c72b | 333 | |
9736d2a9 | 334 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 335 | { |
9736d2a9 | 336 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
337 | } |
338 | ||
19af27af | 339 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 340 | unsigned int size) |
834f3d11 CL |
341 | { |
342 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 343 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
344 | }; |
345 | ||
346 | return x; | |
347 | } | |
348 | ||
19af27af | 349 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 350 | { |
210b5c06 | 351 | return x.x >> OO_SHIFT; |
834f3d11 CL |
352 | } |
353 | ||
19af27af | 354 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 355 | { |
210b5c06 | 356 | return x.x & OO_MASK; |
834f3d11 CL |
357 | } |
358 | ||
881db7fb CL |
359 | /* |
360 | * Per slab locking using the pagelock | |
361 | */ | |
362 | static __always_inline void slab_lock(struct page *page) | |
363 | { | |
48c935ad | 364 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
365 | bit_spin_lock(PG_locked, &page->flags); |
366 | } | |
367 | ||
368 | static __always_inline void slab_unlock(struct page *page) | |
369 | { | |
48c935ad | 370 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
371 | __bit_spin_unlock(PG_locked, &page->flags); |
372 | } | |
373 | ||
1d07171c CL |
374 | /* Interrupts must be disabled (for the fallback code to work right) */ |
375 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
376 | void *freelist_old, unsigned long counters_old, | |
377 | void *freelist_new, unsigned long counters_new, | |
378 | const char *n) | |
379 | { | |
380 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
381 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
382 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 383 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 384 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
385 | freelist_old, counters_old, |
386 | freelist_new, counters_new)) | |
6f6528a1 | 387 | return true; |
1d07171c CL |
388 | } else |
389 | #endif | |
390 | { | |
391 | slab_lock(page); | |
d0e0ac97 CG |
392 | if (page->freelist == freelist_old && |
393 | page->counters == counters_old) { | |
1d07171c | 394 | page->freelist = freelist_new; |
7d27a04b | 395 | page->counters = counters_new; |
1d07171c | 396 | slab_unlock(page); |
6f6528a1 | 397 | return true; |
1d07171c CL |
398 | } |
399 | slab_unlock(page); | |
400 | } | |
401 | ||
402 | cpu_relax(); | |
403 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
404 | ||
405 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 406 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
407 | #endif |
408 | ||
6f6528a1 | 409 | return false; |
1d07171c CL |
410 | } |
411 | ||
b789ef51 CL |
412 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
413 | void *freelist_old, unsigned long counters_old, | |
414 | void *freelist_new, unsigned long counters_new, | |
415 | const char *n) | |
416 | { | |
2565409f HC |
417 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
418 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 419 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 420 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
421 | freelist_old, counters_old, |
422 | freelist_new, counters_new)) | |
6f6528a1 | 423 | return true; |
b789ef51 CL |
424 | } else |
425 | #endif | |
426 | { | |
1d07171c CL |
427 | unsigned long flags; |
428 | ||
429 | local_irq_save(flags); | |
881db7fb | 430 | slab_lock(page); |
d0e0ac97 CG |
431 | if (page->freelist == freelist_old && |
432 | page->counters == counters_old) { | |
b789ef51 | 433 | page->freelist = freelist_new; |
7d27a04b | 434 | page->counters = counters_new; |
881db7fb | 435 | slab_unlock(page); |
1d07171c | 436 | local_irq_restore(flags); |
6f6528a1 | 437 | return true; |
b789ef51 | 438 | } |
881db7fb | 439 | slab_unlock(page); |
1d07171c | 440 | local_irq_restore(flags); |
b789ef51 CL |
441 | } |
442 | ||
443 | cpu_relax(); | |
444 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
445 | ||
446 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 447 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
448 | #endif |
449 | ||
6f6528a1 | 450 | return false; |
b789ef51 CL |
451 | } |
452 | ||
41ecc55b | 453 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 YZ |
454 | static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)]; |
455 | static DEFINE_SPINLOCK(object_map_lock); | |
456 | ||
b3fd64e1 VB |
457 | static void __fill_map(unsigned long *obj_map, struct kmem_cache *s, |
458 | struct page *page) | |
459 | { | |
460 | void *addr = page_address(page); | |
461 | void *p; | |
462 | ||
463 | bitmap_zero(obj_map, page->objects); | |
464 | ||
465 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
466 | set_bit(__obj_to_index(s, addr, p), obj_map); | |
467 | } | |
468 | ||
1f9f78b1 OG |
469 | #if IS_ENABLED(CONFIG_KUNIT) |
470 | static bool slab_add_kunit_errors(void) | |
471 | { | |
472 | struct kunit_resource *resource; | |
473 | ||
474 | if (likely(!current->kunit_test)) | |
475 | return false; | |
476 | ||
477 | resource = kunit_find_named_resource(current->kunit_test, "slab_errors"); | |
478 | if (!resource) | |
479 | return false; | |
480 | ||
481 | (*(int *)resource->data)++; | |
482 | kunit_put_resource(resource); | |
483 | return true; | |
484 | } | |
485 | #else | |
486 | static inline bool slab_add_kunit_errors(void) { return false; } | |
487 | #endif | |
488 | ||
5f80b13a CL |
489 | /* |
490 | * Determine a map of object in use on a page. | |
491 | * | |
881db7fb | 492 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
493 | * not vanish from under us. |
494 | */ | |
90e9f6a6 | 495 | static unsigned long *get_map(struct kmem_cache *s, struct page *page) |
31364c2e | 496 | __acquires(&object_map_lock) |
5f80b13a | 497 | { |
90e9f6a6 YZ |
498 | VM_BUG_ON(!irqs_disabled()); |
499 | ||
500 | spin_lock(&object_map_lock); | |
501 | ||
b3fd64e1 | 502 | __fill_map(object_map, s, page); |
90e9f6a6 YZ |
503 | |
504 | return object_map; | |
505 | } | |
506 | ||
81aba9e0 | 507 | static void put_map(unsigned long *map) __releases(&object_map_lock) |
90e9f6a6 YZ |
508 | { |
509 | VM_BUG_ON(map != object_map); | |
90e9f6a6 | 510 | spin_unlock(&object_map_lock); |
5f80b13a CL |
511 | } |
512 | ||
870b1fbb | 513 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
514 | { |
515 | if (s->flags & SLAB_RED_ZONE) | |
516 | return s->size - s->red_left_pad; | |
517 | ||
518 | return s->size; | |
519 | } | |
520 | ||
521 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
522 | { | |
523 | if (s->flags & SLAB_RED_ZONE) | |
524 | p -= s->red_left_pad; | |
525 | ||
526 | return p; | |
527 | } | |
528 | ||
41ecc55b CL |
529 | /* |
530 | * Debug settings: | |
531 | */ | |
89d3c87e | 532 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 533 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 534 | #else |
d50112ed | 535 | static slab_flags_t slub_debug; |
f0630fff | 536 | #endif |
41ecc55b | 537 | |
e17f1dfb | 538 | static char *slub_debug_string; |
fa5ec8a1 | 539 | static int disable_higher_order_debug; |
41ecc55b | 540 | |
a79316c6 AR |
541 | /* |
542 | * slub is about to manipulate internal object metadata. This memory lies | |
543 | * outside the range of the allocated object, so accessing it would normally | |
544 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
545 | * to tell kasan that these accesses are OK. | |
546 | */ | |
547 | static inline void metadata_access_enable(void) | |
548 | { | |
549 | kasan_disable_current(); | |
550 | } | |
551 | ||
552 | static inline void metadata_access_disable(void) | |
553 | { | |
554 | kasan_enable_current(); | |
555 | } | |
556 | ||
81819f0f CL |
557 | /* |
558 | * Object debugging | |
559 | */ | |
d86bd1be JK |
560 | |
561 | /* Verify that a pointer has an address that is valid within a slab page */ | |
562 | static inline int check_valid_pointer(struct kmem_cache *s, | |
563 | struct page *page, void *object) | |
564 | { | |
565 | void *base; | |
566 | ||
567 | if (!object) | |
568 | return 1; | |
569 | ||
570 | base = page_address(page); | |
338cfaad | 571 | object = kasan_reset_tag(object); |
d86bd1be JK |
572 | object = restore_red_left(s, object); |
573 | if (object < base || object >= base + page->objects * s->size || | |
574 | (object - base) % s->size) { | |
575 | return 0; | |
576 | } | |
577 | ||
578 | return 1; | |
579 | } | |
580 | ||
aa2efd5e DT |
581 | static void print_section(char *level, char *text, u8 *addr, |
582 | unsigned int length) | |
81819f0f | 583 | { |
a79316c6 | 584 | metadata_access_enable(); |
340caf17 KYL |
585 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, |
586 | 16, 1, kasan_reset_tag((void *)addr), length, 1); | |
a79316c6 | 587 | metadata_access_disable(); |
81819f0f CL |
588 | } |
589 | ||
cbfc35a4 WL |
590 | /* |
591 | * See comment in calculate_sizes(). | |
592 | */ | |
593 | static inline bool freeptr_outside_object(struct kmem_cache *s) | |
594 | { | |
595 | return s->offset >= s->inuse; | |
596 | } | |
597 | ||
598 | /* | |
599 | * Return offset of the end of info block which is inuse + free pointer if | |
600 | * not overlapping with object. | |
601 | */ | |
602 | static inline unsigned int get_info_end(struct kmem_cache *s) | |
603 | { | |
604 | if (freeptr_outside_object(s)) | |
605 | return s->inuse + sizeof(void *); | |
606 | else | |
607 | return s->inuse; | |
608 | } | |
609 | ||
81819f0f CL |
610 | static struct track *get_track(struct kmem_cache *s, void *object, |
611 | enum track_item alloc) | |
612 | { | |
613 | struct track *p; | |
614 | ||
cbfc35a4 | 615 | p = object + get_info_end(s); |
81819f0f | 616 | |
aa1ef4d7 | 617 | return kasan_reset_tag(p + alloc); |
81819f0f CL |
618 | } |
619 | ||
620 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 621 | enum track_item alloc, unsigned long addr) |
81819f0f | 622 | { |
1a00df4a | 623 | struct track *p = get_track(s, object, alloc); |
81819f0f | 624 | |
81819f0f | 625 | if (addr) { |
ae14c63a LT |
626 | #ifdef CONFIG_STACKTRACE |
627 | unsigned int nr_entries; | |
628 | ||
629 | metadata_access_enable(); | |
630 | nr_entries = stack_trace_save(kasan_reset_tag(p->addrs), | |
631 | TRACK_ADDRS_COUNT, 3); | |
632 | metadata_access_disable(); | |
633 | ||
634 | if (nr_entries < TRACK_ADDRS_COUNT) | |
635 | p->addrs[nr_entries] = 0; | |
d6543e39 | 636 | #endif |
81819f0f CL |
637 | p->addr = addr; |
638 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 639 | p->pid = current->pid; |
81819f0f | 640 | p->when = jiffies; |
b8ca7ff7 | 641 | } else { |
81819f0f | 642 | memset(p, 0, sizeof(struct track)); |
b8ca7ff7 | 643 | } |
81819f0f CL |
644 | } |
645 | ||
81819f0f CL |
646 | static void init_tracking(struct kmem_cache *s, void *object) |
647 | { | |
24922684 CL |
648 | if (!(s->flags & SLAB_STORE_USER)) |
649 | return; | |
650 | ||
ce71e27c EGM |
651 | set_track(s, object, TRACK_FREE, 0UL); |
652 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
653 | } |
654 | ||
86609d33 | 655 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
656 | { |
657 | if (!t->addr) | |
658 | return; | |
659 | ||
96b94abc | 660 | pr_err("%s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 661 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
ae14c63a | 662 | #ifdef CONFIG_STACKTRACE |
d6543e39 | 663 | { |
ae14c63a LT |
664 | int i; |
665 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
666 | if (t->addrs[i]) | |
667 | pr_err("\t%pS\n", (void *)t->addrs[i]); | |
668 | else | |
669 | break; | |
d6543e39 BG |
670 | } |
671 | #endif | |
24922684 CL |
672 | } |
673 | ||
e42f174e | 674 | void print_tracking(struct kmem_cache *s, void *object) |
24922684 | 675 | { |
86609d33 | 676 | unsigned long pr_time = jiffies; |
24922684 CL |
677 | if (!(s->flags & SLAB_STORE_USER)) |
678 | return; | |
679 | ||
86609d33 CP |
680 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
681 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
682 | } |
683 | ||
684 | static void print_page_info(struct page *page) | |
685 | { | |
96b94abc | 686 | pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n", |
4a8ef190 YS |
687 | page, page->objects, page->inuse, page->freelist, |
688 | page->flags, &page->flags); | |
24922684 CL |
689 | |
690 | } | |
691 | ||
692 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
693 | { | |
ecc42fbe | 694 | struct va_format vaf; |
24922684 | 695 | va_list args; |
24922684 CL |
696 | |
697 | va_start(args, fmt); | |
ecc42fbe FF |
698 | vaf.fmt = fmt; |
699 | vaf.va = &args; | |
f9f58285 | 700 | pr_err("=============================================================================\n"); |
ecc42fbe | 701 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 702 | pr_err("-----------------------------------------------------------------------------\n\n"); |
ecc42fbe | 703 | va_end(args); |
81819f0f CL |
704 | } |
705 | ||
582d1212 | 706 | __printf(2, 3) |
24922684 CL |
707 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
708 | { | |
ecc42fbe | 709 | struct va_format vaf; |
24922684 | 710 | va_list args; |
24922684 | 711 | |
1f9f78b1 OG |
712 | if (slab_add_kunit_errors()) |
713 | return; | |
714 | ||
24922684 | 715 | va_start(args, fmt); |
ecc42fbe FF |
716 | vaf.fmt = fmt; |
717 | vaf.va = &args; | |
718 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 719 | va_end(args); |
24922684 CL |
720 | } |
721 | ||
52f23478 | 722 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 723 | void **freelist, void *nextfree) |
52f23478 DZ |
724 | { |
725 | if ((s->flags & SLAB_CONSISTENCY_CHECKS) && | |
dc07a728 ER |
726 | !check_valid_pointer(s, page, nextfree) && freelist) { |
727 | object_err(s, page, *freelist, "Freechain corrupt"); | |
728 | *freelist = NULL; | |
52f23478 DZ |
729 | slab_fix(s, "Isolate corrupted freechain"); |
730 | return true; | |
731 | } | |
732 | ||
733 | return false; | |
734 | } | |
735 | ||
24922684 | 736 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) |
81819f0f CL |
737 | { |
738 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 739 | u8 *addr = page_address(page); |
24922684 CL |
740 | |
741 | print_tracking(s, p); | |
742 | ||
743 | print_page_info(page); | |
744 | ||
96b94abc | 745 | pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n", |
f9f58285 | 746 | p, p - addr, get_freepointer(s, p)); |
24922684 | 747 | |
d86bd1be | 748 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 749 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
aa2efd5e | 750 | s->red_left_pad); |
d86bd1be | 751 | else if (p > addr + 16) |
aa2efd5e | 752 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 753 | |
8669dbab | 754 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 755 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 756 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 757 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 758 | s->inuse - s->object_size); |
81819f0f | 759 | |
cbfc35a4 | 760 | off = get_info_end(s); |
81819f0f | 761 | |
24922684 | 762 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 763 | off += 2 * sizeof(struct track); |
81819f0f | 764 | |
80a9201a AP |
765 | off += kasan_metadata_size(s); |
766 | ||
d86bd1be | 767 | if (off != size_from_object(s)) |
81819f0f | 768 | /* Beginning of the filler is the free pointer */ |
8669dbab | 769 | print_section(KERN_ERR, "Padding ", p + off, |
aa2efd5e | 770 | size_from_object(s) - off); |
24922684 CL |
771 | |
772 | dump_stack(); | |
81819f0f CL |
773 | } |
774 | ||
75c66def | 775 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
776 | u8 *object, char *reason) |
777 | { | |
1f9f78b1 OG |
778 | if (slab_add_kunit_errors()) |
779 | return; | |
780 | ||
3dc50637 | 781 | slab_bug(s, "%s", reason); |
24922684 | 782 | print_trailer(s, page, object); |
65ebdeef | 783 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
784 | } |
785 | ||
a38965bf | 786 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 787 | const char *fmt, ...) |
81819f0f CL |
788 | { |
789 | va_list args; | |
790 | char buf[100]; | |
791 | ||
1f9f78b1 OG |
792 | if (slab_add_kunit_errors()) |
793 | return; | |
794 | ||
24922684 CL |
795 | va_start(args, fmt); |
796 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 797 | va_end(args); |
3dc50637 | 798 | slab_bug(s, "%s", buf); |
24922684 | 799 | print_page_info(page); |
81819f0f | 800 | dump_stack(); |
65ebdeef | 801 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
802 | } |
803 | ||
f7cb1933 | 804 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f | 805 | { |
aa1ef4d7 | 806 | u8 *p = kasan_reset_tag(object); |
81819f0f | 807 | |
d86bd1be JK |
808 | if (s->flags & SLAB_RED_ZONE) |
809 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
810 | ||
81819f0f | 811 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
812 | memset(p, POISON_FREE, s->object_size - 1); |
813 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
814 | } |
815 | ||
816 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 817 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
818 | } |
819 | ||
24922684 CL |
820 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
821 | void *from, void *to) | |
822 | { | |
582d1212 | 823 | slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data); |
24922684 CL |
824 | memset(from, data, to - from); |
825 | } | |
826 | ||
827 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
828 | u8 *object, char *what, | |
06428780 | 829 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
830 | { |
831 | u8 *fault; | |
832 | u8 *end; | |
e1b70dd1 | 833 | u8 *addr = page_address(page); |
24922684 | 834 | |
a79316c6 | 835 | metadata_access_enable(); |
aa1ef4d7 | 836 | fault = memchr_inv(kasan_reset_tag(start), value, bytes); |
a79316c6 | 837 | metadata_access_disable(); |
24922684 CL |
838 | if (!fault) |
839 | return 1; | |
840 | ||
841 | end = start + bytes; | |
842 | while (end > fault && end[-1] == value) | |
843 | end--; | |
844 | ||
1f9f78b1 OG |
845 | if (slab_add_kunit_errors()) |
846 | goto skip_bug_print; | |
847 | ||
24922684 | 848 | slab_bug(s, "%s overwritten", what); |
96b94abc | 849 | pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", |
e1b70dd1 MC |
850 | fault, end - 1, fault - addr, |
851 | fault[0], value); | |
24922684 | 852 | print_trailer(s, page, object); |
65ebdeef | 853 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
24922684 | 854 | |
1f9f78b1 | 855 | skip_bug_print: |
24922684 CL |
856 | restore_bytes(s, what, value, fault, end); |
857 | return 0; | |
81819f0f CL |
858 | } |
859 | ||
81819f0f CL |
860 | /* |
861 | * Object layout: | |
862 | * | |
863 | * object address | |
864 | * Bytes of the object to be managed. | |
865 | * If the freepointer may overlay the object then the free | |
cbfc35a4 | 866 | * pointer is at the middle of the object. |
672bba3a | 867 | * |
81819f0f CL |
868 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
869 | * 0xa5 (POISON_END) | |
870 | * | |
3b0efdfa | 871 | * object + s->object_size |
81819f0f | 872 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 873 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 874 | * object_size == inuse. |
672bba3a | 875 | * |
81819f0f CL |
876 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
877 | * 0xcc (RED_ACTIVE) for objects in use. | |
878 | * | |
879 | * object + s->inuse | |
672bba3a CL |
880 | * Meta data starts here. |
881 | * | |
81819f0f CL |
882 | * A. Free pointer (if we cannot overwrite object on free) |
883 | * B. Tracking data for SLAB_STORE_USER | |
dc84207d | 884 | * C. Padding to reach required alignment boundary or at minimum |
6446faa2 | 885 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
886 | * before the word boundary. |
887 | * | |
888 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
889 | * |
890 | * object + s->size | |
672bba3a | 891 | * Nothing is used beyond s->size. |
81819f0f | 892 | * |
3b0efdfa | 893 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 894 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
895 | * may be used with merged slabcaches. |
896 | */ | |
897 | ||
81819f0f CL |
898 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
899 | { | |
cbfc35a4 | 900 | unsigned long off = get_info_end(s); /* The end of info */ |
81819f0f CL |
901 | |
902 | if (s->flags & SLAB_STORE_USER) | |
903 | /* We also have user information there */ | |
904 | off += 2 * sizeof(struct track); | |
905 | ||
80a9201a AP |
906 | off += kasan_metadata_size(s); |
907 | ||
d86bd1be | 908 | if (size_from_object(s) == off) |
81819f0f CL |
909 | return 1; |
910 | ||
24922684 | 911 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 912 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
913 | } |
914 | ||
39b26464 | 915 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
916 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
917 | { | |
24922684 CL |
918 | u8 *start; |
919 | u8 *fault; | |
920 | u8 *end; | |
5d682681 | 921 | u8 *pad; |
24922684 CL |
922 | int length; |
923 | int remainder; | |
81819f0f CL |
924 | |
925 | if (!(s->flags & SLAB_POISON)) | |
926 | return 1; | |
927 | ||
a973e9dd | 928 | start = page_address(page); |
a50b854e | 929 | length = page_size(page); |
39b26464 CL |
930 | end = start + length; |
931 | remainder = length % s->size; | |
81819f0f CL |
932 | if (!remainder) |
933 | return 1; | |
934 | ||
5d682681 | 935 | pad = end - remainder; |
a79316c6 | 936 | metadata_access_enable(); |
aa1ef4d7 | 937 | fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder); |
a79316c6 | 938 | metadata_access_disable(); |
24922684 CL |
939 | if (!fault) |
940 | return 1; | |
941 | while (end > fault && end[-1] == POISON_INUSE) | |
942 | end--; | |
943 | ||
e1b70dd1 MC |
944 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu", |
945 | fault, end - 1, fault - start); | |
5d682681 | 946 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 947 | |
5d682681 | 948 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 949 | return 0; |
81819f0f CL |
950 | } |
951 | ||
952 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 953 | void *object, u8 val) |
81819f0f CL |
954 | { |
955 | u8 *p = object; | |
3b0efdfa | 956 | u8 *endobject = object + s->object_size; |
81819f0f CL |
957 | |
958 | if (s->flags & SLAB_RED_ZONE) { | |
8669dbab | 959 | if (!check_bytes_and_report(s, page, object, "Left Redzone", |
d86bd1be JK |
960 | object - s->red_left_pad, val, s->red_left_pad)) |
961 | return 0; | |
962 | ||
8669dbab | 963 | if (!check_bytes_and_report(s, page, object, "Right Redzone", |
3b0efdfa | 964 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 965 | return 0; |
81819f0f | 966 | } else { |
3b0efdfa | 967 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 968 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
969 | endobject, POISON_INUSE, |
970 | s->inuse - s->object_size); | |
3adbefee | 971 | } |
81819f0f CL |
972 | } |
973 | ||
974 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 975 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 976 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 977 | POISON_FREE, s->object_size - 1) || |
8669dbab | 978 | !check_bytes_and_report(s, page, p, "End Poison", |
3b0efdfa | 979 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 980 | return 0; |
81819f0f CL |
981 | /* |
982 | * check_pad_bytes cleans up on its own. | |
983 | */ | |
984 | check_pad_bytes(s, page, p); | |
985 | } | |
986 | ||
cbfc35a4 | 987 | if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE) |
81819f0f CL |
988 | /* |
989 | * Object and freepointer overlap. Cannot check | |
990 | * freepointer while object is allocated. | |
991 | */ | |
992 | return 1; | |
993 | ||
994 | /* Check free pointer validity */ | |
995 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
996 | object_err(s, page, p, "Freepointer corrupt"); | |
997 | /* | |
9f6c708e | 998 | * No choice but to zap it and thus lose the remainder |
81819f0f | 999 | * of the free objects in this slab. May cause |
672bba3a | 1000 | * another error because the object count is now wrong. |
81819f0f | 1001 | */ |
a973e9dd | 1002 | set_freepointer(s, p, NULL); |
81819f0f CL |
1003 | return 0; |
1004 | } | |
1005 | return 1; | |
1006 | } | |
1007 | ||
1008 | static int check_slab(struct kmem_cache *s, struct page *page) | |
1009 | { | |
39b26464 CL |
1010 | int maxobj; |
1011 | ||
81819f0f | 1012 | if (!PageSlab(page)) { |
24922684 | 1013 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
1014 | return 0; |
1015 | } | |
39b26464 | 1016 | |
9736d2a9 | 1017 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
1018 | if (page->objects > maxobj) { |
1019 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 1020 | page->objects, maxobj); |
39b26464 CL |
1021 | return 0; |
1022 | } | |
1023 | if (page->inuse > page->objects) { | |
24922684 | 1024 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 1025 | page->inuse, page->objects); |
81819f0f CL |
1026 | return 0; |
1027 | } | |
1028 | /* Slab_pad_check fixes things up after itself */ | |
1029 | slab_pad_check(s, page); | |
1030 | return 1; | |
1031 | } | |
1032 | ||
1033 | /* | |
672bba3a CL |
1034 | * Determine if a certain object on a page is on the freelist. Must hold the |
1035 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
1036 | */ |
1037 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
1038 | { | |
1039 | int nr = 0; | |
881db7fb | 1040 | void *fp; |
81819f0f | 1041 | void *object = NULL; |
f6edde9c | 1042 | int max_objects; |
81819f0f | 1043 | |
881db7fb | 1044 | fp = page->freelist; |
39b26464 | 1045 | while (fp && nr <= page->objects) { |
81819f0f CL |
1046 | if (fp == search) |
1047 | return 1; | |
1048 | if (!check_valid_pointer(s, page, fp)) { | |
1049 | if (object) { | |
1050 | object_err(s, page, object, | |
1051 | "Freechain corrupt"); | |
a973e9dd | 1052 | set_freepointer(s, object, NULL); |
81819f0f | 1053 | } else { |
24922684 | 1054 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 1055 | page->freelist = NULL; |
39b26464 | 1056 | page->inuse = page->objects; |
24922684 | 1057 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
1058 | return 0; |
1059 | } | |
1060 | break; | |
1061 | } | |
1062 | object = fp; | |
1063 | fp = get_freepointer(s, object); | |
1064 | nr++; | |
1065 | } | |
1066 | ||
9736d2a9 | 1067 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
1068 | if (max_objects > MAX_OBJS_PER_PAGE) |
1069 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
1070 | |
1071 | if (page->objects != max_objects) { | |
756a025f JP |
1072 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
1073 | page->objects, max_objects); | |
224a88be | 1074 | page->objects = max_objects; |
582d1212 | 1075 | slab_fix(s, "Number of objects adjusted"); |
224a88be | 1076 | } |
39b26464 | 1077 | if (page->inuse != page->objects - nr) { |
756a025f JP |
1078 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
1079 | page->inuse, page->objects - nr); | |
39b26464 | 1080 | page->inuse = page->objects - nr; |
582d1212 | 1081 | slab_fix(s, "Object count adjusted"); |
81819f0f CL |
1082 | } |
1083 | return search == NULL; | |
1084 | } | |
1085 | ||
0121c619 CL |
1086 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
1087 | int alloc) | |
3ec09742 CL |
1088 | { |
1089 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1090 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1091 | s->name, |
1092 | alloc ? "alloc" : "free", | |
1093 | object, page->inuse, | |
1094 | page->freelist); | |
1095 | ||
1096 | if (!alloc) | |
aa2efd5e | 1097 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1098 | s->object_size); |
3ec09742 CL |
1099 | |
1100 | dump_stack(); | |
1101 | } | |
1102 | } | |
1103 | ||
643b1138 | 1104 | /* |
672bba3a | 1105 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1106 | */ |
5cc6eee8 CL |
1107 | static void add_full(struct kmem_cache *s, |
1108 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1109 | { |
5cc6eee8 CL |
1110 | if (!(s->flags & SLAB_STORE_USER)) |
1111 | return; | |
1112 | ||
255d0884 | 1113 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1114 | list_add(&page->slab_list, &n->full); |
643b1138 CL |
1115 | } |
1116 | ||
c65c1877 | 1117 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1118 | { |
643b1138 CL |
1119 | if (!(s->flags & SLAB_STORE_USER)) |
1120 | return; | |
1121 | ||
255d0884 | 1122 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1123 | list_del(&page->slab_list); |
643b1138 CL |
1124 | } |
1125 | ||
0f389ec6 CL |
1126 | /* Tracking of the number of slabs for debugging purposes */ |
1127 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1128 | { | |
1129 | struct kmem_cache_node *n = get_node(s, node); | |
1130 | ||
1131 | return atomic_long_read(&n->nr_slabs); | |
1132 | } | |
1133 | ||
26c02cf0 AB |
1134 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1135 | { | |
1136 | return atomic_long_read(&n->nr_slabs); | |
1137 | } | |
1138 | ||
205ab99d | 1139 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1140 | { |
1141 | struct kmem_cache_node *n = get_node(s, node); | |
1142 | ||
1143 | /* | |
1144 | * May be called early in order to allocate a slab for the | |
1145 | * kmem_cache_node structure. Solve the chicken-egg | |
1146 | * dilemma by deferring the increment of the count during | |
1147 | * bootstrap (see early_kmem_cache_node_alloc). | |
1148 | */ | |
338b2642 | 1149 | if (likely(n)) { |
0f389ec6 | 1150 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1151 | atomic_long_add(objects, &n->total_objects); |
1152 | } | |
0f389ec6 | 1153 | } |
205ab99d | 1154 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1155 | { |
1156 | struct kmem_cache_node *n = get_node(s, node); | |
1157 | ||
1158 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1159 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1160 | } |
1161 | ||
1162 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1163 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1164 | void *object) | |
1165 | { | |
8fc8d666 | 1166 | if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)) |
3ec09742 CL |
1167 | return; |
1168 | ||
f7cb1933 | 1169 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1170 | init_tracking(s, object); |
1171 | } | |
1172 | ||
a50b854e MWO |
1173 | static |
1174 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) | |
a7101224 | 1175 | { |
8fc8d666 | 1176 | if (!kmem_cache_debug_flags(s, SLAB_POISON)) |
a7101224 AK |
1177 | return; |
1178 | ||
1179 | metadata_access_enable(); | |
aa1ef4d7 | 1180 | memset(kasan_reset_tag(addr), POISON_INUSE, page_size(page)); |
a7101224 AK |
1181 | metadata_access_disable(); |
1182 | } | |
1183 | ||
becfda68 | 1184 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
278d7756 | 1185 | struct page *page, void *object) |
81819f0f CL |
1186 | { |
1187 | if (!check_slab(s, page)) | |
becfda68 | 1188 | return 0; |
81819f0f | 1189 | |
81819f0f CL |
1190 | if (!check_valid_pointer(s, page, object)) { |
1191 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1192 | return 0; |
81819f0f CL |
1193 | } |
1194 | ||
f7cb1933 | 1195 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1196 | return 0; |
1197 | ||
1198 | return 1; | |
1199 | } | |
1200 | ||
1201 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1202 | struct page *page, | |
1203 | void *object, unsigned long addr) | |
1204 | { | |
1205 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
278d7756 | 1206 | if (!alloc_consistency_checks(s, page, object)) |
becfda68 LA |
1207 | goto bad; |
1208 | } | |
81819f0f | 1209 | |
3ec09742 CL |
1210 | /* Success perform special debug activities for allocs */ |
1211 | if (s->flags & SLAB_STORE_USER) | |
1212 | set_track(s, object, TRACK_ALLOC, addr); | |
1213 | trace(s, page, object, 1); | |
f7cb1933 | 1214 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1215 | return 1; |
3ec09742 | 1216 | |
81819f0f CL |
1217 | bad: |
1218 | if (PageSlab(page)) { | |
1219 | /* | |
1220 | * If this is a slab page then lets do the best we can | |
1221 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1222 | * as used avoids touching the remaining objects. |
81819f0f | 1223 | */ |
24922684 | 1224 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1225 | page->inuse = page->objects; |
a973e9dd | 1226 | page->freelist = NULL; |
81819f0f CL |
1227 | } |
1228 | return 0; | |
1229 | } | |
1230 | ||
becfda68 LA |
1231 | static inline int free_consistency_checks(struct kmem_cache *s, |
1232 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1233 | { |
81819f0f | 1234 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1235 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1236 | return 0; |
81819f0f CL |
1237 | } |
1238 | ||
1239 | if (on_freelist(s, page, object)) { | |
24922684 | 1240 | object_err(s, page, object, "Object already free"); |
becfda68 | 1241 | return 0; |
81819f0f CL |
1242 | } |
1243 | ||
f7cb1933 | 1244 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1245 | return 0; |
81819f0f | 1246 | |
1b4f59e3 | 1247 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1248 | if (!PageSlab(page)) { |
756a025f JP |
1249 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1250 | object); | |
1b4f59e3 | 1251 | } else if (!page->slab_cache) { |
f9f58285 FF |
1252 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1253 | object); | |
70d71228 | 1254 | dump_stack(); |
06428780 | 1255 | } else |
24922684 CL |
1256 | object_err(s, page, object, |
1257 | "page slab pointer corrupt."); | |
becfda68 LA |
1258 | return 0; |
1259 | } | |
1260 | return 1; | |
1261 | } | |
1262 | ||
1263 | /* Supports checking bulk free of a constructed freelist */ | |
1264 | static noinline int free_debug_processing( | |
1265 | struct kmem_cache *s, struct page *page, | |
1266 | void *head, void *tail, int bulk_cnt, | |
1267 | unsigned long addr) | |
1268 | { | |
1269 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1270 | void *object = head; | |
1271 | int cnt = 0; | |
3f649ab7 | 1272 | unsigned long flags; |
becfda68 LA |
1273 | int ret = 0; |
1274 | ||
1275 | spin_lock_irqsave(&n->list_lock, flags); | |
1276 | slab_lock(page); | |
1277 | ||
1278 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1279 | if (!check_slab(s, page)) | |
1280 | goto out; | |
1281 | } | |
1282 | ||
1283 | next_object: | |
1284 | cnt++; | |
1285 | ||
1286 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1287 | if (!free_consistency_checks(s, page, object, addr)) | |
1288 | goto out; | |
81819f0f | 1289 | } |
3ec09742 | 1290 | |
3ec09742 CL |
1291 | if (s->flags & SLAB_STORE_USER) |
1292 | set_track(s, object, TRACK_FREE, addr); | |
1293 | trace(s, page, object, 0); | |
81084651 | 1294 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1295 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1296 | |
1297 | /* Reached end of constructed freelist yet? */ | |
1298 | if (object != tail) { | |
1299 | object = get_freepointer(s, object); | |
1300 | goto next_object; | |
1301 | } | |
804aa132 LA |
1302 | ret = 1; |
1303 | ||
5c2e4bbb | 1304 | out: |
81084651 JDB |
1305 | if (cnt != bulk_cnt) |
1306 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1307 | bulk_cnt, cnt); | |
1308 | ||
881db7fb | 1309 | slab_unlock(page); |
282acb43 | 1310 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1311 | if (!ret) |
1312 | slab_fix(s, "Object at 0x%p not freed", object); | |
1313 | return ret; | |
81819f0f CL |
1314 | } |
1315 | ||
e17f1dfb VB |
1316 | /* |
1317 | * Parse a block of slub_debug options. Blocks are delimited by ';' | |
1318 | * | |
1319 | * @str: start of block | |
1320 | * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified | |
1321 | * @slabs: return start of list of slabs, or NULL when there's no list | |
1322 | * @init: assume this is initial parsing and not per-kmem-create parsing | |
1323 | * | |
1324 | * returns the start of next block if there's any, or NULL | |
1325 | */ | |
1326 | static char * | |
1327 | parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init) | |
41ecc55b | 1328 | { |
e17f1dfb | 1329 | bool higher_order_disable = false; |
f0630fff | 1330 | |
e17f1dfb VB |
1331 | /* Skip any completely empty blocks */ |
1332 | while (*str && *str == ';') | |
1333 | str++; | |
1334 | ||
1335 | if (*str == ',') { | |
f0630fff CL |
1336 | /* |
1337 | * No options but restriction on slabs. This means full | |
1338 | * debugging for slabs matching a pattern. | |
1339 | */ | |
e17f1dfb | 1340 | *flags = DEBUG_DEFAULT_FLAGS; |
f0630fff | 1341 | goto check_slabs; |
e17f1dfb VB |
1342 | } |
1343 | *flags = 0; | |
f0630fff | 1344 | |
e17f1dfb VB |
1345 | /* Determine which debug features should be switched on */ |
1346 | for (; *str && *str != ',' && *str != ';'; str++) { | |
f0630fff | 1347 | switch (tolower(*str)) { |
e17f1dfb VB |
1348 | case '-': |
1349 | *flags = 0; | |
1350 | break; | |
f0630fff | 1351 | case 'f': |
e17f1dfb | 1352 | *flags |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1353 | break; |
1354 | case 'z': | |
e17f1dfb | 1355 | *flags |= SLAB_RED_ZONE; |
f0630fff CL |
1356 | break; |
1357 | case 'p': | |
e17f1dfb | 1358 | *flags |= SLAB_POISON; |
f0630fff CL |
1359 | break; |
1360 | case 'u': | |
e17f1dfb | 1361 | *flags |= SLAB_STORE_USER; |
f0630fff CL |
1362 | break; |
1363 | case 't': | |
e17f1dfb | 1364 | *flags |= SLAB_TRACE; |
f0630fff | 1365 | break; |
4c13dd3b | 1366 | case 'a': |
e17f1dfb | 1367 | *flags |= SLAB_FAILSLAB; |
4c13dd3b | 1368 | break; |
08303a73 CA |
1369 | case 'o': |
1370 | /* | |
1371 | * Avoid enabling debugging on caches if its minimum | |
1372 | * order would increase as a result. | |
1373 | */ | |
e17f1dfb | 1374 | higher_order_disable = true; |
08303a73 | 1375 | break; |
f0630fff | 1376 | default: |
e17f1dfb VB |
1377 | if (init) |
1378 | pr_err("slub_debug option '%c' unknown. skipped\n", *str); | |
f0630fff | 1379 | } |
41ecc55b | 1380 | } |
f0630fff | 1381 | check_slabs: |
41ecc55b | 1382 | if (*str == ',') |
e17f1dfb VB |
1383 | *slabs = ++str; |
1384 | else | |
1385 | *slabs = NULL; | |
1386 | ||
1387 | /* Skip over the slab list */ | |
1388 | while (*str && *str != ';') | |
1389 | str++; | |
1390 | ||
1391 | /* Skip any completely empty blocks */ | |
1392 | while (*str && *str == ';') | |
1393 | str++; | |
1394 | ||
1395 | if (init && higher_order_disable) | |
1396 | disable_higher_order_debug = 1; | |
1397 | ||
1398 | if (*str) | |
1399 | return str; | |
1400 | else | |
1401 | return NULL; | |
1402 | } | |
1403 | ||
1404 | static int __init setup_slub_debug(char *str) | |
1405 | { | |
1406 | slab_flags_t flags; | |
a7f1d485 | 1407 | slab_flags_t global_flags; |
e17f1dfb VB |
1408 | char *saved_str; |
1409 | char *slab_list; | |
1410 | bool global_slub_debug_changed = false; | |
1411 | bool slab_list_specified = false; | |
1412 | ||
a7f1d485 | 1413 | global_flags = DEBUG_DEFAULT_FLAGS; |
e17f1dfb VB |
1414 | if (*str++ != '=' || !*str) |
1415 | /* | |
1416 | * No options specified. Switch on full debugging. | |
1417 | */ | |
1418 | goto out; | |
1419 | ||
1420 | saved_str = str; | |
1421 | while (str) { | |
1422 | str = parse_slub_debug_flags(str, &flags, &slab_list, true); | |
1423 | ||
1424 | if (!slab_list) { | |
a7f1d485 | 1425 | global_flags = flags; |
e17f1dfb VB |
1426 | global_slub_debug_changed = true; |
1427 | } else { | |
1428 | slab_list_specified = true; | |
1429 | } | |
1430 | } | |
1431 | ||
1432 | /* | |
1433 | * For backwards compatibility, a single list of flags with list of | |
a7f1d485 VB |
1434 | * slabs means debugging is only changed for those slabs, so the global |
1435 | * slub_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending | |
1436 | * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as | |
e17f1dfb VB |
1437 | * long as there is no option specifying flags without a slab list. |
1438 | */ | |
1439 | if (slab_list_specified) { | |
1440 | if (!global_slub_debug_changed) | |
a7f1d485 | 1441 | global_flags = slub_debug; |
e17f1dfb VB |
1442 | slub_debug_string = saved_str; |
1443 | } | |
f0630fff | 1444 | out: |
a7f1d485 | 1445 | slub_debug = global_flags; |
ca0cab65 VB |
1446 | if (slub_debug != 0 || slub_debug_string) |
1447 | static_branch_enable(&slub_debug_enabled); | |
02ac47d0 SB |
1448 | else |
1449 | static_branch_disable(&slub_debug_enabled); | |
6471384a AP |
1450 | if ((static_branch_unlikely(&init_on_alloc) || |
1451 | static_branch_unlikely(&init_on_free)) && | |
1452 | (slub_debug & SLAB_POISON)) | |
1453 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
41ecc55b CL |
1454 | return 1; |
1455 | } | |
1456 | ||
1457 | __setup("slub_debug", setup_slub_debug); | |
1458 | ||
c5fd3ca0 AT |
1459 | /* |
1460 | * kmem_cache_flags - apply debugging options to the cache | |
1461 | * @object_size: the size of an object without meta data | |
1462 | * @flags: flags to set | |
1463 | * @name: name of the cache | |
c5fd3ca0 AT |
1464 | * |
1465 | * Debug option(s) are applied to @flags. In addition to the debug | |
1466 | * option(s), if a slab name (or multiple) is specified i.e. | |
1467 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1468 | * then only the select slabs will receive the debug option(s). | |
1469 | */ | |
0293d1fd | 1470 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1471 | slab_flags_t flags, const char *name) |
41ecc55b | 1472 | { |
c5fd3ca0 AT |
1473 | char *iter; |
1474 | size_t len; | |
e17f1dfb VB |
1475 | char *next_block; |
1476 | slab_flags_t block_flags; | |
ca220593 JB |
1477 | slab_flags_t slub_debug_local = slub_debug; |
1478 | ||
1479 | /* | |
1480 | * If the slab cache is for debugging (e.g. kmemleak) then | |
1481 | * don't store user (stack trace) information by default, | |
1482 | * but let the user enable it via the command line below. | |
1483 | */ | |
1484 | if (flags & SLAB_NOLEAKTRACE) | |
1485 | slub_debug_local &= ~SLAB_STORE_USER; | |
c5fd3ca0 | 1486 | |
c5fd3ca0 | 1487 | len = strlen(name); |
e17f1dfb VB |
1488 | next_block = slub_debug_string; |
1489 | /* Go through all blocks of debug options, see if any matches our slab's name */ | |
1490 | while (next_block) { | |
1491 | next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false); | |
1492 | if (!iter) | |
1493 | continue; | |
1494 | /* Found a block that has a slab list, search it */ | |
1495 | while (*iter) { | |
1496 | char *end, *glob; | |
1497 | size_t cmplen; | |
1498 | ||
1499 | end = strchrnul(iter, ','); | |
1500 | if (next_block && next_block < end) | |
1501 | end = next_block - 1; | |
1502 | ||
1503 | glob = strnchr(iter, end - iter, '*'); | |
1504 | if (glob) | |
1505 | cmplen = glob - iter; | |
1506 | else | |
1507 | cmplen = max_t(size_t, len, (end - iter)); | |
c5fd3ca0 | 1508 | |
e17f1dfb VB |
1509 | if (!strncmp(name, iter, cmplen)) { |
1510 | flags |= block_flags; | |
1511 | return flags; | |
1512 | } | |
c5fd3ca0 | 1513 | |
e17f1dfb VB |
1514 | if (!*end || *end == ';') |
1515 | break; | |
1516 | iter = end + 1; | |
c5fd3ca0 | 1517 | } |
c5fd3ca0 | 1518 | } |
ba0268a8 | 1519 | |
ca220593 | 1520 | return flags | slub_debug_local; |
41ecc55b | 1521 | } |
b4a64718 | 1522 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1523 | static inline void setup_object_debug(struct kmem_cache *s, |
1524 | struct page *page, void *object) {} | |
a50b854e MWO |
1525 | static inline |
1526 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {} | |
41ecc55b | 1527 | |
3ec09742 | 1528 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1529 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1530 | |
282acb43 | 1531 | static inline int free_debug_processing( |
81084651 JDB |
1532 | struct kmem_cache *s, struct page *page, |
1533 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1534 | unsigned long addr) { return 0; } |
41ecc55b | 1535 | |
41ecc55b CL |
1536 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1537 | { return 1; } | |
1538 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1539 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1540 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1541 | struct page *page) {} | |
c65c1877 PZ |
1542 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1543 | struct page *page) {} | |
0293d1fd | 1544 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1545 | slab_flags_t flags, const char *name) |
ba0268a8 CL |
1546 | { |
1547 | return flags; | |
1548 | } | |
41ecc55b | 1549 | #define slub_debug 0 |
0f389ec6 | 1550 | |
fdaa45e9 IM |
1551 | #define disable_higher_order_debug 0 |
1552 | ||
0f389ec6 CL |
1553 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1554 | { return 0; } | |
26c02cf0 AB |
1555 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1556 | { return 0; } | |
205ab99d CL |
1557 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1558 | int objects) {} | |
1559 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1560 | int objects) {} | |
7d550c56 | 1561 | |
52f23478 | 1562 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 1563 | void **freelist, void *nextfree) |
52f23478 DZ |
1564 | { |
1565 | return false; | |
1566 | } | |
02e72cc6 AR |
1567 | #endif /* CONFIG_SLUB_DEBUG */ |
1568 | ||
1569 | /* | |
1570 | * Hooks for other subsystems that check memory allocations. In a typical | |
1571 | * production configuration these hooks all should produce no code at all. | |
1572 | */ | |
0116523c | 1573 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1574 | { |
53128245 | 1575 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1576 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1577 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1578 | return ptr; |
d56791b3 RB |
1579 | } |
1580 | ||
ee3ce779 | 1581 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1582 | { |
1583 | kmemleak_free(x); | |
027b37b5 | 1584 | kasan_kfree_large(x); |
d56791b3 RB |
1585 | } |
1586 | ||
d57a964e AK |
1587 | static __always_inline bool slab_free_hook(struct kmem_cache *s, |
1588 | void *x, bool init) | |
d56791b3 RB |
1589 | { |
1590 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1591 | |
84048039 | 1592 | debug_check_no_locks_freed(x, s->object_size); |
02e72cc6 | 1593 | |
02e72cc6 AR |
1594 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
1595 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1596 | |
cfbe1636 ME |
1597 | /* Use KCSAN to help debug racy use-after-free. */ |
1598 | if (!(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
1599 | __kcsan_check_access(x, s->object_size, | |
1600 | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT); | |
1601 | ||
d57a964e AK |
1602 | /* |
1603 | * As memory initialization might be integrated into KASAN, | |
1604 | * kasan_slab_free and initialization memset's must be | |
1605 | * kept together to avoid discrepancies in behavior. | |
1606 | * | |
1607 | * The initialization memset's clear the object and the metadata, | |
1608 | * but don't touch the SLAB redzone. | |
1609 | */ | |
1610 | if (init) { | |
1611 | int rsize; | |
1612 | ||
1613 | if (!kasan_has_integrated_init()) | |
1614 | memset(kasan_reset_tag(x), 0, s->object_size); | |
1615 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0; | |
1616 | memset((char *)kasan_reset_tag(x) + s->inuse, 0, | |
1617 | s->size - s->inuse - rsize); | |
1618 | } | |
1619 | /* KASAN might put x into memory quarantine, delaying its reuse. */ | |
1620 | return kasan_slab_free(s, x, init); | |
02e72cc6 | 1621 | } |
205ab99d | 1622 | |
c3895391 AK |
1623 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1624 | void **head, void **tail) | |
81084651 | 1625 | { |
6471384a AP |
1626 | |
1627 | void *object; | |
1628 | void *next = *head; | |
1629 | void *old_tail = *tail ? *tail : *head; | |
6471384a | 1630 | |
b89fb5ef | 1631 | if (is_kfence_address(next)) { |
d57a964e | 1632 | slab_free_hook(s, next, false); |
b89fb5ef AP |
1633 | return true; |
1634 | } | |
1635 | ||
aea4df4c LA |
1636 | /* Head and tail of the reconstructed freelist */ |
1637 | *head = NULL; | |
1638 | *tail = NULL; | |
1b7e816f | 1639 | |
aea4df4c LA |
1640 | do { |
1641 | object = next; | |
1642 | next = get_freepointer(s, object); | |
1643 | ||
c3895391 | 1644 | /* If object's reuse doesn't have to be delayed */ |
d57a964e | 1645 | if (!slab_free_hook(s, object, slab_want_init_on_free(s))) { |
c3895391 AK |
1646 | /* Move object to the new freelist */ |
1647 | set_freepointer(s, object, *head); | |
1648 | *head = object; | |
1649 | if (!*tail) | |
1650 | *tail = object; | |
1651 | } | |
1652 | } while (object != old_tail); | |
1653 | ||
1654 | if (*head == *tail) | |
1655 | *tail = NULL; | |
1656 | ||
1657 | return *head != NULL; | |
81084651 JDB |
1658 | } |
1659 | ||
4d176711 | 1660 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1661 | void *object) |
1662 | { | |
1663 | setup_object_debug(s, page, object); | |
4d176711 | 1664 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1665 | if (unlikely(s->ctor)) { |
1666 | kasan_unpoison_object_data(s, object); | |
1667 | s->ctor(object); | |
1668 | kasan_poison_object_data(s, object); | |
1669 | } | |
4d176711 | 1670 | return object; |
588f8ba9 TG |
1671 | } |
1672 | ||
81819f0f CL |
1673 | /* |
1674 | * Slab allocation and freeing | |
1675 | */ | |
5dfb4175 VD |
1676 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1677 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1678 | { |
5dfb4175 | 1679 | struct page *page; |
19af27af | 1680 | unsigned int order = oo_order(oo); |
65c3376a | 1681 | |
2154a336 | 1682 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1683 | page = alloc_pages(flags, order); |
65c3376a | 1684 | else |
96db800f | 1685 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1686 | |
5dfb4175 | 1687 | return page; |
65c3376a CL |
1688 | } |
1689 | ||
210e7a43 TG |
1690 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1691 | /* Pre-initialize the random sequence cache */ | |
1692 | static int init_cache_random_seq(struct kmem_cache *s) | |
1693 | { | |
19af27af | 1694 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1695 | int err; |
210e7a43 | 1696 | |
a810007a SR |
1697 | /* Bailout if already initialised */ |
1698 | if (s->random_seq) | |
1699 | return 0; | |
1700 | ||
210e7a43 TG |
1701 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1702 | if (err) { | |
1703 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1704 | s->name); | |
1705 | return err; | |
1706 | } | |
1707 | ||
1708 | /* Transform to an offset on the set of pages */ | |
1709 | if (s->random_seq) { | |
19af27af AD |
1710 | unsigned int i; |
1711 | ||
210e7a43 TG |
1712 | for (i = 0; i < count; i++) |
1713 | s->random_seq[i] *= s->size; | |
1714 | } | |
1715 | return 0; | |
1716 | } | |
1717 | ||
1718 | /* Initialize each random sequence freelist per cache */ | |
1719 | static void __init init_freelist_randomization(void) | |
1720 | { | |
1721 | struct kmem_cache *s; | |
1722 | ||
1723 | mutex_lock(&slab_mutex); | |
1724 | ||
1725 | list_for_each_entry(s, &slab_caches, list) | |
1726 | init_cache_random_seq(s); | |
1727 | ||
1728 | mutex_unlock(&slab_mutex); | |
1729 | } | |
1730 | ||
1731 | /* Get the next entry on the pre-computed freelist randomized */ | |
1732 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1733 | unsigned long *pos, void *start, | |
1734 | unsigned long page_limit, | |
1735 | unsigned long freelist_count) | |
1736 | { | |
1737 | unsigned int idx; | |
1738 | ||
1739 | /* | |
1740 | * If the target page allocation failed, the number of objects on the | |
1741 | * page might be smaller than the usual size defined by the cache. | |
1742 | */ | |
1743 | do { | |
1744 | idx = s->random_seq[*pos]; | |
1745 | *pos += 1; | |
1746 | if (*pos >= freelist_count) | |
1747 | *pos = 0; | |
1748 | } while (unlikely(idx >= page_limit)); | |
1749 | ||
1750 | return (char *)start + idx; | |
1751 | } | |
1752 | ||
1753 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1754 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1755 | { | |
1756 | void *start; | |
1757 | void *cur; | |
1758 | void *next; | |
1759 | unsigned long idx, pos, page_limit, freelist_count; | |
1760 | ||
1761 | if (page->objects < 2 || !s->random_seq) | |
1762 | return false; | |
1763 | ||
1764 | freelist_count = oo_objects(s->oo); | |
1765 | pos = get_random_int() % freelist_count; | |
1766 | ||
1767 | page_limit = page->objects * s->size; | |
1768 | start = fixup_red_left(s, page_address(page)); | |
1769 | ||
1770 | /* First entry is used as the base of the freelist */ | |
1771 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1772 | freelist_count); | |
4d176711 | 1773 | cur = setup_object(s, page, cur); |
210e7a43 TG |
1774 | page->freelist = cur; |
1775 | ||
1776 | for (idx = 1; idx < page->objects; idx++) { | |
210e7a43 TG |
1777 | next = next_freelist_entry(s, page, &pos, start, page_limit, |
1778 | freelist_count); | |
4d176711 | 1779 | next = setup_object(s, page, next); |
210e7a43 TG |
1780 | set_freepointer(s, cur, next); |
1781 | cur = next; | |
1782 | } | |
210e7a43 TG |
1783 | set_freepointer(s, cur, NULL); |
1784 | ||
1785 | return true; | |
1786 | } | |
1787 | #else | |
1788 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1789 | { | |
1790 | return 0; | |
1791 | } | |
1792 | static inline void init_freelist_randomization(void) { } | |
1793 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1794 | { | |
1795 | return false; | |
1796 | } | |
1797 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1798 | ||
81819f0f CL |
1799 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1800 | { | |
06428780 | 1801 | struct page *page; |
834f3d11 | 1802 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1803 | gfp_t alloc_gfp; |
4d176711 | 1804 | void *start, *p, *next; |
a50b854e | 1805 | int idx; |
210e7a43 | 1806 | bool shuffle; |
81819f0f | 1807 | |
7e0528da CL |
1808 | flags &= gfp_allowed_mask; |
1809 | ||
b7a49f0d | 1810 | flags |= s->allocflags; |
e12ba74d | 1811 | |
ba52270d PE |
1812 | /* |
1813 | * Let the initial higher-order allocation fail under memory pressure | |
1814 | * so we fall-back to the minimum order allocation. | |
1815 | */ | |
1816 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1817 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1818 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1819 | |
5dfb4175 | 1820 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1821 | if (unlikely(!page)) { |
1822 | oo = s->min; | |
80c3a998 | 1823 | alloc_gfp = flags; |
65c3376a CL |
1824 | /* |
1825 | * Allocation may have failed due to fragmentation. | |
1826 | * Try a lower order alloc if possible | |
1827 | */ | |
5dfb4175 | 1828 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1829 | if (unlikely(!page)) |
1830 | goto out; | |
1831 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1832 | } |
5a896d9e | 1833 | |
834f3d11 | 1834 | page->objects = oo_objects(oo); |
81819f0f | 1835 | |
2e9bd483 | 1836 | account_slab_page(page, oo_order(oo), s, flags); |
1f3147b4 | 1837 | |
1b4f59e3 | 1838 | page->slab_cache = s; |
c03f94cc | 1839 | __SetPageSlab(page); |
2f064f34 | 1840 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1841 | SetPageSlabPfmemalloc(page); |
81819f0f | 1842 | |
a7101224 | 1843 | kasan_poison_slab(page); |
81819f0f | 1844 | |
a7101224 | 1845 | start = page_address(page); |
81819f0f | 1846 | |
a50b854e | 1847 | setup_page_debug(s, page, start); |
0316bec2 | 1848 | |
210e7a43 TG |
1849 | shuffle = shuffle_freelist(s, page); |
1850 | ||
1851 | if (!shuffle) { | |
4d176711 AK |
1852 | start = fixup_red_left(s, start); |
1853 | start = setup_object(s, page, start); | |
1854 | page->freelist = start; | |
18e50661 AK |
1855 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1856 | next = p + s->size; | |
1857 | next = setup_object(s, page, next); | |
1858 | set_freepointer(s, p, next); | |
1859 | p = next; | |
1860 | } | |
1861 | set_freepointer(s, p, NULL); | |
81819f0f | 1862 | } |
81819f0f | 1863 | |
e6e82ea1 | 1864 | page->inuse = page->objects; |
8cb0a506 | 1865 | page->frozen = 1; |
588f8ba9 | 1866 | |
81819f0f | 1867 | out: |
588f8ba9 TG |
1868 | if (!page) |
1869 | return NULL; | |
1870 | ||
588f8ba9 TG |
1871 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1872 | ||
81819f0f CL |
1873 | return page; |
1874 | } | |
1875 | ||
588f8ba9 TG |
1876 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1877 | { | |
44405099 LL |
1878 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
1879 | flags = kmalloc_fix_flags(flags); | |
588f8ba9 | 1880 | |
53a0de06 VB |
1881 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
1882 | ||
588f8ba9 TG |
1883 | return allocate_slab(s, |
1884 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1885 | } | |
1886 | ||
81819f0f CL |
1887 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1888 | { | |
834f3d11 CL |
1889 | int order = compound_order(page); |
1890 | int pages = 1 << order; | |
81819f0f | 1891 | |
8fc8d666 | 1892 | if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) { |
81819f0f CL |
1893 | void *p; |
1894 | ||
1895 | slab_pad_check(s, page); | |
224a88be CL |
1896 | for_each_object(p, s, page_address(page), |
1897 | page->objects) | |
f7cb1933 | 1898 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1899 | } |
1900 | ||
072bb0aa | 1901 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1902 | __ClearPageSlab(page); |
0c06dd75 VB |
1903 | /* In union with page->mapping where page allocator expects NULL */ |
1904 | page->slab_cache = NULL; | |
1eb5ac64 NP |
1905 | if (current->reclaim_state) |
1906 | current->reclaim_state->reclaimed_slab += pages; | |
74d555be | 1907 | unaccount_slab_page(page, order, s); |
27ee57c9 | 1908 | __free_pages(page, order); |
81819f0f CL |
1909 | } |
1910 | ||
1911 | static void rcu_free_slab(struct rcu_head *h) | |
1912 | { | |
bf68c214 | 1913 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1914 | |
1b4f59e3 | 1915 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1916 | } |
1917 | ||
1918 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1919 | { | |
5f0d5a3a | 1920 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1921 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1922 | } else |
1923 | __free_slab(s, page); | |
1924 | } | |
1925 | ||
1926 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1927 | { | |
205ab99d | 1928 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1929 | free_slab(s, page); |
1930 | } | |
1931 | ||
1932 | /* | |
5cc6eee8 | 1933 | * Management of partially allocated slabs. |
81819f0f | 1934 | */ |
1e4dd946 SR |
1935 | static inline void |
1936 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1937 | { |
e95eed57 | 1938 | n->nr_partial++; |
136333d1 | 1939 | if (tail == DEACTIVATE_TO_TAIL) |
916ac052 | 1940 | list_add_tail(&page->slab_list, &n->partial); |
7c2e132c | 1941 | else |
916ac052 | 1942 | list_add(&page->slab_list, &n->partial); |
81819f0f CL |
1943 | } |
1944 | ||
1e4dd946 SR |
1945 | static inline void add_partial(struct kmem_cache_node *n, |
1946 | struct page *page, int tail) | |
62e346a8 | 1947 | { |
c65c1877 | 1948 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1949 | __add_partial(n, page, tail); |
1950 | } | |
c65c1877 | 1951 | |
1e4dd946 SR |
1952 | static inline void remove_partial(struct kmem_cache_node *n, |
1953 | struct page *page) | |
1954 | { | |
1955 | lockdep_assert_held(&n->list_lock); | |
916ac052 | 1956 | list_del(&page->slab_list); |
52b4b950 | 1957 | n->nr_partial--; |
1e4dd946 SR |
1958 | } |
1959 | ||
81819f0f | 1960 | /* |
7ced3719 CL |
1961 | * Remove slab from the partial list, freeze it and |
1962 | * return the pointer to the freelist. | |
81819f0f | 1963 | * |
497b66f2 | 1964 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1965 | */ |
497b66f2 | 1966 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1967 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1968 | int mode, int *objects) |
81819f0f | 1969 | { |
2cfb7455 CL |
1970 | void *freelist; |
1971 | unsigned long counters; | |
1972 | struct page new; | |
1973 | ||
c65c1877 PZ |
1974 | lockdep_assert_held(&n->list_lock); |
1975 | ||
2cfb7455 CL |
1976 | /* |
1977 | * Zap the freelist and set the frozen bit. | |
1978 | * The old freelist is the list of objects for the | |
1979 | * per cpu allocation list. | |
1980 | */ | |
7ced3719 CL |
1981 | freelist = page->freelist; |
1982 | counters = page->counters; | |
1983 | new.counters = counters; | |
633b0764 | 1984 | *objects = new.objects - new.inuse; |
23910c50 | 1985 | if (mode) { |
7ced3719 | 1986 | new.inuse = page->objects; |
23910c50 PE |
1987 | new.freelist = NULL; |
1988 | } else { | |
1989 | new.freelist = freelist; | |
1990 | } | |
2cfb7455 | 1991 | |
a0132ac0 | 1992 | VM_BUG_ON(new.frozen); |
7ced3719 | 1993 | new.frozen = 1; |
2cfb7455 | 1994 | |
7ced3719 | 1995 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1996 | freelist, counters, |
02d7633f | 1997 | new.freelist, new.counters, |
7ced3719 | 1998 | "acquire_slab")) |
7ced3719 | 1999 | return NULL; |
2cfb7455 CL |
2000 | |
2001 | remove_partial(n, page); | |
7ced3719 | 2002 | WARN_ON(!freelist); |
49e22585 | 2003 | return freelist; |
81819f0f CL |
2004 | } |
2005 | ||
633b0764 | 2006 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 2007 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 2008 | |
81819f0f | 2009 | /* |
672bba3a | 2010 | * Try to allocate a partial slab from a specific node. |
81819f0f | 2011 | */ |
8ba00bb6 | 2012 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
4b1f449d | 2013 | struct page **ret_page, gfp_t gfpflags) |
81819f0f | 2014 | { |
49e22585 CL |
2015 | struct page *page, *page2; |
2016 | void *object = NULL; | |
e5d9998f | 2017 | unsigned int available = 0; |
4b1f449d | 2018 | unsigned long flags; |
633b0764 | 2019 | int objects; |
81819f0f CL |
2020 | |
2021 | /* | |
2022 | * Racy check. If we mistakenly see no partial slabs then we | |
2023 | * just allocate an empty slab. If we mistakenly try to get a | |
70b6d25e | 2024 | * partial slab and there is none available then get_partial() |
672bba3a | 2025 | * will return NULL. |
81819f0f CL |
2026 | */ |
2027 | if (!n || !n->nr_partial) | |
2028 | return NULL; | |
2029 | ||
4b1f449d | 2030 | spin_lock_irqsave(&n->list_lock, flags); |
916ac052 | 2031 | list_for_each_entry_safe(page, page2, &n->partial, slab_list) { |
8ba00bb6 | 2032 | void *t; |
49e22585 | 2033 | |
4b1f449d | 2034 | if (!pfmemalloc_match(page, gfpflags)) |
8ba00bb6 JK |
2035 | continue; |
2036 | ||
633b0764 | 2037 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 | 2038 | if (!t) |
9b1ea29b | 2039 | break; |
49e22585 | 2040 | |
633b0764 | 2041 | available += objects; |
12d79634 | 2042 | if (!object) { |
75c8ff28 | 2043 | *ret_page = page; |
49e22585 | 2044 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 2045 | object = t; |
49e22585 | 2046 | } else { |
633b0764 | 2047 | put_cpu_partial(s, page, 0); |
8028dcea | 2048 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 2049 | } |
345c905d | 2050 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 2051 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
2052 | break; |
2053 | ||
497b66f2 | 2054 | } |
4b1f449d | 2055 | spin_unlock_irqrestore(&n->list_lock, flags); |
497b66f2 | 2056 | return object; |
81819f0f CL |
2057 | } |
2058 | ||
2059 | /* | |
672bba3a | 2060 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 2061 | */ |
de3ec035 | 2062 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
75c8ff28 | 2063 | struct page **ret_page) |
81819f0f CL |
2064 | { |
2065 | #ifdef CONFIG_NUMA | |
2066 | struct zonelist *zonelist; | |
dd1a239f | 2067 | struct zoneref *z; |
54a6eb5c | 2068 | struct zone *zone; |
97a225e6 | 2069 | enum zone_type highest_zoneidx = gfp_zone(flags); |
497b66f2 | 2070 | void *object; |
cc9a6c87 | 2071 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
2072 | |
2073 | /* | |
672bba3a CL |
2074 | * The defrag ratio allows a configuration of the tradeoffs between |
2075 | * inter node defragmentation and node local allocations. A lower | |
2076 | * defrag_ratio increases the tendency to do local allocations | |
2077 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 2078 | * |
672bba3a CL |
2079 | * If the defrag_ratio is set to 0 then kmalloc() always |
2080 | * returns node local objects. If the ratio is higher then kmalloc() | |
2081 | * may return off node objects because partial slabs are obtained | |
2082 | * from other nodes and filled up. | |
81819f0f | 2083 | * |
43efd3ea LP |
2084 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
2085 | * (which makes defrag_ratio = 1000) then every (well almost) | |
2086 | * allocation will first attempt to defrag slab caches on other nodes. | |
2087 | * This means scanning over all nodes to look for partial slabs which | |
2088 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 2089 | * with available objects. |
81819f0f | 2090 | */ |
9824601e CL |
2091 | if (!s->remote_node_defrag_ratio || |
2092 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
2093 | return NULL; |
2094 | ||
cc9a6c87 | 2095 | do { |
d26914d1 | 2096 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2097 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
97a225e6 | 2098 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
cc9a6c87 MG |
2099 | struct kmem_cache_node *n; |
2100 | ||
2101 | n = get_node(s, zone_to_nid(zone)); | |
2102 | ||
dee2f8aa | 2103 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 2104 | n->nr_partial > s->min_partial) { |
75c8ff28 | 2105 | object = get_partial_node(s, n, ret_page, flags); |
cc9a6c87 MG |
2106 | if (object) { |
2107 | /* | |
d26914d1 MG |
2108 | * Don't check read_mems_allowed_retry() |
2109 | * here - if mems_allowed was updated in | |
2110 | * parallel, that was a harmless race | |
2111 | * between allocation and the cpuset | |
2112 | * update | |
cc9a6c87 | 2113 | */ |
cc9a6c87 MG |
2114 | return object; |
2115 | } | |
c0ff7453 | 2116 | } |
81819f0f | 2117 | } |
d26914d1 | 2118 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 2119 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
2120 | return NULL; |
2121 | } | |
2122 | ||
2123 | /* | |
2124 | * Get a partial page, lock it and return it. | |
2125 | */ | |
497b66f2 | 2126 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
75c8ff28 | 2127 | struct page **ret_page) |
81819f0f | 2128 | { |
497b66f2 | 2129 | void *object; |
a561ce00 JK |
2130 | int searchnode = node; |
2131 | ||
2132 | if (node == NUMA_NO_NODE) | |
2133 | searchnode = numa_mem_id(); | |
81819f0f | 2134 | |
75c8ff28 | 2135 | object = get_partial_node(s, get_node(s, searchnode), ret_page, flags); |
497b66f2 CL |
2136 | if (object || node != NUMA_NO_NODE) |
2137 | return object; | |
81819f0f | 2138 | |
75c8ff28 | 2139 | return get_any_partial(s, flags, ret_page); |
81819f0f CL |
2140 | } |
2141 | ||
923717cb | 2142 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2143 | /* |
0d645ed1 | 2144 | * Calculate the next globally unique transaction for disambiguation |
8a5ec0ba CL |
2145 | * during cmpxchg. The transactions start with the cpu number and are then |
2146 | * incremented by CONFIG_NR_CPUS. | |
2147 | */ | |
2148 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
2149 | #else | |
2150 | /* | |
2151 | * No preemption supported therefore also no need to check for | |
2152 | * different cpus. | |
2153 | */ | |
2154 | #define TID_STEP 1 | |
2155 | #endif | |
2156 | ||
2157 | static inline unsigned long next_tid(unsigned long tid) | |
2158 | { | |
2159 | return tid + TID_STEP; | |
2160 | } | |
2161 | ||
9d5f0be0 | 2162 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2163 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2164 | { | |
2165 | return tid % TID_STEP; | |
2166 | } | |
2167 | ||
2168 | static inline unsigned long tid_to_event(unsigned long tid) | |
2169 | { | |
2170 | return tid / TID_STEP; | |
2171 | } | |
9d5f0be0 | 2172 | #endif |
8a5ec0ba CL |
2173 | |
2174 | static inline unsigned int init_tid(int cpu) | |
2175 | { | |
2176 | return cpu; | |
2177 | } | |
2178 | ||
2179 | static inline void note_cmpxchg_failure(const char *n, | |
2180 | const struct kmem_cache *s, unsigned long tid) | |
2181 | { | |
2182 | #ifdef SLUB_DEBUG_CMPXCHG | |
2183 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2184 | ||
f9f58285 | 2185 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2186 | |
923717cb | 2187 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2188 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2189 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2190 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2191 | else | |
2192 | #endif | |
2193 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2194 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2195 | tid_to_event(tid), tid_to_event(actual_tid)); |
2196 | else | |
f9f58285 | 2197 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2198 | actual_tid, tid, next_tid(tid)); |
2199 | #endif | |
4fdccdfb | 2200 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2201 | } |
2202 | ||
788e1aad | 2203 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2204 | { |
8a5ec0ba CL |
2205 | int cpu; |
2206 | ||
2207 | for_each_possible_cpu(cpu) | |
2208 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2209 | } |
2cfb7455 | 2210 | |
81819f0f | 2211 | /* |
a019d201 VB |
2212 | * Finishes removing the cpu slab. Merges cpu's freelist with page's freelist, |
2213 | * unfreezes the slabs and puts it on the proper list. | |
2214 | * Assumes the slab has been already safely taken away from kmem_cache_cpu | |
2215 | * by the caller. | |
81819f0f | 2216 | */ |
d0e0ac97 | 2217 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
a019d201 | 2218 | void *freelist) |
81819f0f | 2219 | { |
2cfb7455 | 2220 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 | 2221 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
d930ff03 | 2222 | int lock = 0, free_delta = 0; |
2cfb7455 | 2223 | enum slab_modes l = M_NONE, m = M_NONE; |
d930ff03 | 2224 | void *nextfree, *freelist_iter, *freelist_tail; |
136333d1 | 2225 | int tail = DEACTIVATE_TO_HEAD; |
3406e91b | 2226 | unsigned long flags = 0; |
2cfb7455 CL |
2227 | struct page new; |
2228 | struct page old; | |
2229 | ||
2230 | if (page->freelist) { | |
84e554e6 | 2231 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2232 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2233 | } |
2234 | ||
894b8788 | 2235 | /* |
d930ff03 VB |
2236 | * Stage one: Count the objects on cpu's freelist as free_delta and |
2237 | * remember the last object in freelist_tail for later splicing. | |
2cfb7455 | 2238 | */ |
d930ff03 VB |
2239 | freelist_tail = NULL; |
2240 | freelist_iter = freelist; | |
2241 | while (freelist_iter) { | |
2242 | nextfree = get_freepointer(s, freelist_iter); | |
2cfb7455 | 2243 | |
52f23478 DZ |
2244 | /* |
2245 | * If 'nextfree' is invalid, it is possible that the object at | |
d930ff03 VB |
2246 | * 'freelist_iter' is already corrupted. So isolate all objects |
2247 | * starting at 'freelist_iter' by skipping them. | |
52f23478 | 2248 | */ |
d930ff03 | 2249 | if (freelist_corrupted(s, page, &freelist_iter, nextfree)) |
52f23478 DZ |
2250 | break; |
2251 | ||
d930ff03 VB |
2252 | freelist_tail = freelist_iter; |
2253 | free_delta++; | |
2cfb7455 | 2254 | |
d930ff03 | 2255 | freelist_iter = nextfree; |
2cfb7455 CL |
2256 | } |
2257 | ||
894b8788 | 2258 | /* |
d930ff03 VB |
2259 | * Stage two: Unfreeze the page while splicing the per-cpu |
2260 | * freelist to the head of page's freelist. | |
2261 | * | |
2262 | * Ensure that the page is unfrozen while the list presence | |
2263 | * reflects the actual number of objects during unfreeze. | |
2cfb7455 CL |
2264 | * |
2265 | * We setup the list membership and then perform a cmpxchg | |
2266 | * with the count. If there is a mismatch then the page | |
2267 | * is not unfrozen but the page is on the wrong list. | |
2268 | * | |
2269 | * Then we restart the process which may have to remove | |
2270 | * the page from the list that we just put it on again | |
2271 | * because the number of objects in the slab may have | |
2272 | * changed. | |
894b8788 | 2273 | */ |
2cfb7455 | 2274 | redo: |
894b8788 | 2275 | |
d930ff03 VB |
2276 | old.freelist = READ_ONCE(page->freelist); |
2277 | old.counters = READ_ONCE(page->counters); | |
a0132ac0 | 2278 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2279 | |
2cfb7455 CL |
2280 | /* Determine target state of the slab */ |
2281 | new.counters = old.counters; | |
d930ff03 VB |
2282 | if (freelist_tail) { |
2283 | new.inuse -= free_delta; | |
2284 | set_freepointer(s, freelist_tail, old.freelist); | |
2cfb7455 CL |
2285 | new.freelist = freelist; |
2286 | } else | |
2287 | new.freelist = old.freelist; | |
2288 | ||
2289 | new.frozen = 0; | |
2290 | ||
8a5b20ae | 2291 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2292 | m = M_FREE; |
2293 | else if (new.freelist) { | |
2294 | m = M_PARTIAL; | |
2295 | if (!lock) { | |
2296 | lock = 1; | |
2297 | /* | |
8bb4e7a2 | 2298 | * Taking the spinlock removes the possibility |
2cfb7455 CL |
2299 | * that acquire_slab() will see a slab page that |
2300 | * is frozen | |
2301 | */ | |
3406e91b | 2302 | spin_lock_irqsave(&n->list_lock, flags); |
2cfb7455 CL |
2303 | } |
2304 | } else { | |
2305 | m = M_FULL; | |
965c4848 | 2306 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER) && !lock) { |
2cfb7455 CL |
2307 | lock = 1; |
2308 | /* | |
2309 | * This also ensures that the scanning of full | |
2310 | * slabs from diagnostic functions will not see | |
2311 | * any frozen slabs. | |
2312 | */ | |
3406e91b | 2313 | spin_lock_irqsave(&n->list_lock, flags); |
2cfb7455 CL |
2314 | } |
2315 | } | |
2316 | ||
2317 | if (l != m) { | |
2cfb7455 | 2318 | if (l == M_PARTIAL) |
2cfb7455 | 2319 | remove_partial(n, page); |
2cfb7455 | 2320 | else if (l == M_FULL) |
c65c1877 | 2321 | remove_full(s, n, page); |
2cfb7455 | 2322 | |
88349a28 | 2323 | if (m == M_PARTIAL) |
2cfb7455 | 2324 | add_partial(n, page, tail); |
88349a28 | 2325 | else if (m == M_FULL) |
2cfb7455 | 2326 | add_full(s, n, page); |
2cfb7455 CL |
2327 | } |
2328 | ||
2329 | l = m; | |
3406e91b | 2330 | if (!cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2331 | old.freelist, old.counters, |
2332 | new.freelist, new.counters, | |
2333 | "unfreezing slab")) | |
2334 | goto redo; | |
2335 | ||
2cfb7455 | 2336 | if (lock) |
3406e91b | 2337 | spin_unlock_irqrestore(&n->list_lock, flags); |
2cfb7455 | 2338 | |
88349a28 WY |
2339 | if (m == M_PARTIAL) |
2340 | stat(s, tail); | |
2341 | else if (m == M_FULL) | |
2342 | stat(s, DEACTIVATE_FULL); | |
2343 | else if (m == M_FREE) { | |
2cfb7455 CL |
2344 | stat(s, DEACTIVATE_EMPTY); |
2345 | discard_slab(s, page); | |
2346 | stat(s, FREE_SLAB); | |
894b8788 | 2347 | } |
81819f0f CL |
2348 | } |
2349 | ||
345c905d | 2350 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
fc1455f4 VB |
2351 | static void __unfreeze_partials(struct kmem_cache *s, struct page *partial_page) |
2352 | { | |
43d77867 | 2353 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
fc1455f4 | 2354 | struct page *page, *discard_page = NULL; |
7cf9f3ba | 2355 | unsigned long flags = 0; |
49e22585 | 2356 | |
c2f973ba | 2357 | while (partial_page) { |
49e22585 CL |
2358 | struct page new; |
2359 | struct page old; | |
2360 | ||
c2f973ba VB |
2361 | page = partial_page; |
2362 | partial_page = page->next; | |
43d77867 JK |
2363 | |
2364 | n2 = get_node(s, page_to_nid(page)); | |
2365 | if (n != n2) { | |
2366 | if (n) | |
7cf9f3ba | 2367 | spin_unlock_irqrestore(&n->list_lock, flags); |
43d77867 JK |
2368 | |
2369 | n = n2; | |
7cf9f3ba | 2370 | spin_lock_irqsave(&n->list_lock, flags); |
43d77867 | 2371 | } |
49e22585 CL |
2372 | |
2373 | do { | |
2374 | ||
2375 | old.freelist = page->freelist; | |
2376 | old.counters = page->counters; | |
a0132ac0 | 2377 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2378 | |
2379 | new.counters = old.counters; | |
2380 | new.freelist = old.freelist; | |
2381 | ||
2382 | new.frozen = 0; | |
2383 | ||
d24ac77f | 2384 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2385 | old.freelist, old.counters, |
2386 | new.freelist, new.counters, | |
2387 | "unfreezing slab")); | |
2388 | ||
8a5b20ae | 2389 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2390 | page->next = discard_page; |
2391 | discard_page = page; | |
43d77867 JK |
2392 | } else { |
2393 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2394 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2395 | } |
2396 | } | |
2397 | ||
2398 | if (n) | |
7cf9f3ba | 2399 | spin_unlock_irqrestore(&n->list_lock, flags); |
8de06a6f | 2400 | |
9ada1934 SL |
2401 | while (discard_page) { |
2402 | page = discard_page; | |
2403 | discard_page = discard_page->next; | |
2404 | ||
2405 | stat(s, DEACTIVATE_EMPTY); | |
2406 | discard_slab(s, page); | |
2407 | stat(s, FREE_SLAB); | |
2408 | } | |
fc1455f4 | 2409 | } |
f3ab8b6b | 2410 | |
fc1455f4 VB |
2411 | /* |
2412 | * Unfreeze all the cpu partial slabs. | |
2413 | */ | |
2414 | static void unfreeze_partials(struct kmem_cache *s) | |
2415 | { | |
2416 | struct page *partial_page; | |
2417 | unsigned long flags; | |
2418 | ||
2419 | local_irq_save(flags); | |
2420 | partial_page = this_cpu_read(s->cpu_slab->partial); | |
2421 | this_cpu_write(s->cpu_slab->partial, NULL); | |
2422 | local_irq_restore(flags); | |
2423 | ||
2424 | if (partial_page) | |
2425 | __unfreeze_partials(s, partial_page); | |
2426 | } | |
2427 | ||
2428 | static void unfreeze_partials_cpu(struct kmem_cache *s, | |
2429 | struct kmem_cache_cpu *c) | |
2430 | { | |
2431 | struct page *partial_page; | |
2432 | ||
2433 | partial_page = slub_percpu_partial(c); | |
2434 | c->partial = NULL; | |
2435 | ||
2436 | if (partial_page) | |
2437 | __unfreeze_partials(s, partial_page); | |
49e22585 CL |
2438 | } |
2439 | ||
fc1455f4 VB |
2440 | #else /* CONFIG_SLUB_CPU_PARTIAL */ |
2441 | ||
2442 | static inline void unfreeze_partials(struct kmem_cache *s) { } | |
2443 | static inline void unfreeze_partials_cpu(struct kmem_cache *s, | |
2444 | struct kmem_cache_cpu *c) { } | |
2445 | ||
2446 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ | |
2447 | ||
49e22585 | 2448 | /* |
9234bae9 WY |
2449 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a |
2450 | * partial page slot if available. | |
49e22585 CL |
2451 | * |
2452 | * If we did not find a slot then simply move all the partials to the | |
2453 | * per node partial list. | |
2454 | */ | |
633b0764 | 2455 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2456 | { |
345c905d | 2457 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2458 | struct page *oldpage; |
2459 | int pages; | |
2460 | int pobjects; | |
2461 | ||
d6e0b7fa | 2462 | preempt_disable(); |
49e22585 CL |
2463 | do { |
2464 | pages = 0; | |
2465 | pobjects = 0; | |
2466 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2467 | ||
2468 | if (oldpage) { | |
2469 | pobjects = oldpage->pobjects; | |
2470 | pages = oldpage->pages; | |
bbd4e305 | 2471 | if (drain && pobjects > slub_cpu_partial(s)) { |
49e22585 CL |
2472 | /* |
2473 | * partial array is full. Move the existing | |
2474 | * set to the per node partial list. | |
2475 | */ | |
fc1455f4 | 2476 | unfreeze_partials(s); |
e24fc410 | 2477 | oldpage = NULL; |
49e22585 CL |
2478 | pobjects = 0; |
2479 | pages = 0; | |
8028dcea | 2480 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2481 | } |
2482 | } | |
2483 | ||
2484 | pages++; | |
2485 | pobjects += page->objects - page->inuse; | |
2486 | ||
2487 | page->pages = pages; | |
2488 | page->pobjects = pobjects; | |
2489 | page->next = oldpage; | |
2490 | ||
d0e0ac97 CG |
2491 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2492 | != oldpage); | |
d6e0b7fa | 2493 | preempt_enable(); |
6dfd1b65 | 2494 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2495 | } |
2496 | ||
dfb4f096 | 2497 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2498 | { |
a019d201 VB |
2499 | void *freelist = c->freelist; |
2500 | struct page *page = c->page; | |
c17dda40 | 2501 | |
a019d201 VB |
2502 | c->page = NULL; |
2503 | c->freelist = NULL; | |
c17dda40 | 2504 | c->tid = next_tid(c->tid); |
a019d201 VB |
2505 | |
2506 | deactivate_slab(s, page, freelist); | |
2507 | ||
2508 | stat(s, CPUSLAB_FLUSH); | |
81819f0f CL |
2509 | } |
2510 | ||
0c710013 | 2511 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2512 | { |
9dfc6e68 | 2513 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2514 | |
1265ef2d WY |
2515 | if (c->page) |
2516 | flush_slab(s, c); | |
49e22585 | 2517 | |
fc1455f4 | 2518 | unfreeze_partials_cpu(s, c); |
81819f0f CL |
2519 | } |
2520 | ||
fc1455f4 VB |
2521 | /* |
2522 | * Flush cpu slab. | |
2523 | * | |
2524 | * Called from IPI handler with interrupts disabled. | |
2525 | */ | |
81819f0f CL |
2526 | static void flush_cpu_slab(void *d) |
2527 | { | |
2528 | struct kmem_cache *s = d; | |
fc1455f4 VB |
2529 | struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab); |
2530 | ||
2531 | if (c->page) | |
2532 | flush_slab(s, c); | |
81819f0f | 2533 | |
fc1455f4 | 2534 | unfreeze_partials(s); |
81819f0f CL |
2535 | } |
2536 | ||
a8364d55 GBY |
2537 | static bool has_cpu_slab(int cpu, void *info) |
2538 | { | |
2539 | struct kmem_cache *s = info; | |
2540 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2541 | ||
a93cf07b | 2542 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2543 | } |
2544 | ||
81819f0f CL |
2545 | static void flush_all(struct kmem_cache *s) |
2546 | { | |
cb923159 | 2547 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1); |
81819f0f CL |
2548 | } |
2549 | ||
a96a87bf SAS |
2550 | /* |
2551 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2552 | * necessary. | |
2553 | */ | |
2554 | static int slub_cpu_dead(unsigned int cpu) | |
2555 | { | |
2556 | struct kmem_cache *s; | |
a96a87bf SAS |
2557 | |
2558 | mutex_lock(&slab_mutex); | |
0e7ac738 | 2559 | list_for_each_entry(s, &slab_caches, list) |
a96a87bf | 2560 | __flush_cpu_slab(s, cpu); |
a96a87bf SAS |
2561 | mutex_unlock(&slab_mutex); |
2562 | return 0; | |
2563 | } | |
2564 | ||
dfb4f096 CL |
2565 | /* |
2566 | * Check if the objects in a per cpu structure fit numa | |
2567 | * locality expectations. | |
2568 | */ | |
57d437d2 | 2569 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2570 | { |
2571 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2572 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2573 | return 0; |
2574 | #endif | |
2575 | return 1; | |
2576 | } | |
2577 | ||
9a02d699 | 2578 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2579 | static int count_free(struct page *page) |
2580 | { | |
2581 | return page->objects - page->inuse; | |
2582 | } | |
2583 | ||
9a02d699 DR |
2584 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2585 | { | |
2586 | return atomic_long_read(&n->total_objects); | |
2587 | } | |
2588 | #endif /* CONFIG_SLUB_DEBUG */ | |
2589 | ||
2590 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2591 | static unsigned long count_partial(struct kmem_cache_node *n, |
2592 | int (*get_count)(struct page *)) | |
2593 | { | |
2594 | unsigned long flags; | |
2595 | unsigned long x = 0; | |
2596 | struct page *page; | |
2597 | ||
2598 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 2599 | list_for_each_entry(page, &n->partial, slab_list) |
781b2ba6 PE |
2600 | x += get_count(page); |
2601 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2602 | return x; | |
2603 | } | |
9a02d699 | 2604 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2605 | |
781b2ba6 PE |
2606 | static noinline void |
2607 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2608 | { | |
9a02d699 DR |
2609 | #ifdef CONFIG_SLUB_DEBUG |
2610 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2611 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2612 | int node; |
fa45dc25 | 2613 | struct kmem_cache_node *n; |
781b2ba6 | 2614 | |
9a02d699 DR |
2615 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2616 | return; | |
2617 | ||
5b3810e5 VB |
2618 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2619 | nid, gfpflags, &gfpflags); | |
19af27af | 2620 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2621 | s->name, s->object_size, s->size, oo_order(s->oo), |
2622 | oo_order(s->min)); | |
781b2ba6 | 2623 | |
3b0efdfa | 2624 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2625 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2626 | s->name); | |
fa5ec8a1 | 2627 | |
fa45dc25 | 2628 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2629 | unsigned long nr_slabs; |
2630 | unsigned long nr_objs; | |
2631 | unsigned long nr_free; | |
2632 | ||
26c02cf0 AB |
2633 | nr_free = count_partial(n, count_free); |
2634 | nr_slabs = node_nr_slabs(n); | |
2635 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2636 | |
f9f58285 | 2637 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2638 | node, nr_slabs, nr_objs, nr_free); |
2639 | } | |
9a02d699 | 2640 | #endif |
781b2ba6 PE |
2641 | } |
2642 | ||
072bb0aa MG |
2643 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2644 | { | |
2645 | if (unlikely(PageSlabPfmemalloc(page))) | |
2646 | return gfp_pfmemalloc_allowed(gfpflags); | |
2647 | ||
2648 | return true; | |
2649 | } | |
2650 | ||
0b303fb4 VB |
2651 | /* |
2652 | * A variant of pfmemalloc_match() that tests page flags without asserting | |
2653 | * PageSlab. Intended for opportunistic checks before taking a lock and | |
2654 | * rechecking that nobody else freed the page under us. | |
2655 | */ | |
2656 | static inline bool pfmemalloc_match_unsafe(struct page *page, gfp_t gfpflags) | |
2657 | { | |
2658 | if (unlikely(__PageSlabPfmemalloc(page))) | |
2659 | return gfp_pfmemalloc_allowed(gfpflags); | |
2660 | ||
2661 | return true; | |
2662 | } | |
2663 | ||
213eeb9f | 2664 | /* |
d0e0ac97 CG |
2665 | * Check the page->freelist of a page and either transfer the freelist to the |
2666 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2667 | * |
2668 | * The page is still frozen if the return value is not NULL. | |
2669 | * | |
2670 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2671 | * |
2672 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2673 | */ |
2674 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2675 | { | |
2676 | struct page new; | |
2677 | unsigned long counters; | |
2678 | void *freelist; | |
2679 | ||
2680 | do { | |
2681 | freelist = page->freelist; | |
2682 | counters = page->counters; | |
6faa6833 | 2683 | |
213eeb9f | 2684 | new.counters = counters; |
a0132ac0 | 2685 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2686 | |
2687 | new.inuse = page->objects; | |
2688 | new.frozen = freelist != NULL; | |
2689 | ||
d24ac77f | 2690 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2691 | freelist, counters, |
2692 | NULL, new.counters, | |
2693 | "get_freelist")); | |
2694 | ||
2695 | return freelist; | |
2696 | } | |
2697 | ||
81819f0f | 2698 | /* |
894b8788 CL |
2699 | * Slow path. The lockless freelist is empty or we need to perform |
2700 | * debugging duties. | |
2701 | * | |
894b8788 CL |
2702 | * Processing is still very fast if new objects have been freed to the |
2703 | * regular freelist. In that case we simply take over the regular freelist | |
2704 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2705 | * |
894b8788 CL |
2706 | * If that is not working then we fall back to the partial lists. We take the |
2707 | * first element of the freelist as the object to allocate now and move the | |
2708 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2709 | * |
894b8788 | 2710 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2711 | * we need to allocate a new slab. This is the slowest path since it involves |
2712 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 | 2713 | * |
e500059b | 2714 | * Version of __slab_alloc to use when we know that preemption is |
a380a3c7 | 2715 | * already disabled (which is the case for bulk allocation). |
81819f0f | 2716 | */ |
a380a3c7 | 2717 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2718 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2719 | { |
6faa6833 | 2720 | void *freelist; |
f6e7def7 | 2721 | struct page *page; |
e500059b | 2722 | unsigned long flags; |
81819f0f | 2723 | |
9f986d99 AW |
2724 | stat(s, ALLOC_SLOWPATH); |
2725 | ||
0b303fb4 VB |
2726 | reread_page: |
2727 | ||
2728 | page = READ_ONCE(c->page); | |
0715e6c5 VB |
2729 | if (!page) { |
2730 | /* | |
2731 | * if the node is not online or has no normal memory, just | |
2732 | * ignore the node constraint | |
2733 | */ | |
2734 | if (unlikely(node != NUMA_NO_NODE && | |
7e1fa93d | 2735 | !node_isset(node, slab_nodes))) |
0715e6c5 | 2736 | node = NUMA_NO_NODE; |
81819f0f | 2737 | goto new_slab; |
0715e6c5 | 2738 | } |
49e22585 | 2739 | redo: |
6faa6833 | 2740 | |
57d437d2 | 2741 | if (unlikely(!node_match(page, node))) { |
0715e6c5 VB |
2742 | /* |
2743 | * same as above but node_match() being false already | |
2744 | * implies node != NUMA_NO_NODE | |
2745 | */ | |
7e1fa93d | 2746 | if (!node_isset(node, slab_nodes)) { |
0715e6c5 VB |
2747 | node = NUMA_NO_NODE; |
2748 | goto redo; | |
2749 | } else { | |
a561ce00 | 2750 | stat(s, ALLOC_NODE_MISMATCH); |
0b303fb4 | 2751 | goto deactivate_slab; |
a561ce00 | 2752 | } |
fc59c053 | 2753 | } |
6446faa2 | 2754 | |
072bb0aa MG |
2755 | /* |
2756 | * By rights, we should be searching for a slab page that was | |
2757 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2758 | * information when the page leaves the per-cpu allocator | |
2759 | */ | |
0b303fb4 VB |
2760 | if (unlikely(!pfmemalloc_match_unsafe(page, gfpflags))) |
2761 | goto deactivate_slab; | |
072bb0aa | 2762 | |
0b303fb4 VB |
2763 | /* must check again c->page in case IRQ handler changed it */ |
2764 | local_irq_save(flags); | |
2765 | if (unlikely(page != c->page)) { | |
2766 | local_irq_restore(flags); | |
2767 | goto reread_page; | |
2768 | } | |
6faa6833 CL |
2769 | freelist = c->freelist; |
2770 | if (freelist) | |
73736e03 | 2771 | goto load_freelist; |
03e404af | 2772 | |
f6e7def7 | 2773 | freelist = get_freelist(s, page); |
6446faa2 | 2774 | |
6faa6833 | 2775 | if (!freelist) { |
03e404af | 2776 | c->page = NULL; |
fa417ab7 | 2777 | local_irq_restore(flags); |
03e404af | 2778 | stat(s, DEACTIVATE_BYPASS); |
fc59c053 | 2779 | goto new_slab; |
03e404af | 2780 | } |
6446faa2 | 2781 | |
84e554e6 | 2782 | stat(s, ALLOC_REFILL); |
6446faa2 | 2783 | |
894b8788 | 2784 | load_freelist: |
0b303fb4 VB |
2785 | |
2786 | lockdep_assert_irqs_disabled(); | |
2787 | ||
507effea CL |
2788 | /* |
2789 | * freelist is pointing to the list of objects to be used. | |
2790 | * page is pointing to the page from which the objects are obtained. | |
2791 | * That page must be frozen for per cpu allocations to work. | |
2792 | */ | |
a0132ac0 | 2793 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2794 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2795 | c->tid = next_tid(c->tid); |
e500059b | 2796 | local_irq_restore(flags); |
6faa6833 | 2797 | return freelist; |
81819f0f | 2798 | |
0b303fb4 VB |
2799 | deactivate_slab: |
2800 | ||
2801 | local_irq_save(flags); | |
2802 | if (page != c->page) { | |
2803 | local_irq_restore(flags); | |
2804 | goto reread_page; | |
2805 | } | |
a019d201 VB |
2806 | freelist = c->freelist; |
2807 | c->page = NULL; | |
2808 | c->freelist = NULL; | |
fa417ab7 | 2809 | local_irq_restore(flags); |
cfdf836e | 2810 | deactivate_slab(s, page, freelist); |
0b303fb4 | 2811 | |
81819f0f | 2812 | new_slab: |
2cfb7455 | 2813 | |
a93cf07b | 2814 | if (slub_percpu_partial(c)) { |
fa417ab7 VB |
2815 | local_irq_save(flags); |
2816 | if (unlikely(c->page)) { | |
2817 | local_irq_restore(flags); | |
2818 | goto reread_page; | |
2819 | } | |
4b1f449d VB |
2820 | if (unlikely(!slub_percpu_partial(c))) { |
2821 | local_irq_restore(flags); | |
fa417ab7 | 2822 | goto new_objects; /* stolen by an IRQ handler */ |
4b1f449d | 2823 | } |
fa417ab7 | 2824 | |
a93cf07b WY |
2825 | page = c->page = slub_percpu_partial(c); |
2826 | slub_set_percpu_partial(c, page); | |
0b303fb4 | 2827 | local_irq_restore(flags); |
49e22585 | 2828 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2829 | goto redo; |
81819f0f CL |
2830 | } |
2831 | ||
fa417ab7 VB |
2832 | new_objects: |
2833 | ||
75c8ff28 | 2834 | freelist = get_partial(s, gfpflags, node, &page); |
3f2b77e3 | 2835 | if (freelist) |
2a904905 VB |
2836 | goto check_new_page; |
2837 | ||
e500059b | 2838 | put_cpu_ptr(s->cpu_slab); |
53a0de06 | 2839 | page = new_slab(s, gfpflags, node); |
e500059b | 2840 | c = get_cpu_ptr(s->cpu_slab); |
01ad8a7b | 2841 | |
53a0de06 | 2842 | if (unlikely(!page)) { |
9a02d699 | 2843 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2844 | return NULL; |
81819f0f | 2845 | } |
2cfb7455 | 2846 | |
53a0de06 VB |
2847 | /* |
2848 | * No other reference to the page yet so we can | |
2849 | * muck around with it freely without cmpxchg | |
2850 | */ | |
2851 | freelist = page->freelist; | |
2852 | page->freelist = NULL; | |
2853 | ||
2854 | stat(s, ALLOC_SLAB); | |
53a0de06 | 2855 | |
2a904905 | 2856 | check_new_page: |
2cfb7455 | 2857 | |
1572df7c | 2858 | if (kmem_cache_debug(s)) { |
fa417ab7 | 2859 | if (!alloc_debug_processing(s, page, freelist, addr)) { |
1572df7c VB |
2860 | /* Slab failed checks. Next slab needed */ |
2861 | goto new_slab; | |
fa417ab7 | 2862 | } else { |
1572df7c VB |
2863 | /* |
2864 | * For debug case, we don't load freelist so that all | |
2865 | * allocations go through alloc_debug_processing() | |
2866 | */ | |
2867 | goto return_single; | |
fa417ab7 | 2868 | } |
1572df7c VB |
2869 | } |
2870 | ||
2871 | if (unlikely(!pfmemalloc_match(page, gfpflags))) | |
2872 | /* | |
2873 | * For !pfmemalloc_match() case we don't load freelist so that | |
2874 | * we don't make further mismatched allocations easier. | |
2875 | */ | |
2876 | goto return_single; | |
2877 | ||
cfdf836e VB |
2878 | retry_load_page: |
2879 | ||
9f101ee8 | 2880 | local_irq_save(flags); |
cfdf836e VB |
2881 | if (unlikely(c->page)) { |
2882 | void *flush_freelist = c->freelist; | |
2883 | struct page *flush_page = c->page; | |
2884 | ||
2885 | c->page = NULL; | |
2886 | c->freelist = NULL; | |
2887 | c->tid = next_tid(c->tid); | |
2888 | ||
2889 | local_irq_restore(flags); | |
2890 | ||
2891 | deactivate_slab(s, flush_page, flush_freelist); | |
2892 | ||
2893 | stat(s, CPUSLAB_FLUSH); | |
2894 | ||
2895 | goto retry_load_page; | |
2896 | } | |
3f2b77e3 VB |
2897 | c->page = page; |
2898 | ||
1572df7c VB |
2899 | goto load_freelist; |
2900 | ||
2901 | return_single: | |
894b8788 | 2902 | |
a019d201 | 2903 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
6faa6833 | 2904 | return freelist; |
894b8788 CL |
2905 | } |
2906 | ||
a380a3c7 | 2907 | /* |
e500059b VB |
2908 | * A wrapper for ___slab_alloc() for contexts where preemption is not yet |
2909 | * disabled. Compensates for possible cpu changes by refetching the per cpu area | |
2910 | * pointer. | |
a380a3c7 CL |
2911 | */ |
2912 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2913 | unsigned long addr, struct kmem_cache_cpu *c) | |
2914 | { | |
2915 | void *p; | |
a380a3c7 | 2916 | |
e500059b | 2917 | #ifdef CONFIG_PREEMPT_COUNT |
a380a3c7 CL |
2918 | /* |
2919 | * We may have been preempted and rescheduled on a different | |
e500059b | 2920 | * cpu before disabling preemption. Need to reload cpu area |
a380a3c7 CL |
2921 | * pointer. |
2922 | */ | |
e500059b | 2923 | c = get_cpu_ptr(s->cpu_slab); |
a380a3c7 CL |
2924 | #endif |
2925 | ||
2926 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
e500059b VB |
2927 | #ifdef CONFIG_PREEMPT_COUNT |
2928 | put_cpu_ptr(s->cpu_slab); | |
2929 | #endif | |
a380a3c7 CL |
2930 | return p; |
2931 | } | |
2932 | ||
0f181f9f AP |
2933 | /* |
2934 | * If the object has been wiped upon free, make sure it's fully initialized by | |
2935 | * zeroing out freelist pointer. | |
2936 | */ | |
2937 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
2938 | void *obj) | |
2939 | { | |
2940 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
ce5716c6 AK |
2941 | memset((void *)((char *)kasan_reset_tag(obj) + s->offset), |
2942 | 0, sizeof(void *)); | |
0f181f9f AP |
2943 | } |
2944 | ||
894b8788 CL |
2945 | /* |
2946 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2947 | * have the fastpath folded into their functions. So no function call | |
2948 | * overhead for requests that can be satisfied on the fastpath. | |
2949 | * | |
2950 | * The fastpath works by first checking if the lockless freelist can be used. | |
2951 | * If not then __slab_alloc is called for slow processing. | |
2952 | * | |
2953 | * Otherwise we can simply pick the next object from the lockless free list. | |
2954 | */ | |
2b847c3c | 2955 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
b89fb5ef | 2956 | gfp_t gfpflags, int node, unsigned long addr, size_t orig_size) |
894b8788 | 2957 | { |
03ec0ed5 | 2958 | void *object; |
dfb4f096 | 2959 | struct kmem_cache_cpu *c; |
57d437d2 | 2960 | struct page *page; |
8a5ec0ba | 2961 | unsigned long tid; |
964d4bd3 | 2962 | struct obj_cgroup *objcg = NULL; |
da844b78 | 2963 | bool init = false; |
1f84260c | 2964 | |
964d4bd3 | 2965 | s = slab_pre_alloc_hook(s, &objcg, 1, gfpflags); |
8135be5a | 2966 | if (!s) |
773ff60e | 2967 | return NULL; |
b89fb5ef AP |
2968 | |
2969 | object = kfence_alloc(s, orig_size, gfpflags); | |
2970 | if (unlikely(object)) | |
2971 | goto out; | |
2972 | ||
8a5ec0ba | 2973 | redo: |
8a5ec0ba CL |
2974 | /* |
2975 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2976 | * enabled. We may switch back and forth between cpus while | |
2977 | * reading from one cpu area. That does not matter as long | |
2978 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2979 | * |
9b4bc85a VB |
2980 | * We must guarantee that tid and kmem_cache_cpu are retrieved on the |
2981 | * same cpu. We read first the kmem_cache_cpu pointer and use it to read | |
2982 | * the tid. If we are preempted and switched to another cpu between the | |
2983 | * two reads, it's OK as the two are still associated with the same cpu | |
2984 | * and cmpxchg later will validate the cpu. | |
8a5ec0ba | 2985 | */ |
9b4bc85a VB |
2986 | c = raw_cpu_ptr(s->cpu_slab); |
2987 | tid = READ_ONCE(c->tid); | |
9aabf810 JK |
2988 | |
2989 | /* | |
2990 | * Irqless object alloc/free algorithm used here depends on sequence | |
2991 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2992 | * on c to guarantee that object and page associated with previous tid | |
2993 | * won't be used with current tid. If we fetch tid first, object and | |
2994 | * page could be one associated with next tid and our alloc/free | |
2995 | * request will be failed. In this case, we will retry. So, no problem. | |
2996 | */ | |
2997 | barrier(); | |
8a5ec0ba | 2998 | |
8a5ec0ba CL |
2999 | /* |
3000 | * The transaction ids are globally unique per cpu and per operation on | |
3001 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
3002 | * occurs on the right processor and that there was no operation on the | |
3003 | * linked list in between. | |
3004 | */ | |
8a5ec0ba | 3005 | |
9dfc6e68 | 3006 | object = c->freelist; |
57d437d2 | 3007 | page = c->page; |
22e4663e | 3008 | if (unlikely(!object || !page || !node_match(page, node))) { |
dfb4f096 | 3009 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 | 3010 | } else { |
0ad9500e ED |
3011 | void *next_object = get_freepointer_safe(s, object); |
3012 | ||
8a5ec0ba | 3013 | /* |
25985edc | 3014 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
3015 | * operation and if we are on the right processor. |
3016 | * | |
d0e0ac97 CG |
3017 | * The cmpxchg does the following atomically (without lock |
3018 | * semantics!) | |
8a5ec0ba CL |
3019 | * 1. Relocate first pointer to the current per cpu area. |
3020 | * 2. Verify that tid and freelist have not been changed | |
3021 | * 3. If they were not changed replace tid and freelist | |
3022 | * | |
d0e0ac97 CG |
3023 | * Since this is without lock semantics the protection is only |
3024 | * against code executing on this cpu *not* from access by | |
3025 | * other cpus. | |
8a5ec0ba | 3026 | */ |
933393f5 | 3027 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
3028 | s->cpu_slab->freelist, s->cpu_slab->tid, |
3029 | object, tid, | |
0ad9500e | 3030 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
3031 | |
3032 | note_cmpxchg_failure("slab_alloc", s, tid); | |
3033 | goto redo; | |
3034 | } | |
0ad9500e | 3035 | prefetch_freepointer(s, next_object); |
84e554e6 | 3036 | stat(s, ALLOC_FASTPATH); |
894b8788 | 3037 | } |
0f181f9f | 3038 | |
ce5716c6 | 3039 | maybe_wipe_obj_freeptr(s, object); |
da844b78 | 3040 | init = slab_want_init_on_alloc(gfpflags, s); |
d07dbea4 | 3041 | |
b89fb5ef | 3042 | out: |
da844b78 | 3043 | slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init); |
5a896d9e | 3044 | |
894b8788 | 3045 | return object; |
81819f0f CL |
3046 | } |
3047 | ||
2b847c3c | 3048 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
b89fb5ef | 3049 | gfp_t gfpflags, unsigned long addr, size_t orig_size) |
2b847c3c | 3050 | { |
b89fb5ef | 3051 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr, orig_size); |
2b847c3c EG |
3052 | } |
3053 | ||
81819f0f CL |
3054 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
3055 | { | |
b89fb5ef | 3056 | void *ret = slab_alloc(s, gfpflags, _RET_IP_, s->object_size); |
5b882be4 | 3057 | |
d0e0ac97 CG |
3058 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
3059 | s->size, gfpflags); | |
5b882be4 EGM |
3060 | |
3061 | return ret; | |
81819f0f CL |
3062 | } |
3063 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3064 | ||
0f24f128 | 3065 | #ifdef CONFIG_TRACING |
4a92379b RK |
3066 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
3067 | { | |
b89fb5ef | 3068 | void *ret = slab_alloc(s, gfpflags, _RET_IP_, size); |
4a92379b | 3069 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 3070 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
3071 | return ret; |
3072 | } | |
3073 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
3074 | #endif |
3075 | ||
81819f0f CL |
3076 | #ifdef CONFIG_NUMA |
3077 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
3078 | { | |
b89fb5ef | 3079 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, s->object_size); |
5b882be4 | 3080 | |
ca2b84cb | 3081 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 3082 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
3083 | |
3084 | return ret; | |
81819f0f CL |
3085 | } |
3086 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 3087 | |
0f24f128 | 3088 | #ifdef CONFIG_TRACING |
4a92379b | 3089 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 3090 | gfp_t gfpflags, |
4a92379b | 3091 | int node, size_t size) |
5b882be4 | 3092 | { |
b89fb5ef | 3093 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, size); |
4a92379b RK |
3094 | |
3095 | trace_kmalloc_node(_RET_IP_, ret, | |
3096 | size, s->size, gfpflags, node); | |
0316bec2 | 3097 | |
0116523c | 3098 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 3099 | return ret; |
5b882be4 | 3100 | } |
4a92379b | 3101 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 3102 | #endif |
6dfd1b65 | 3103 | #endif /* CONFIG_NUMA */ |
5b882be4 | 3104 | |
81819f0f | 3105 | /* |
94e4d712 | 3106 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 3107 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 3108 | * |
894b8788 CL |
3109 | * So we still attempt to reduce cache line usage. Just take the slab |
3110 | * lock and free the item. If there is no additional partial page | |
3111 | * handling required then we can return immediately. | |
81819f0f | 3112 | */ |
894b8788 | 3113 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
3114 | void *head, void *tail, int cnt, |
3115 | unsigned long addr) | |
3116 | ||
81819f0f CL |
3117 | { |
3118 | void *prior; | |
2cfb7455 | 3119 | int was_frozen; |
2cfb7455 CL |
3120 | struct page new; |
3121 | unsigned long counters; | |
3122 | struct kmem_cache_node *n = NULL; | |
3f649ab7 | 3123 | unsigned long flags; |
81819f0f | 3124 | |
8a5ec0ba | 3125 | stat(s, FREE_SLOWPATH); |
81819f0f | 3126 | |
b89fb5ef AP |
3127 | if (kfence_free(head)) |
3128 | return; | |
3129 | ||
19c7ff9e | 3130 | if (kmem_cache_debug(s) && |
282acb43 | 3131 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 3132 | return; |
6446faa2 | 3133 | |
2cfb7455 | 3134 | do { |
837d678d JK |
3135 | if (unlikely(n)) { |
3136 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3137 | n = NULL; | |
3138 | } | |
2cfb7455 CL |
3139 | prior = page->freelist; |
3140 | counters = page->counters; | |
81084651 | 3141 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
3142 | new.counters = counters; |
3143 | was_frozen = new.frozen; | |
81084651 | 3144 | new.inuse -= cnt; |
837d678d | 3145 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 3146 | |
c65c1877 | 3147 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
3148 | |
3149 | /* | |
d0e0ac97 CG |
3150 | * Slab was on no list before and will be |
3151 | * partially empty | |
3152 | * We can defer the list move and instead | |
3153 | * freeze it. | |
49e22585 CL |
3154 | */ |
3155 | new.frozen = 1; | |
3156 | ||
c65c1877 | 3157 | } else { /* Needs to be taken off a list */ |
49e22585 | 3158 | |
b455def2 | 3159 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
3160 | /* |
3161 | * Speculatively acquire the list_lock. | |
3162 | * If the cmpxchg does not succeed then we may | |
3163 | * drop the list_lock without any processing. | |
3164 | * | |
3165 | * Otherwise the list_lock will synchronize with | |
3166 | * other processors updating the list of slabs. | |
3167 | */ | |
3168 | spin_lock_irqsave(&n->list_lock, flags); | |
3169 | ||
3170 | } | |
2cfb7455 | 3171 | } |
81819f0f | 3172 | |
2cfb7455 CL |
3173 | } while (!cmpxchg_double_slab(s, page, |
3174 | prior, counters, | |
81084651 | 3175 | head, new.counters, |
2cfb7455 | 3176 | "__slab_free")); |
81819f0f | 3177 | |
2cfb7455 | 3178 | if (likely(!n)) { |
49e22585 | 3179 | |
c270cf30 AW |
3180 | if (likely(was_frozen)) { |
3181 | /* | |
3182 | * The list lock was not taken therefore no list | |
3183 | * activity can be necessary. | |
3184 | */ | |
3185 | stat(s, FREE_FROZEN); | |
3186 | } else if (new.frozen) { | |
3187 | /* | |
3188 | * If we just froze the page then put it onto the | |
3189 | * per cpu partial list. | |
3190 | */ | |
49e22585 | 3191 | put_cpu_partial(s, page, 1); |
8028dcea AS |
3192 | stat(s, CPU_PARTIAL_FREE); |
3193 | } | |
c270cf30 | 3194 | |
b455def2 L |
3195 | return; |
3196 | } | |
81819f0f | 3197 | |
8a5b20ae | 3198 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
3199 | goto slab_empty; |
3200 | ||
81819f0f | 3201 | /* |
837d678d JK |
3202 | * Objects left in the slab. If it was not on the partial list before |
3203 | * then add it. | |
81819f0f | 3204 | */ |
345c905d | 3205 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
a4d3f891 | 3206 | remove_full(s, n, page); |
837d678d JK |
3207 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
3208 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 3209 | } |
80f08c19 | 3210 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
3211 | return; |
3212 | ||
3213 | slab_empty: | |
a973e9dd | 3214 | if (prior) { |
81819f0f | 3215 | /* |
6fbabb20 | 3216 | * Slab on the partial list. |
81819f0f | 3217 | */ |
5cc6eee8 | 3218 | remove_partial(n, page); |
84e554e6 | 3219 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 3220 | } else { |
6fbabb20 | 3221 | /* Slab must be on the full list */ |
c65c1877 PZ |
3222 | remove_full(s, n, page); |
3223 | } | |
2cfb7455 | 3224 | |
80f08c19 | 3225 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 3226 | stat(s, FREE_SLAB); |
81819f0f | 3227 | discard_slab(s, page); |
81819f0f CL |
3228 | } |
3229 | ||
894b8788 CL |
3230 | /* |
3231 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
3232 | * can perform fastpath freeing without additional function calls. | |
3233 | * | |
3234 | * The fastpath is only possible if we are freeing to the current cpu slab | |
3235 | * of this processor. This typically the case if we have just allocated | |
3236 | * the item before. | |
3237 | * | |
3238 | * If fastpath is not possible then fall back to __slab_free where we deal | |
3239 | * with all sorts of special processing. | |
81084651 JDB |
3240 | * |
3241 | * Bulk free of a freelist with several objects (all pointing to the | |
3242 | * same page) possible by specifying head and tail ptr, plus objects | |
3243 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 3244 | */ |
80a9201a AP |
3245 | static __always_inline void do_slab_free(struct kmem_cache *s, |
3246 | struct page *page, void *head, void *tail, | |
3247 | int cnt, unsigned long addr) | |
894b8788 | 3248 | { |
81084651 | 3249 | void *tail_obj = tail ? : head; |
dfb4f096 | 3250 | struct kmem_cache_cpu *c; |
8a5ec0ba | 3251 | unsigned long tid; |
964d4bd3 | 3252 | |
d1b2cf6c | 3253 | memcg_slab_free_hook(s, &head, 1); |
8a5ec0ba CL |
3254 | redo: |
3255 | /* | |
3256 | * Determine the currently cpus per cpu slab. | |
3257 | * The cpu may change afterward. However that does not matter since | |
3258 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 3259 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 3260 | */ |
9b4bc85a VB |
3261 | c = raw_cpu_ptr(s->cpu_slab); |
3262 | tid = READ_ONCE(c->tid); | |
c016b0bd | 3263 | |
9aabf810 JK |
3264 | /* Same with comment on barrier() in slab_alloc_node() */ |
3265 | barrier(); | |
c016b0bd | 3266 | |
442b06bc | 3267 | if (likely(page == c->page)) { |
5076190d LT |
3268 | void **freelist = READ_ONCE(c->freelist); |
3269 | ||
3270 | set_freepointer(s, tail_obj, freelist); | |
8a5ec0ba | 3271 | |
933393f5 | 3272 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba | 3273 | s->cpu_slab->freelist, s->cpu_slab->tid, |
5076190d | 3274 | freelist, tid, |
81084651 | 3275 | head, next_tid(tid)))) { |
8a5ec0ba CL |
3276 | |
3277 | note_cmpxchg_failure("slab_free", s, tid); | |
3278 | goto redo; | |
3279 | } | |
84e554e6 | 3280 | stat(s, FREE_FASTPATH); |
894b8788 | 3281 | } else |
81084651 | 3282 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 3283 | |
894b8788 CL |
3284 | } |
3285 | ||
80a9201a AP |
3286 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
3287 | void *head, void *tail, int cnt, | |
3288 | unsigned long addr) | |
3289 | { | |
80a9201a | 3290 | /* |
c3895391 AK |
3291 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3292 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3293 | */ |
c3895391 AK |
3294 | if (slab_free_freelist_hook(s, &head, &tail)) |
3295 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
3296 | } |
3297 | ||
2bd926b4 | 3298 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3299 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3300 | { | |
3301 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3302 | } | |
3303 | #endif | |
3304 | ||
81819f0f CL |
3305 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3306 | { | |
b9ce5ef4 GC |
3307 | s = cache_from_obj(s, x); |
3308 | if (!s) | |
79576102 | 3309 | return; |
81084651 | 3310 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
3544de8e | 3311 | trace_kmem_cache_free(_RET_IP_, x, s->name); |
81819f0f CL |
3312 | } |
3313 | EXPORT_SYMBOL(kmem_cache_free); | |
3314 | ||
d0ecd894 | 3315 | struct detached_freelist { |
fbd02630 | 3316 | struct page *page; |
d0ecd894 JDB |
3317 | void *tail; |
3318 | void *freelist; | |
3319 | int cnt; | |
376bf125 | 3320 | struct kmem_cache *s; |
d0ecd894 | 3321 | }; |
fbd02630 | 3322 | |
1ed7ce57 | 3323 | static inline void free_nonslab_page(struct page *page, void *object) |
f227f0fa SB |
3324 | { |
3325 | unsigned int order = compound_order(page); | |
3326 | ||
3327 | VM_BUG_ON_PAGE(!PageCompound(page), page); | |
1ed7ce57 | 3328 | kfree_hook(object); |
f227f0fa SB |
3329 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, -(PAGE_SIZE << order)); |
3330 | __free_pages(page, order); | |
3331 | } | |
3332 | ||
d0ecd894 JDB |
3333 | /* |
3334 | * This function progressively scans the array with free objects (with | |
3335 | * a limited look ahead) and extract objects belonging to the same | |
3336 | * page. It builds a detached freelist directly within the given | |
3337 | * page/objects. This can happen without any need for | |
3338 | * synchronization, because the objects are owned by running process. | |
3339 | * The freelist is build up as a single linked list in the objects. | |
3340 | * The idea is, that this detached freelist can then be bulk | |
3341 | * transferred to the real freelist(s), but only requiring a single | |
3342 | * synchronization primitive. Look ahead in the array is limited due | |
3343 | * to performance reasons. | |
3344 | */ | |
376bf125 JDB |
3345 | static inline |
3346 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3347 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3348 | { |
3349 | size_t first_skipped_index = 0; | |
3350 | int lookahead = 3; | |
3351 | void *object; | |
ca257195 | 3352 | struct page *page; |
fbd02630 | 3353 | |
d0ecd894 JDB |
3354 | /* Always re-init detached_freelist */ |
3355 | df->page = NULL; | |
fbd02630 | 3356 | |
d0ecd894 JDB |
3357 | do { |
3358 | object = p[--size]; | |
ca257195 | 3359 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3360 | } while (!object && size); |
3eed034d | 3361 | |
d0ecd894 JDB |
3362 | if (!object) |
3363 | return 0; | |
fbd02630 | 3364 | |
ca257195 JDB |
3365 | page = virt_to_head_page(object); |
3366 | if (!s) { | |
3367 | /* Handle kalloc'ed objects */ | |
3368 | if (unlikely(!PageSlab(page))) { | |
1ed7ce57 | 3369 | free_nonslab_page(page, object); |
ca257195 JDB |
3370 | p[size] = NULL; /* mark object processed */ |
3371 | return size; | |
3372 | } | |
3373 | /* Derive kmem_cache from object */ | |
3374 | df->s = page->slab_cache; | |
3375 | } else { | |
3376 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3377 | } | |
376bf125 | 3378 | |
b89fb5ef | 3379 | if (is_kfence_address(object)) { |
d57a964e | 3380 | slab_free_hook(df->s, object, false); |
b89fb5ef AP |
3381 | __kfence_free(object); |
3382 | p[size] = NULL; /* mark object processed */ | |
3383 | return size; | |
3384 | } | |
3385 | ||
d0ecd894 | 3386 | /* Start new detached freelist */ |
ca257195 | 3387 | df->page = page; |
376bf125 | 3388 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3389 | df->tail = object; |
3390 | df->freelist = object; | |
3391 | p[size] = NULL; /* mark object processed */ | |
3392 | df->cnt = 1; | |
3393 | ||
3394 | while (size) { | |
3395 | object = p[--size]; | |
3396 | if (!object) | |
3397 | continue; /* Skip processed objects */ | |
3398 | ||
3399 | /* df->page is always set at this point */ | |
3400 | if (df->page == virt_to_head_page(object)) { | |
3401 | /* Opportunity build freelist */ | |
376bf125 | 3402 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3403 | df->freelist = object; |
3404 | df->cnt++; | |
3405 | p[size] = NULL; /* mark object processed */ | |
3406 | ||
3407 | continue; | |
fbd02630 | 3408 | } |
d0ecd894 JDB |
3409 | |
3410 | /* Limit look ahead search */ | |
3411 | if (!--lookahead) | |
3412 | break; | |
3413 | ||
3414 | if (!first_skipped_index) | |
3415 | first_skipped_index = size + 1; | |
fbd02630 | 3416 | } |
d0ecd894 JDB |
3417 | |
3418 | return first_skipped_index; | |
3419 | } | |
3420 | ||
d0ecd894 | 3421 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3422 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3423 | { |
3424 | if (WARN_ON(!size)) | |
3425 | return; | |
3426 | ||
d1b2cf6c | 3427 | memcg_slab_free_hook(s, p, size); |
d0ecd894 JDB |
3428 | do { |
3429 | struct detached_freelist df; | |
3430 | ||
3431 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3432 | if (!df.page) |
d0ecd894 JDB |
3433 | continue; |
3434 | ||
457c82c3 | 3435 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt, _RET_IP_); |
d0ecd894 | 3436 | } while (likely(size)); |
484748f0 CL |
3437 | } |
3438 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3439 | ||
994eb764 | 3440 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3441 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3442 | void **p) | |
484748f0 | 3443 | { |
994eb764 JDB |
3444 | struct kmem_cache_cpu *c; |
3445 | int i; | |
964d4bd3 | 3446 | struct obj_cgroup *objcg = NULL; |
994eb764 | 3447 | |
03ec0ed5 | 3448 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3449 | s = slab_pre_alloc_hook(s, &objcg, size, flags); |
03ec0ed5 JDB |
3450 | if (unlikely(!s)) |
3451 | return false; | |
994eb764 JDB |
3452 | /* |
3453 | * Drain objects in the per cpu slab, while disabling local | |
3454 | * IRQs, which protects against PREEMPT and interrupts | |
3455 | * handlers invoking normal fastpath. | |
3456 | */ | |
e500059b | 3457 | c = get_cpu_ptr(s->cpu_slab); |
994eb764 | 3458 | local_irq_disable(); |
994eb764 JDB |
3459 | |
3460 | for (i = 0; i < size; i++) { | |
b89fb5ef | 3461 | void *object = kfence_alloc(s, s->object_size, flags); |
994eb764 | 3462 | |
b89fb5ef AP |
3463 | if (unlikely(object)) { |
3464 | p[i] = object; | |
3465 | continue; | |
3466 | } | |
3467 | ||
3468 | object = c->freelist; | |
ebe909e0 | 3469 | if (unlikely(!object)) { |
fd4d9c7d JH |
3470 | /* |
3471 | * We may have removed an object from c->freelist using | |
3472 | * the fastpath in the previous iteration; in that case, | |
3473 | * c->tid has not been bumped yet. | |
3474 | * Since ___slab_alloc() may reenable interrupts while | |
3475 | * allocating memory, we should bump c->tid now. | |
3476 | */ | |
3477 | c->tid = next_tid(c->tid); | |
3478 | ||
e500059b VB |
3479 | local_irq_enable(); |
3480 | ||
ebe909e0 JDB |
3481 | /* |
3482 | * Invoking slow path likely have side-effect | |
3483 | * of re-populating per CPU c->freelist | |
3484 | */ | |
87098373 | 3485 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3486 | _RET_IP_, c); |
87098373 CL |
3487 | if (unlikely(!p[i])) |
3488 | goto error; | |
3489 | ||
ebe909e0 | 3490 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
3491 | maybe_wipe_obj_freeptr(s, p[i]); |
3492 | ||
e500059b VB |
3493 | local_irq_disable(); |
3494 | ||
ebe909e0 JDB |
3495 | continue; /* goto for-loop */ |
3496 | } | |
994eb764 JDB |
3497 | c->freelist = get_freepointer(s, object); |
3498 | p[i] = object; | |
0f181f9f | 3499 | maybe_wipe_obj_freeptr(s, p[i]); |
994eb764 JDB |
3500 | } |
3501 | c->tid = next_tid(c->tid); | |
3502 | local_irq_enable(); | |
e500059b | 3503 | put_cpu_ptr(s->cpu_slab); |
994eb764 | 3504 | |
da844b78 AK |
3505 | /* |
3506 | * memcg and kmem_cache debug support and memory initialization. | |
3507 | * Done outside of the IRQ disabled fastpath loop. | |
3508 | */ | |
3509 | slab_post_alloc_hook(s, objcg, flags, size, p, | |
3510 | slab_want_init_on_alloc(flags, s)); | |
865762a8 | 3511 | return i; |
87098373 | 3512 | error: |
e500059b | 3513 | put_cpu_ptr(s->cpu_slab); |
da844b78 | 3514 | slab_post_alloc_hook(s, objcg, flags, i, p, false); |
03ec0ed5 | 3515 | __kmem_cache_free_bulk(s, i, p); |
865762a8 | 3516 | return 0; |
484748f0 CL |
3517 | } |
3518 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3519 | ||
3520 | ||
81819f0f | 3521 | /* |
672bba3a CL |
3522 | * Object placement in a slab is made very easy because we always start at |
3523 | * offset 0. If we tune the size of the object to the alignment then we can | |
3524 | * get the required alignment by putting one properly sized object after | |
3525 | * another. | |
81819f0f CL |
3526 | * |
3527 | * Notice that the allocation order determines the sizes of the per cpu | |
3528 | * caches. Each processor has always one slab available for allocations. | |
3529 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3530 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3531 | * locking overhead. |
81819f0f CL |
3532 | */ |
3533 | ||
3534 | /* | |
f0953a1b | 3535 | * Minimum / Maximum order of slab pages. This influences locking overhead |
81819f0f CL |
3536 | * and slab fragmentation. A higher order reduces the number of partial slabs |
3537 | * and increases the number of allocations possible without having to | |
3538 | * take the list_lock. | |
3539 | */ | |
19af27af AD |
3540 | static unsigned int slub_min_order; |
3541 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3542 | static unsigned int slub_min_objects; | |
81819f0f | 3543 | |
81819f0f CL |
3544 | /* |
3545 | * Calculate the order of allocation given an slab object size. | |
3546 | * | |
672bba3a CL |
3547 | * The order of allocation has significant impact on performance and other |
3548 | * system components. Generally order 0 allocations should be preferred since | |
3549 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3550 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3551 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3552 | * would be wasted. |
3553 | * | |
3554 | * In order to reach satisfactory performance we must ensure that a minimum | |
3555 | * number of objects is in one slab. Otherwise we may generate too much | |
3556 | * activity on the partial lists which requires taking the list_lock. This is | |
3557 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3558 | * |
672bba3a CL |
3559 | * slub_max_order specifies the order where we begin to stop considering the |
3560 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3561 | * we try to keep the page order as low as possible. So we accept more waste | |
3562 | * of space in favor of a small page order. | |
81819f0f | 3563 | * |
672bba3a CL |
3564 | * Higher order allocations also allow the placement of more objects in a |
3565 | * slab and thereby reduce object handling overhead. If the user has | |
dc84207d | 3566 | * requested a higher minimum order then we start with that one instead of |
672bba3a | 3567 | * the smallest order which will fit the object. |
81819f0f | 3568 | */ |
19af27af AD |
3569 | static inline unsigned int slab_order(unsigned int size, |
3570 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3571 | unsigned int fract_leftover) |
81819f0f | 3572 | { |
19af27af AD |
3573 | unsigned int min_order = slub_min_order; |
3574 | unsigned int order; | |
81819f0f | 3575 | |
9736d2a9 | 3576 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3577 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3578 | |
9736d2a9 | 3579 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3580 | order <= max_order; order++) { |
81819f0f | 3581 | |
19af27af AD |
3582 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3583 | unsigned int rem; | |
81819f0f | 3584 | |
9736d2a9 | 3585 | rem = slab_size % size; |
81819f0f | 3586 | |
5e6d444e | 3587 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3588 | break; |
81819f0f | 3589 | } |
672bba3a | 3590 | |
81819f0f CL |
3591 | return order; |
3592 | } | |
3593 | ||
9736d2a9 | 3594 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3595 | { |
19af27af AD |
3596 | unsigned int order; |
3597 | unsigned int min_objects; | |
3598 | unsigned int max_objects; | |
3286222f | 3599 | unsigned int nr_cpus; |
5e6d444e CL |
3600 | |
3601 | /* | |
3602 | * Attempt to find best configuration for a slab. This | |
3603 | * works by first attempting to generate a layout with | |
3604 | * the best configuration and backing off gradually. | |
3605 | * | |
422ff4d7 | 3606 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3607 | * we reduce the minimum objects required in a slab. |
3608 | */ | |
3609 | min_objects = slub_min_objects; | |
3286222f VB |
3610 | if (!min_objects) { |
3611 | /* | |
3612 | * Some architectures will only update present cpus when | |
3613 | * onlining them, so don't trust the number if it's just 1. But | |
3614 | * we also don't want to use nr_cpu_ids always, as on some other | |
3615 | * architectures, there can be many possible cpus, but never | |
3616 | * onlined. Here we compromise between trying to avoid too high | |
3617 | * order on systems that appear larger than they are, and too | |
3618 | * low order on systems that appear smaller than they are. | |
3619 | */ | |
3620 | nr_cpus = num_present_cpus(); | |
3621 | if (nr_cpus <= 1) | |
3622 | nr_cpus = nr_cpu_ids; | |
3623 | min_objects = 4 * (fls(nr_cpus) + 1); | |
3624 | } | |
9736d2a9 | 3625 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3626 | min_objects = min(min_objects, max_objects); |
3627 | ||
5e6d444e | 3628 | while (min_objects > 1) { |
19af27af AD |
3629 | unsigned int fraction; |
3630 | ||
c124f5b5 | 3631 | fraction = 16; |
5e6d444e CL |
3632 | while (fraction >= 4) { |
3633 | order = slab_order(size, min_objects, | |
9736d2a9 | 3634 | slub_max_order, fraction); |
5e6d444e CL |
3635 | if (order <= slub_max_order) |
3636 | return order; | |
3637 | fraction /= 2; | |
3638 | } | |
5086c389 | 3639 | min_objects--; |
5e6d444e CL |
3640 | } |
3641 | ||
3642 | /* | |
3643 | * We were unable to place multiple objects in a slab. Now | |
3644 | * lets see if we can place a single object there. | |
3645 | */ | |
9736d2a9 | 3646 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3647 | if (order <= slub_max_order) |
3648 | return order; | |
3649 | ||
3650 | /* | |
3651 | * Doh this slab cannot be placed using slub_max_order. | |
3652 | */ | |
9736d2a9 | 3653 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3654 | if (order < MAX_ORDER) |
5e6d444e CL |
3655 | return order; |
3656 | return -ENOSYS; | |
3657 | } | |
3658 | ||
5595cffc | 3659 | static void |
4053497d | 3660 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3661 | { |
3662 | n->nr_partial = 0; | |
81819f0f CL |
3663 | spin_lock_init(&n->list_lock); |
3664 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3665 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3666 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3667 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3668 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3669 | #endif |
81819f0f CL |
3670 | } |
3671 | ||
55136592 | 3672 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3673 | { |
6c182dc0 | 3674 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3675 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3676 | |
8a5ec0ba | 3677 | /* |
d4d84fef CM |
3678 | * Must align to double word boundary for the double cmpxchg |
3679 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3680 | */ |
d4d84fef CM |
3681 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3682 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3683 | |
3684 | if (!s->cpu_slab) | |
3685 | return 0; | |
3686 | ||
3687 | init_kmem_cache_cpus(s); | |
4c93c355 | 3688 | |
8a5ec0ba | 3689 | return 1; |
4c93c355 | 3690 | } |
4c93c355 | 3691 | |
51df1142 CL |
3692 | static struct kmem_cache *kmem_cache_node; |
3693 | ||
81819f0f CL |
3694 | /* |
3695 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3696 | * slab on the node for this slabcache. There are no concurrent accesses | |
3697 | * possible. | |
3698 | * | |
721ae22a ZYW |
3699 | * Note that this function only works on the kmem_cache_node |
3700 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3701 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3702 | */ |
55136592 | 3703 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3704 | { |
3705 | struct page *page; | |
3706 | struct kmem_cache_node *n; | |
3707 | ||
51df1142 | 3708 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3709 | |
51df1142 | 3710 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3711 | |
3712 | BUG_ON(!page); | |
a2f92ee7 | 3713 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3714 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3715 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3716 | } |
3717 | ||
81819f0f CL |
3718 | n = page->freelist; |
3719 | BUG_ON(!n); | |
8ab1372f | 3720 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3721 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3722 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3723 | #endif |
da844b78 | 3724 | n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false); |
12b22386 AK |
3725 | page->freelist = get_freepointer(kmem_cache_node, n); |
3726 | page->inuse = 1; | |
3727 | page->frozen = 0; | |
3728 | kmem_cache_node->node[node] = n; | |
4053497d | 3729 | init_kmem_cache_node(n); |
51df1142 | 3730 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3731 | |
67b6c900 | 3732 | /* |
1e4dd946 SR |
3733 | * No locks need to be taken here as it has just been |
3734 | * initialized and there is no concurrent access. | |
67b6c900 | 3735 | */ |
1e4dd946 | 3736 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3737 | } |
3738 | ||
3739 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3740 | { | |
3741 | int node; | |
fa45dc25 | 3742 | struct kmem_cache_node *n; |
81819f0f | 3743 | |
fa45dc25 | 3744 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3745 | s->node[node] = NULL; |
ea37df54 | 3746 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3747 | } |
3748 | } | |
3749 | ||
52b4b950 DS |
3750 | void __kmem_cache_release(struct kmem_cache *s) |
3751 | { | |
210e7a43 | 3752 | cache_random_seq_destroy(s); |
52b4b950 DS |
3753 | free_percpu(s->cpu_slab); |
3754 | free_kmem_cache_nodes(s); | |
3755 | } | |
3756 | ||
55136592 | 3757 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3758 | { |
3759 | int node; | |
81819f0f | 3760 | |
7e1fa93d | 3761 | for_each_node_mask(node, slab_nodes) { |
81819f0f CL |
3762 | struct kmem_cache_node *n; |
3763 | ||
73367bd8 | 3764 | if (slab_state == DOWN) { |
55136592 | 3765 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3766 | continue; |
3767 | } | |
51df1142 | 3768 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3769 | GFP_KERNEL, node); |
81819f0f | 3770 | |
73367bd8 AD |
3771 | if (!n) { |
3772 | free_kmem_cache_nodes(s); | |
3773 | return 0; | |
81819f0f | 3774 | } |
73367bd8 | 3775 | |
4053497d | 3776 | init_kmem_cache_node(n); |
ea37df54 | 3777 | s->node[node] = n; |
81819f0f CL |
3778 | } |
3779 | return 1; | |
3780 | } | |
81819f0f | 3781 | |
c0bdb232 | 3782 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3783 | { |
3784 | if (min < MIN_PARTIAL) | |
3785 | min = MIN_PARTIAL; | |
3786 | else if (min > MAX_PARTIAL) | |
3787 | min = MAX_PARTIAL; | |
3788 | s->min_partial = min; | |
3789 | } | |
3790 | ||
e6d0e1dc WY |
3791 | static void set_cpu_partial(struct kmem_cache *s) |
3792 | { | |
3793 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3794 | /* | |
3795 | * cpu_partial determined the maximum number of objects kept in the | |
3796 | * per cpu partial lists of a processor. | |
3797 | * | |
3798 | * Per cpu partial lists mainly contain slabs that just have one | |
3799 | * object freed. If they are used for allocation then they can be | |
3800 | * filled up again with minimal effort. The slab will never hit the | |
3801 | * per node partial lists and therefore no locking will be required. | |
3802 | * | |
3803 | * This setting also determines | |
3804 | * | |
3805 | * A) The number of objects from per cpu partial slabs dumped to the | |
3806 | * per node list when we reach the limit. | |
3807 | * B) The number of objects in cpu partial slabs to extract from the | |
3808 | * per node list when we run out of per cpu objects. We only fetch | |
3809 | * 50% to keep some capacity around for frees. | |
3810 | */ | |
3811 | if (!kmem_cache_has_cpu_partial(s)) | |
bbd4e305 | 3812 | slub_set_cpu_partial(s, 0); |
e6d0e1dc | 3813 | else if (s->size >= PAGE_SIZE) |
bbd4e305 | 3814 | slub_set_cpu_partial(s, 2); |
e6d0e1dc | 3815 | else if (s->size >= 1024) |
bbd4e305 | 3816 | slub_set_cpu_partial(s, 6); |
e6d0e1dc | 3817 | else if (s->size >= 256) |
bbd4e305 | 3818 | slub_set_cpu_partial(s, 13); |
e6d0e1dc | 3819 | else |
bbd4e305 | 3820 | slub_set_cpu_partial(s, 30); |
e6d0e1dc WY |
3821 | #endif |
3822 | } | |
3823 | ||
81819f0f CL |
3824 | /* |
3825 | * calculate_sizes() determines the order and the distribution of data within | |
3826 | * a slab object. | |
3827 | */ | |
06b285dc | 3828 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3829 | { |
d50112ed | 3830 | slab_flags_t flags = s->flags; |
be4a7988 | 3831 | unsigned int size = s->object_size; |
19af27af | 3832 | unsigned int order; |
81819f0f | 3833 | |
d8b42bf5 CL |
3834 | /* |
3835 | * Round up object size to the next word boundary. We can only | |
3836 | * place the free pointer at word boundaries and this determines | |
3837 | * the possible location of the free pointer. | |
3838 | */ | |
3839 | size = ALIGN(size, sizeof(void *)); | |
3840 | ||
3841 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3842 | /* |
3843 | * Determine if we can poison the object itself. If the user of | |
3844 | * the slab may touch the object after free or before allocation | |
3845 | * then we should never poison the object itself. | |
3846 | */ | |
5f0d5a3a | 3847 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3848 | !s->ctor) |
81819f0f CL |
3849 | s->flags |= __OBJECT_POISON; |
3850 | else | |
3851 | s->flags &= ~__OBJECT_POISON; | |
3852 | ||
81819f0f CL |
3853 | |
3854 | /* | |
672bba3a | 3855 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3856 | * end of the object and the free pointer. If not then add an |
672bba3a | 3857 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3858 | */ |
3b0efdfa | 3859 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3860 | size += sizeof(void *); |
41ecc55b | 3861 | #endif |
81819f0f CL |
3862 | |
3863 | /* | |
672bba3a | 3864 | * With that we have determined the number of bytes in actual use |
e41a49fa | 3865 | * by the object and redzoning. |
81819f0f CL |
3866 | */ |
3867 | s->inuse = size; | |
3868 | ||
74c1d3e0 KC |
3869 | if ((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
3870 | ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) || | |
3871 | s->ctor) { | |
81819f0f CL |
3872 | /* |
3873 | * Relocate free pointer after the object if it is not | |
3874 | * permitted to overwrite the first word of the object on | |
3875 | * kmem_cache_free. | |
3876 | * | |
3877 | * This is the case if we do RCU, have a constructor or | |
74c1d3e0 KC |
3878 | * destructor, are poisoning the objects, or are |
3879 | * redzoning an object smaller than sizeof(void *). | |
cbfc35a4 WL |
3880 | * |
3881 | * The assumption that s->offset >= s->inuse means free | |
3882 | * pointer is outside of the object is used in the | |
3883 | * freeptr_outside_object() function. If that is no | |
3884 | * longer true, the function needs to be modified. | |
81819f0f CL |
3885 | */ |
3886 | s->offset = size; | |
3887 | size += sizeof(void *); | |
e41a49fa | 3888 | } else { |
3202fa62 KC |
3889 | /* |
3890 | * Store freelist pointer near middle of object to keep | |
3891 | * it away from the edges of the object to avoid small | |
3892 | * sized over/underflows from neighboring allocations. | |
3893 | */ | |
e41a49fa | 3894 | s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *)); |
81819f0f CL |
3895 | } |
3896 | ||
c12b3c62 | 3897 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3898 | if (flags & SLAB_STORE_USER) |
3899 | /* | |
3900 | * Need to store information about allocs and frees after | |
3901 | * the object. | |
3902 | */ | |
3903 | size += 2 * sizeof(struct track); | |
80a9201a | 3904 | #endif |
81819f0f | 3905 | |
80a9201a AP |
3906 | kasan_cache_create(s, &size, &s->flags); |
3907 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3908 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3909 | /* |
3910 | * Add some empty padding so that we can catch | |
3911 | * overwrites from earlier objects rather than let | |
3912 | * tracking information or the free pointer be | |
0211a9c8 | 3913 | * corrupted if a user writes before the start |
81819f0f CL |
3914 | * of the object. |
3915 | */ | |
3916 | size += sizeof(void *); | |
d86bd1be JK |
3917 | |
3918 | s->red_left_pad = sizeof(void *); | |
3919 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3920 | size += s->red_left_pad; | |
3921 | } | |
41ecc55b | 3922 | #endif |
672bba3a | 3923 | |
81819f0f CL |
3924 | /* |
3925 | * SLUB stores one object immediately after another beginning from | |
3926 | * offset 0. In order to align the objects we have to simply size | |
3927 | * each object to conform to the alignment. | |
3928 | */ | |
45906855 | 3929 | size = ALIGN(size, s->align); |
81819f0f | 3930 | s->size = size; |
4138fdfc | 3931 | s->reciprocal_size = reciprocal_value(size); |
06b285dc CL |
3932 | if (forced_order >= 0) |
3933 | order = forced_order; | |
3934 | else | |
9736d2a9 | 3935 | order = calculate_order(size); |
81819f0f | 3936 | |
19af27af | 3937 | if ((int)order < 0) |
81819f0f CL |
3938 | return 0; |
3939 | ||
b7a49f0d | 3940 | s->allocflags = 0; |
834f3d11 | 3941 | if (order) |
b7a49f0d CL |
3942 | s->allocflags |= __GFP_COMP; |
3943 | ||
3944 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3945 | s->allocflags |= GFP_DMA; |
b7a49f0d | 3946 | |
6d6ea1e9 NB |
3947 | if (s->flags & SLAB_CACHE_DMA32) |
3948 | s->allocflags |= GFP_DMA32; | |
3949 | ||
b7a49f0d CL |
3950 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
3951 | s->allocflags |= __GFP_RECLAIMABLE; | |
3952 | ||
81819f0f CL |
3953 | /* |
3954 | * Determine the number of objects per slab | |
3955 | */ | |
9736d2a9 MW |
3956 | s->oo = oo_make(order, size); |
3957 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3958 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3959 | s->max = s->oo; | |
81819f0f | 3960 | |
834f3d11 | 3961 | return !!oo_objects(s->oo); |
81819f0f CL |
3962 | } |
3963 | ||
d50112ed | 3964 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3965 | { |
37540008 | 3966 | s->flags = kmem_cache_flags(s->size, flags, s->name); |
2482ddec KC |
3967 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3968 | s->random = get_random_long(); | |
3969 | #endif | |
81819f0f | 3970 | |
06b285dc | 3971 | if (!calculate_sizes(s, -1)) |
81819f0f | 3972 | goto error; |
3de47213 DR |
3973 | if (disable_higher_order_debug) { |
3974 | /* | |
3975 | * Disable debugging flags that store metadata if the min slab | |
3976 | * order increased. | |
3977 | */ | |
3b0efdfa | 3978 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3979 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3980 | s->offset = 0; | |
3981 | if (!calculate_sizes(s, -1)) | |
3982 | goto error; | |
3983 | } | |
3984 | } | |
81819f0f | 3985 | |
2565409f HC |
3986 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3987 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3988 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3989 | /* Enable fast mode */ |
3990 | s->flags |= __CMPXCHG_DOUBLE; | |
3991 | #endif | |
3992 | ||
3b89d7d8 DR |
3993 | /* |
3994 | * The larger the object size is, the more pages we want on the partial | |
3995 | * list to avoid pounding the page allocator excessively. | |
3996 | */ | |
49e22585 CL |
3997 | set_min_partial(s, ilog2(s->size) / 2); |
3998 | ||
e6d0e1dc | 3999 | set_cpu_partial(s); |
49e22585 | 4000 | |
81819f0f | 4001 | #ifdef CONFIG_NUMA |
e2cb96b7 | 4002 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 4003 | #endif |
210e7a43 TG |
4004 | |
4005 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
4006 | if (slab_state >= UP) { | |
4007 | if (init_cache_random_seq(s)) | |
4008 | goto error; | |
4009 | } | |
4010 | ||
55136592 | 4011 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 4012 | goto error; |
81819f0f | 4013 | |
55136592 | 4014 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 4015 | return 0; |
ff12059e | 4016 | |
4c93c355 | 4017 | free_kmem_cache_nodes(s); |
81819f0f | 4018 | error: |
278b1bb1 | 4019 | return -EINVAL; |
81819f0f | 4020 | } |
81819f0f | 4021 | |
33b12c38 | 4022 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
55860d96 | 4023 | const char *text) |
33b12c38 CL |
4024 | { |
4025 | #ifdef CONFIG_SLUB_DEBUG | |
4026 | void *addr = page_address(page); | |
55860d96 | 4027 | unsigned long *map; |
33b12c38 | 4028 | void *p; |
aa456c7a | 4029 | |
945cf2b6 | 4030 | slab_err(s, page, text, s->name); |
33b12c38 | 4031 | slab_lock(page); |
33b12c38 | 4032 | |
90e9f6a6 | 4033 | map = get_map(s, page); |
33b12c38 CL |
4034 | for_each_object(p, s, addr, page->objects) { |
4035 | ||
4138fdfc | 4036 | if (!test_bit(__obj_to_index(s, addr, p), map)) { |
96b94abc | 4037 | pr_err("Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
4038 | print_tracking(s, p); |
4039 | } | |
4040 | } | |
55860d96 | 4041 | put_map(map); |
33b12c38 CL |
4042 | slab_unlock(page); |
4043 | #endif | |
4044 | } | |
4045 | ||
81819f0f | 4046 | /* |
599870b1 | 4047 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
4048 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
4049 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 4050 | */ |
599870b1 | 4051 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 4052 | { |
60398923 | 4053 | LIST_HEAD(discard); |
81819f0f CL |
4054 | struct page *page, *h; |
4055 | ||
52b4b950 DS |
4056 | BUG_ON(irqs_disabled()); |
4057 | spin_lock_irq(&n->list_lock); | |
916ac052 | 4058 | list_for_each_entry_safe(page, h, &n->partial, slab_list) { |
81819f0f | 4059 | if (!page->inuse) { |
52b4b950 | 4060 | remove_partial(n, page); |
916ac052 | 4061 | list_add(&page->slab_list, &discard); |
33b12c38 CL |
4062 | } else { |
4063 | list_slab_objects(s, page, | |
55860d96 | 4064 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 4065 | } |
33b12c38 | 4066 | } |
52b4b950 | 4067 | spin_unlock_irq(&n->list_lock); |
60398923 | 4068 | |
916ac052 | 4069 | list_for_each_entry_safe(page, h, &discard, slab_list) |
60398923 | 4070 | discard_slab(s, page); |
81819f0f CL |
4071 | } |
4072 | ||
f9e13c0a SB |
4073 | bool __kmem_cache_empty(struct kmem_cache *s) |
4074 | { | |
4075 | int node; | |
4076 | struct kmem_cache_node *n; | |
4077 | ||
4078 | for_each_kmem_cache_node(s, node, n) | |
4079 | if (n->nr_partial || slabs_node(s, node)) | |
4080 | return false; | |
4081 | return true; | |
4082 | } | |
4083 | ||
81819f0f | 4084 | /* |
672bba3a | 4085 | * Release all resources used by a slab cache. |
81819f0f | 4086 | */ |
52b4b950 | 4087 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
4088 | { |
4089 | int node; | |
fa45dc25 | 4090 | struct kmem_cache_node *n; |
81819f0f CL |
4091 | |
4092 | flush_all(s); | |
81819f0f | 4093 | /* Attempt to free all objects */ |
fa45dc25 | 4094 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
4095 | free_partial(s, n); |
4096 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
4097 | return 1; |
4098 | } | |
81819f0f CL |
4099 | return 0; |
4100 | } | |
4101 | ||
5bb1bb35 | 4102 | #ifdef CONFIG_PRINTK |
8e7f37f2 PM |
4103 | void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page) |
4104 | { | |
4105 | void *base; | |
4106 | int __maybe_unused i; | |
4107 | unsigned int objnr; | |
4108 | void *objp; | |
4109 | void *objp0; | |
4110 | struct kmem_cache *s = page->slab_cache; | |
4111 | struct track __maybe_unused *trackp; | |
4112 | ||
4113 | kpp->kp_ptr = object; | |
4114 | kpp->kp_page = page; | |
4115 | kpp->kp_slab_cache = s; | |
4116 | base = page_address(page); | |
4117 | objp0 = kasan_reset_tag(object); | |
4118 | #ifdef CONFIG_SLUB_DEBUG | |
4119 | objp = restore_red_left(s, objp0); | |
4120 | #else | |
4121 | objp = objp0; | |
4122 | #endif | |
4123 | objnr = obj_to_index(s, page, objp); | |
4124 | kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp); | |
4125 | objp = base + s->size * objnr; | |
4126 | kpp->kp_objp = objp; | |
4127 | if (WARN_ON_ONCE(objp < base || objp >= base + page->objects * s->size || (objp - base) % s->size) || | |
4128 | !(s->flags & SLAB_STORE_USER)) | |
4129 | return; | |
4130 | #ifdef CONFIG_SLUB_DEBUG | |
0cbc124b | 4131 | objp = fixup_red_left(s, objp); |
8e7f37f2 PM |
4132 | trackp = get_track(s, objp, TRACK_ALLOC); |
4133 | kpp->kp_ret = (void *)trackp->addr; | |
ae14c63a LT |
4134 | #ifdef CONFIG_STACKTRACE |
4135 | for (i = 0; i < KS_ADDRS_COUNT && i < TRACK_ADDRS_COUNT; i++) { | |
4136 | kpp->kp_stack[i] = (void *)trackp->addrs[i]; | |
4137 | if (!kpp->kp_stack[i]) | |
4138 | break; | |
4139 | } | |
78869146 | 4140 | |
ae14c63a LT |
4141 | trackp = get_track(s, objp, TRACK_FREE); |
4142 | for (i = 0; i < KS_ADDRS_COUNT && i < TRACK_ADDRS_COUNT; i++) { | |
4143 | kpp->kp_free_stack[i] = (void *)trackp->addrs[i]; | |
4144 | if (!kpp->kp_free_stack[i]) | |
4145 | break; | |
e548eaa1 | 4146 | } |
8e7f37f2 PM |
4147 | #endif |
4148 | #endif | |
4149 | } | |
5bb1bb35 | 4150 | #endif |
8e7f37f2 | 4151 | |
81819f0f CL |
4152 | /******************************************************************** |
4153 | * Kmalloc subsystem | |
4154 | *******************************************************************/ | |
4155 | ||
81819f0f CL |
4156 | static int __init setup_slub_min_order(char *str) |
4157 | { | |
19af27af | 4158 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
4159 | |
4160 | return 1; | |
4161 | } | |
4162 | ||
4163 | __setup("slub_min_order=", setup_slub_min_order); | |
4164 | ||
4165 | static int __init setup_slub_max_order(char *str) | |
4166 | { | |
19af27af AD |
4167 | get_option(&str, (int *)&slub_max_order); |
4168 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
4169 | |
4170 | return 1; | |
4171 | } | |
4172 | ||
4173 | __setup("slub_max_order=", setup_slub_max_order); | |
4174 | ||
4175 | static int __init setup_slub_min_objects(char *str) | |
4176 | { | |
19af27af | 4177 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
4178 | |
4179 | return 1; | |
4180 | } | |
4181 | ||
4182 | __setup("slub_min_objects=", setup_slub_min_objects); | |
4183 | ||
81819f0f CL |
4184 | void *__kmalloc(size_t size, gfp_t flags) |
4185 | { | |
aadb4bc4 | 4186 | struct kmem_cache *s; |
5b882be4 | 4187 | void *ret; |
81819f0f | 4188 | |
95a05b42 | 4189 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 4190 | return kmalloc_large(size, flags); |
aadb4bc4 | 4191 | |
2c59dd65 | 4192 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4193 | |
4194 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4195 | return s; |
4196 | ||
b89fb5ef | 4197 | ret = slab_alloc(s, flags, _RET_IP_, size); |
5b882be4 | 4198 | |
ca2b84cb | 4199 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 4200 | |
0116523c | 4201 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4202 | |
5b882be4 | 4203 | return ret; |
81819f0f CL |
4204 | } |
4205 | EXPORT_SYMBOL(__kmalloc); | |
4206 | ||
5d1f57e4 | 4207 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
4208 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
4209 | { | |
b1eeab67 | 4210 | struct page *page; |
e4f7c0b4 | 4211 | void *ptr = NULL; |
6a486c0a | 4212 | unsigned int order = get_order(size); |
f619cfe1 | 4213 | |
75f296d9 | 4214 | flags |= __GFP_COMP; |
6a486c0a VB |
4215 | page = alloc_pages_node(node, flags, order); |
4216 | if (page) { | |
e4f7c0b4 | 4217 | ptr = page_address(page); |
96403bfe MS |
4218 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
4219 | PAGE_SIZE << order); | |
6a486c0a | 4220 | } |
e4f7c0b4 | 4221 | |
0116523c | 4222 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
4223 | } |
4224 | ||
81819f0f CL |
4225 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
4226 | { | |
aadb4bc4 | 4227 | struct kmem_cache *s; |
5b882be4 | 4228 | void *ret; |
81819f0f | 4229 | |
95a05b42 | 4230 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
4231 | ret = kmalloc_large_node(size, flags, node); |
4232 | ||
ca2b84cb EGM |
4233 | trace_kmalloc_node(_RET_IP_, ret, |
4234 | size, PAGE_SIZE << get_order(size), | |
4235 | flags, node); | |
5b882be4 EGM |
4236 | |
4237 | return ret; | |
4238 | } | |
aadb4bc4 | 4239 | |
2c59dd65 | 4240 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4241 | |
4242 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4243 | return s; |
4244 | ||
b89fb5ef | 4245 | ret = slab_alloc_node(s, flags, node, _RET_IP_, size); |
5b882be4 | 4246 | |
ca2b84cb | 4247 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 4248 | |
0116523c | 4249 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4250 | |
5b882be4 | 4251 | return ret; |
81819f0f CL |
4252 | } |
4253 | EXPORT_SYMBOL(__kmalloc_node); | |
6dfd1b65 | 4254 | #endif /* CONFIG_NUMA */ |
81819f0f | 4255 | |
ed18adc1 KC |
4256 | #ifdef CONFIG_HARDENED_USERCOPY |
4257 | /* | |
afcc90f8 KC |
4258 | * Rejects incorrectly sized objects and objects that are to be copied |
4259 | * to/from userspace but do not fall entirely within the containing slab | |
4260 | * cache's usercopy region. | |
ed18adc1 KC |
4261 | * |
4262 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4263 | * to indicate an error. | |
4264 | */ | |
f4e6e289 KC |
4265 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
4266 | bool to_user) | |
ed18adc1 KC |
4267 | { |
4268 | struct kmem_cache *s; | |
44065b2e | 4269 | unsigned int offset; |
ed18adc1 | 4270 | size_t object_size; |
b89fb5ef | 4271 | bool is_kfence = is_kfence_address(ptr); |
ed18adc1 | 4272 | |
96fedce2 AK |
4273 | ptr = kasan_reset_tag(ptr); |
4274 | ||
ed18adc1 KC |
4275 | /* Find object and usable object size. */ |
4276 | s = page->slab_cache; | |
ed18adc1 KC |
4277 | |
4278 | /* Reject impossible pointers. */ | |
4279 | if (ptr < page_address(page)) | |
f4e6e289 KC |
4280 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
4281 | to_user, 0, n); | |
ed18adc1 KC |
4282 | |
4283 | /* Find offset within object. */ | |
b89fb5ef AP |
4284 | if (is_kfence) |
4285 | offset = ptr - kfence_object_start(ptr); | |
4286 | else | |
4287 | offset = (ptr - page_address(page)) % s->size; | |
ed18adc1 KC |
4288 | |
4289 | /* Adjust for redzone and reject if within the redzone. */ | |
b89fb5ef | 4290 | if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) { |
ed18adc1 | 4291 | if (offset < s->red_left_pad) |
f4e6e289 KC |
4292 | usercopy_abort("SLUB object in left red zone", |
4293 | s->name, to_user, offset, n); | |
ed18adc1 KC |
4294 | offset -= s->red_left_pad; |
4295 | } | |
4296 | ||
afcc90f8 KC |
4297 | /* Allow address range falling entirely within usercopy region. */ |
4298 | if (offset >= s->useroffset && | |
4299 | offset - s->useroffset <= s->usersize && | |
4300 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 4301 | return; |
ed18adc1 | 4302 | |
afcc90f8 KC |
4303 | /* |
4304 | * If the copy is still within the allocated object, produce | |
4305 | * a warning instead of rejecting the copy. This is intended | |
4306 | * to be a temporary method to find any missing usercopy | |
4307 | * whitelists. | |
4308 | */ | |
4309 | object_size = slab_ksize(s); | |
2d891fbc KC |
4310 | if (usercopy_fallback && |
4311 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
4312 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
4313 | return; | |
4314 | } | |
ed18adc1 | 4315 | |
f4e6e289 | 4316 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
4317 | } |
4318 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4319 | ||
10d1f8cb | 4320 | size_t __ksize(const void *object) |
81819f0f | 4321 | { |
272c1d21 | 4322 | struct page *page; |
81819f0f | 4323 | |
ef8b4520 | 4324 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
4325 | return 0; |
4326 | ||
294a80a8 | 4327 | page = virt_to_head_page(object); |
294a80a8 | 4328 | |
76994412 PE |
4329 | if (unlikely(!PageSlab(page))) { |
4330 | WARN_ON(!PageCompound(page)); | |
a50b854e | 4331 | return page_size(page); |
76994412 | 4332 | } |
81819f0f | 4333 | |
1b4f59e3 | 4334 | return slab_ksize(page->slab_cache); |
81819f0f | 4335 | } |
10d1f8cb | 4336 | EXPORT_SYMBOL(__ksize); |
81819f0f CL |
4337 | |
4338 | void kfree(const void *x) | |
4339 | { | |
81819f0f | 4340 | struct page *page; |
5bb983b0 | 4341 | void *object = (void *)x; |
81819f0f | 4342 | |
2121db74 PE |
4343 | trace_kfree(_RET_IP_, x); |
4344 | ||
2408c550 | 4345 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
4346 | return; |
4347 | ||
b49af68f | 4348 | page = virt_to_head_page(x); |
aadb4bc4 | 4349 | if (unlikely(!PageSlab(page))) { |
1ed7ce57 | 4350 | free_nonslab_page(page, object); |
aadb4bc4 CL |
4351 | return; |
4352 | } | |
81084651 | 4353 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
4354 | } |
4355 | EXPORT_SYMBOL(kfree); | |
4356 | ||
832f37f5 VD |
4357 | #define SHRINK_PROMOTE_MAX 32 |
4358 | ||
2086d26a | 4359 | /* |
832f37f5 VD |
4360 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
4361 | * up most to the head of the partial lists. New allocations will then | |
4362 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
4363 | * |
4364 | * The slabs with the least items are placed last. This results in them | |
4365 | * being allocated from last increasing the chance that the last objects | |
4366 | * are freed in them. | |
2086d26a | 4367 | */ |
c9fc5864 | 4368 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
4369 | { |
4370 | int node; | |
4371 | int i; | |
4372 | struct kmem_cache_node *n; | |
4373 | struct page *page; | |
4374 | struct page *t; | |
832f37f5 VD |
4375 | struct list_head discard; |
4376 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 4377 | unsigned long flags; |
ce3712d7 | 4378 | int ret = 0; |
2086d26a | 4379 | |
2086d26a | 4380 | flush_all(s); |
fa45dc25 | 4381 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
4382 | INIT_LIST_HEAD(&discard); |
4383 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
4384 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
4385 | |
4386 | spin_lock_irqsave(&n->list_lock, flags); | |
4387 | ||
4388 | /* | |
832f37f5 | 4389 | * Build lists of slabs to discard or promote. |
2086d26a | 4390 | * |
672bba3a CL |
4391 | * Note that concurrent frees may occur while we hold the |
4392 | * list_lock. page->inuse here is the upper limit. | |
2086d26a | 4393 | */ |
916ac052 | 4394 | list_for_each_entry_safe(page, t, &n->partial, slab_list) { |
832f37f5 VD |
4395 | int free = page->objects - page->inuse; |
4396 | ||
4397 | /* Do not reread page->inuse */ | |
4398 | barrier(); | |
4399 | ||
4400 | /* We do not keep full slabs on the list */ | |
4401 | BUG_ON(free <= 0); | |
4402 | ||
4403 | if (free == page->objects) { | |
916ac052 | 4404 | list_move(&page->slab_list, &discard); |
69cb8e6b | 4405 | n->nr_partial--; |
832f37f5 | 4406 | } else if (free <= SHRINK_PROMOTE_MAX) |
916ac052 | 4407 | list_move(&page->slab_list, promote + free - 1); |
2086d26a CL |
4408 | } |
4409 | ||
2086d26a | 4410 | /* |
832f37f5 VD |
4411 | * Promote the slabs filled up most to the head of the |
4412 | * partial list. | |
2086d26a | 4413 | */ |
832f37f5 VD |
4414 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4415 | list_splice(promote + i, &n->partial); | |
2086d26a | 4416 | |
2086d26a | 4417 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4418 | |
4419 | /* Release empty slabs */ | |
916ac052 | 4420 | list_for_each_entry_safe(page, t, &discard, slab_list) |
69cb8e6b | 4421 | discard_slab(s, page); |
ce3712d7 VD |
4422 | |
4423 | if (slabs_node(s, node)) | |
4424 | ret = 1; | |
2086d26a CL |
4425 | } |
4426 | ||
ce3712d7 | 4427 | return ret; |
2086d26a | 4428 | } |
2086d26a | 4429 | |
b9049e23 YG |
4430 | static int slab_mem_going_offline_callback(void *arg) |
4431 | { | |
4432 | struct kmem_cache *s; | |
4433 | ||
18004c5d | 4434 | mutex_lock(&slab_mutex); |
b9049e23 | 4435 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4436 | __kmem_cache_shrink(s); |
18004c5d | 4437 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4438 | |
4439 | return 0; | |
4440 | } | |
4441 | ||
4442 | static void slab_mem_offline_callback(void *arg) | |
4443 | { | |
b9049e23 YG |
4444 | struct memory_notify *marg = arg; |
4445 | int offline_node; | |
4446 | ||
b9d5ab25 | 4447 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4448 | |
4449 | /* | |
4450 | * If the node still has available memory. we need kmem_cache_node | |
4451 | * for it yet. | |
4452 | */ | |
4453 | if (offline_node < 0) | |
4454 | return; | |
4455 | ||
18004c5d | 4456 | mutex_lock(&slab_mutex); |
7e1fa93d | 4457 | node_clear(offline_node, slab_nodes); |
666716fd VB |
4458 | /* |
4459 | * We no longer free kmem_cache_node structures here, as it would be | |
4460 | * racy with all get_node() users, and infeasible to protect them with | |
4461 | * slab_mutex. | |
4462 | */ | |
18004c5d | 4463 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4464 | } |
4465 | ||
4466 | static int slab_mem_going_online_callback(void *arg) | |
4467 | { | |
4468 | struct kmem_cache_node *n; | |
4469 | struct kmem_cache *s; | |
4470 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4471 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4472 | int ret = 0; |
4473 | ||
4474 | /* | |
4475 | * If the node's memory is already available, then kmem_cache_node is | |
4476 | * already created. Nothing to do. | |
4477 | */ | |
4478 | if (nid < 0) | |
4479 | return 0; | |
4480 | ||
4481 | /* | |
0121c619 | 4482 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4483 | * allocate a kmem_cache_node structure in order to bring the node |
4484 | * online. | |
4485 | */ | |
18004c5d | 4486 | mutex_lock(&slab_mutex); |
b9049e23 | 4487 | list_for_each_entry(s, &slab_caches, list) { |
666716fd VB |
4488 | /* |
4489 | * The structure may already exist if the node was previously | |
4490 | * onlined and offlined. | |
4491 | */ | |
4492 | if (get_node(s, nid)) | |
4493 | continue; | |
b9049e23 YG |
4494 | /* |
4495 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4496 | * since memory is not yet available from the node that | |
4497 | * is brought up. | |
4498 | */ | |
8de66a0c | 4499 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4500 | if (!n) { |
4501 | ret = -ENOMEM; | |
4502 | goto out; | |
4503 | } | |
4053497d | 4504 | init_kmem_cache_node(n); |
b9049e23 YG |
4505 | s->node[nid] = n; |
4506 | } | |
7e1fa93d VB |
4507 | /* |
4508 | * Any cache created after this point will also have kmem_cache_node | |
4509 | * initialized for the new node. | |
4510 | */ | |
4511 | node_set(nid, slab_nodes); | |
b9049e23 | 4512 | out: |
18004c5d | 4513 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4514 | return ret; |
4515 | } | |
4516 | ||
4517 | static int slab_memory_callback(struct notifier_block *self, | |
4518 | unsigned long action, void *arg) | |
4519 | { | |
4520 | int ret = 0; | |
4521 | ||
4522 | switch (action) { | |
4523 | case MEM_GOING_ONLINE: | |
4524 | ret = slab_mem_going_online_callback(arg); | |
4525 | break; | |
4526 | case MEM_GOING_OFFLINE: | |
4527 | ret = slab_mem_going_offline_callback(arg); | |
4528 | break; | |
4529 | case MEM_OFFLINE: | |
4530 | case MEM_CANCEL_ONLINE: | |
4531 | slab_mem_offline_callback(arg); | |
4532 | break; | |
4533 | case MEM_ONLINE: | |
4534 | case MEM_CANCEL_OFFLINE: | |
4535 | break; | |
4536 | } | |
dc19f9db KH |
4537 | if (ret) |
4538 | ret = notifier_from_errno(ret); | |
4539 | else | |
4540 | ret = NOTIFY_OK; | |
b9049e23 YG |
4541 | return ret; |
4542 | } | |
4543 | ||
3ac38faa AM |
4544 | static struct notifier_block slab_memory_callback_nb = { |
4545 | .notifier_call = slab_memory_callback, | |
4546 | .priority = SLAB_CALLBACK_PRI, | |
4547 | }; | |
b9049e23 | 4548 | |
81819f0f CL |
4549 | /******************************************************************** |
4550 | * Basic setup of slabs | |
4551 | *******************************************************************/ | |
4552 | ||
51df1142 CL |
4553 | /* |
4554 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4555 | * the page allocator. Allocate them properly then fix up the pointers |
4556 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4557 | */ |
4558 | ||
dffb4d60 | 4559 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4560 | { |
4561 | int node; | |
dffb4d60 | 4562 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4563 | struct kmem_cache_node *n; |
51df1142 | 4564 | |
dffb4d60 | 4565 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4566 | |
7d557b3c GC |
4567 | /* |
4568 | * This runs very early, and only the boot processor is supposed to be | |
4569 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4570 | * IPIs around. | |
4571 | */ | |
4572 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4573 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4574 | struct page *p; |
4575 | ||
916ac052 | 4576 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 4577 | p->slab_cache = s; |
51df1142 | 4578 | |
607bf324 | 4579 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 4580 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 4581 | p->slab_cache = s; |
51df1142 | 4582 | #endif |
51df1142 | 4583 | } |
dffb4d60 CL |
4584 | list_add(&s->list, &slab_caches); |
4585 | return s; | |
51df1142 CL |
4586 | } |
4587 | ||
81819f0f CL |
4588 | void __init kmem_cache_init(void) |
4589 | { | |
dffb4d60 CL |
4590 | static __initdata struct kmem_cache boot_kmem_cache, |
4591 | boot_kmem_cache_node; | |
7e1fa93d | 4592 | int node; |
51df1142 | 4593 | |
fc8d8620 SG |
4594 | if (debug_guardpage_minorder()) |
4595 | slub_max_order = 0; | |
4596 | ||
79270291 SB |
4597 | /* Print slub debugging pointers without hashing */ |
4598 | if (__slub_debug_enabled()) | |
4599 | no_hash_pointers_enable(NULL); | |
4600 | ||
dffb4d60 CL |
4601 | kmem_cache_node = &boot_kmem_cache_node; |
4602 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4603 | |
7e1fa93d VB |
4604 | /* |
4605 | * Initialize the nodemask for which we will allocate per node | |
4606 | * structures. Here we don't need taking slab_mutex yet. | |
4607 | */ | |
4608 | for_each_node_state(node, N_NORMAL_MEMORY) | |
4609 | node_set(node, slab_nodes); | |
4610 | ||
dffb4d60 | 4611 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4612 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4613 | |
3ac38faa | 4614 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4615 | |
4616 | /* Able to allocate the per node structures */ | |
4617 | slab_state = PARTIAL; | |
4618 | ||
dffb4d60 CL |
4619 | create_boot_cache(kmem_cache, "kmem_cache", |
4620 | offsetof(struct kmem_cache, node) + | |
4621 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4622 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4623 | |
dffb4d60 | 4624 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4625 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4626 | |
4627 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4628 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4629 | create_kmalloc_caches(0); |
81819f0f | 4630 | |
210e7a43 TG |
4631 | /* Setup random freelists for each cache */ |
4632 | init_freelist_randomization(); | |
4633 | ||
a96a87bf SAS |
4634 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4635 | slub_cpu_dead); | |
81819f0f | 4636 | |
b9726c26 | 4637 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4638 | cache_line_size(), |
81819f0f CL |
4639 | slub_min_order, slub_max_order, slub_min_objects, |
4640 | nr_cpu_ids, nr_node_ids); | |
4641 | } | |
4642 | ||
7e85ee0c PE |
4643 | void __init kmem_cache_init_late(void) |
4644 | { | |
7e85ee0c PE |
4645 | } |
4646 | ||
2633d7a0 | 4647 | struct kmem_cache * |
f4957d5b | 4648 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4649 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4650 | { |
10befea9 | 4651 | struct kmem_cache *s; |
81819f0f | 4652 | |
a44cb944 | 4653 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4654 | if (s) { |
4655 | s->refcount++; | |
84d0ddd6 | 4656 | |
81819f0f CL |
4657 | /* |
4658 | * Adjust the object sizes so that we clear | |
4659 | * the complete object on kzalloc. | |
4660 | */ | |
1b473f29 | 4661 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4662 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4663 | |
7b8f3b66 | 4664 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4665 | s->refcount--; |
cbb79694 | 4666 | s = NULL; |
7b8f3b66 | 4667 | } |
a0e1d1be | 4668 | } |
6446faa2 | 4669 | |
cbb79694 CL |
4670 | return s; |
4671 | } | |
84c1cf62 | 4672 | |
d50112ed | 4673 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4674 | { |
aac3a166 PE |
4675 | int err; |
4676 | ||
4677 | err = kmem_cache_open(s, flags); | |
4678 | if (err) | |
4679 | return err; | |
20cea968 | 4680 | |
45530c44 CL |
4681 | /* Mutex is not taken during early boot */ |
4682 | if (slab_state <= UP) | |
4683 | return 0; | |
4684 | ||
aac3a166 | 4685 | err = sysfs_slab_add(s); |
aac3a166 | 4686 | if (err) |
52b4b950 | 4687 | __kmem_cache_release(s); |
20cea968 | 4688 | |
64dd6849 FM |
4689 | if (s->flags & SLAB_STORE_USER) |
4690 | debugfs_slab_add(s); | |
4691 | ||
aac3a166 | 4692 | return err; |
81819f0f | 4693 | } |
81819f0f | 4694 | |
ce71e27c | 4695 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4696 | { |
aadb4bc4 | 4697 | struct kmem_cache *s; |
94b528d0 | 4698 | void *ret; |
aadb4bc4 | 4699 | |
95a05b42 | 4700 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4701 | return kmalloc_large(size, gfpflags); |
4702 | ||
2c59dd65 | 4703 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4704 | |
2408c550 | 4705 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4706 | return s; |
81819f0f | 4707 | |
b89fb5ef | 4708 | ret = slab_alloc(s, gfpflags, caller, size); |
94b528d0 | 4709 | |
25985edc | 4710 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4711 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4712 | |
4713 | return ret; | |
81819f0f | 4714 | } |
fd7cb575 | 4715 | EXPORT_SYMBOL(__kmalloc_track_caller); |
81819f0f | 4716 | |
5d1f57e4 | 4717 | #ifdef CONFIG_NUMA |
81819f0f | 4718 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4719 | int node, unsigned long caller) |
81819f0f | 4720 | { |
aadb4bc4 | 4721 | struct kmem_cache *s; |
94b528d0 | 4722 | void *ret; |
aadb4bc4 | 4723 | |
95a05b42 | 4724 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4725 | ret = kmalloc_large_node(size, gfpflags, node); |
4726 | ||
4727 | trace_kmalloc_node(caller, ret, | |
4728 | size, PAGE_SIZE << get_order(size), | |
4729 | gfpflags, node); | |
4730 | ||
4731 | return ret; | |
4732 | } | |
eada35ef | 4733 | |
2c59dd65 | 4734 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4735 | |
2408c550 | 4736 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4737 | return s; |
81819f0f | 4738 | |
b89fb5ef | 4739 | ret = slab_alloc_node(s, gfpflags, node, caller, size); |
94b528d0 | 4740 | |
25985edc | 4741 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4742 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4743 | |
4744 | return ret; | |
81819f0f | 4745 | } |
fd7cb575 | 4746 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
5d1f57e4 | 4747 | #endif |
81819f0f | 4748 | |
ab4d5ed5 | 4749 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4750 | static int count_inuse(struct page *page) |
4751 | { | |
4752 | return page->inuse; | |
4753 | } | |
4754 | ||
4755 | static int count_total(struct page *page) | |
4756 | { | |
4757 | return page->objects; | |
4758 | } | |
ab4d5ed5 | 4759 | #endif |
205ab99d | 4760 | |
ab4d5ed5 | 4761 | #ifdef CONFIG_SLUB_DEBUG |
0a19e7dd VB |
4762 | static void validate_slab(struct kmem_cache *s, struct page *page, |
4763 | unsigned long *obj_map) | |
53e15af0 CL |
4764 | { |
4765 | void *p; | |
a973e9dd | 4766 | void *addr = page_address(page); |
90e9f6a6 YZ |
4767 | |
4768 | slab_lock(page); | |
53e15af0 | 4769 | |
dd98afd4 | 4770 | if (!check_slab(s, page) || !on_freelist(s, page, NULL)) |
90e9f6a6 | 4771 | goto unlock; |
53e15af0 CL |
4772 | |
4773 | /* Now we know that a valid freelist exists */ | |
0a19e7dd | 4774 | __fill_map(obj_map, s, page); |
5f80b13a | 4775 | for_each_object(p, s, addr, page->objects) { |
0a19e7dd | 4776 | u8 val = test_bit(__obj_to_index(s, addr, p), obj_map) ? |
dd98afd4 | 4777 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; |
53e15af0 | 4778 | |
dd98afd4 YZ |
4779 | if (!check_object(s, page, p, val)) |
4780 | break; | |
4781 | } | |
90e9f6a6 | 4782 | unlock: |
881db7fb | 4783 | slab_unlock(page); |
53e15af0 CL |
4784 | } |
4785 | ||
434e245d | 4786 | static int validate_slab_node(struct kmem_cache *s, |
0a19e7dd | 4787 | struct kmem_cache_node *n, unsigned long *obj_map) |
53e15af0 CL |
4788 | { |
4789 | unsigned long count = 0; | |
4790 | struct page *page; | |
4791 | unsigned long flags; | |
4792 | ||
4793 | spin_lock_irqsave(&n->list_lock, flags); | |
4794 | ||
916ac052 | 4795 | list_for_each_entry(page, &n->partial, slab_list) { |
0a19e7dd | 4796 | validate_slab(s, page, obj_map); |
53e15af0 CL |
4797 | count++; |
4798 | } | |
1f9f78b1 | 4799 | if (count != n->nr_partial) { |
f9f58285 FF |
4800 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4801 | s->name, count, n->nr_partial); | |
1f9f78b1 OG |
4802 | slab_add_kunit_errors(); |
4803 | } | |
53e15af0 CL |
4804 | |
4805 | if (!(s->flags & SLAB_STORE_USER)) | |
4806 | goto out; | |
4807 | ||
916ac052 | 4808 | list_for_each_entry(page, &n->full, slab_list) { |
0a19e7dd | 4809 | validate_slab(s, page, obj_map); |
53e15af0 CL |
4810 | count++; |
4811 | } | |
1f9f78b1 | 4812 | if (count != atomic_long_read(&n->nr_slabs)) { |
f9f58285 FF |
4813 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4814 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
1f9f78b1 OG |
4815 | slab_add_kunit_errors(); |
4816 | } | |
53e15af0 CL |
4817 | |
4818 | out: | |
4819 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4820 | return count; | |
4821 | } | |
4822 | ||
1f9f78b1 | 4823 | long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4824 | { |
4825 | int node; | |
4826 | unsigned long count = 0; | |
fa45dc25 | 4827 | struct kmem_cache_node *n; |
0a19e7dd VB |
4828 | unsigned long *obj_map; |
4829 | ||
4830 | obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL); | |
4831 | if (!obj_map) | |
4832 | return -ENOMEM; | |
53e15af0 CL |
4833 | |
4834 | flush_all(s); | |
fa45dc25 | 4835 | for_each_kmem_cache_node(s, node, n) |
0a19e7dd VB |
4836 | count += validate_slab_node(s, n, obj_map); |
4837 | ||
4838 | bitmap_free(obj_map); | |
90e9f6a6 | 4839 | |
53e15af0 CL |
4840 | return count; |
4841 | } | |
1f9f78b1 OG |
4842 | EXPORT_SYMBOL(validate_slab_cache); |
4843 | ||
64dd6849 | 4844 | #ifdef CONFIG_DEBUG_FS |
88a420e4 | 4845 | /* |
672bba3a | 4846 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4847 | * and freed. |
4848 | */ | |
4849 | ||
4850 | struct location { | |
4851 | unsigned long count; | |
ce71e27c | 4852 | unsigned long addr; |
45edfa58 CL |
4853 | long long sum_time; |
4854 | long min_time; | |
4855 | long max_time; | |
4856 | long min_pid; | |
4857 | long max_pid; | |
174596a0 | 4858 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4859 | nodemask_t nodes; |
88a420e4 CL |
4860 | }; |
4861 | ||
4862 | struct loc_track { | |
4863 | unsigned long max; | |
4864 | unsigned long count; | |
4865 | struct location *loc; | |
4866 | }; | |
4867 | ||
64dd6849 FM |
4868 | static struct dentry *slab_debugfs_root; |
4869 | ||
88a420e4 CL |
4870 | static void free_loc_track(struct loc_track *t) |
4871 | { | |
4872 | if (t->max) | |
4873 | free_pages((unsigned long)t->loc, | |
4874 | get_order(sizeof(struct location) * t->max)); | |
4875 | } | |
4876 | ||
68dff6a9 | 4877 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4878 | { |
4879 | struct location *l; | |
4880 | int order; | |
4881 | ||
88a420e4 CL |
4882 | order = get_order(sizeof(struct location) * max); |
4883 | ||
68dff6a9 | 4884 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4885 | if (!l) |
4886 | return 0; | |
4887 | ||
4888 | if (t->count) { | |
4889 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4890 | free_loc_track(t); | |
4891 | } | |
4892 | t->max = max; | |
4893 | t->loc = l; | |
4894 | return 1; | |
4895 | } | |
4896 | ||
4897 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4898 | const struct track *track) |
88a420e4 CL |
4899 | { |
4900 | long start, end, pos; | |
4901 | struct location *l; | |
ce71e27c | 4902 | unsigned long caddr; |
45edfa58 | 4903 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4904 | |
4905 | start = -1; | |
4906 | end = t->count; | |
4907 | ||
4908 | for ( ; ; ) { | |
4909 | pos = start + (end - start + 1) / 2; | |
4910 | ||
4911 | /* | |
4912 | * There is nothing at "end". If we end up there | |
4913 | * we need to add something to before end. | |
4914 | */ | |
4915 | if (pos == end) | |
4916 | break; | |
4917 | ||
4918 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4919 | if (track->addr == caddr) { |
4920 | ||
4921 | l = &t->loc[pos]; | |
4922 | l->count++; | |
4923 | if (track->when) { | |
4924 | l->sum_time += age; | |
4925 | if (age < l->min_time) | |
4926 | l->min_time = age; | |
4927 | if (age > l->max_time) | |
4928 | l->max_time = age; | |
4929 | ||
4930 | if (track->pid < l->min_pid) | |
4931 | l->min_pid = track->pid; | |
4932 | if (track->pid > l->max_pid) | |
4933 | l->max_pid = track->pid; | |
4934 | ||
174596a0 RR |
4935 | cpumask_set_cpu(track->cpu, |
4936 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4937 | } |
4938 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4939 | return 1; |
4940 | } | |
4941 | ||
45edfa58 | 4942 | if (track->addr < caddr) |
88a420e4 CL |
4943 | end = pos; |
4944 | else | |
4945 | start = pos; | |
4946 | } | |
4947 | ||
4948 | /* | |
672bba3a | 4949 | * Not found. Insert new tracking element. |
88a420e4 | 4950 | */ |
68dff6a9 | 4951 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4952 | return 0; |
4953 | ||
4954 | l = t->loc + pos; | |
4955 | if (pos < t->count) | |
4956 | memmove(l + 1, l, | |
4957 | (t->count - pos) * sizeof(struct location)); | |
4958 | t->count++; | |
4959 | l->count = 1; | |
45edfa58 CL |
4960 | l->addr = track->addr; |
4961 | l->sum_time = age; | |
4962 | l->min_time = age; | |
4963 | l->max_time = age; | |
4964 | l->min_pid = track->pid; | |
4965 | l->max_pid = track->pid; | |
174596a0 RR |
4966 | cpumask_clear(to_cpumask(l->cpus)); |
4967 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4968 | nodes_clear(l->nodes); |
4969 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4970 | return 1; |
4971 | } | |
4972 | ||
4973 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
b3fd64e1 VB |
4974 | struct page *page, enum track_item alloc, |
4975 | unsigned long *obj_map) | |
88a420e4 | 4976 | { |
a973e9dd | 4977 | void *addr = page_address(page); |
88a420e4 CL |
4978 | void *p; |
4979 | ||
b3fd64e1 VB |
4980 | __fill_map(obj_map, s, page); |
4981 | ||
224a88be | 4982 | for_each_object(p, s, addr, page->objects) |
b3fd64e1 | 4983 | if (!test_bit(__obj_to_index(s, addr, p), obj_map)) |
45edfa58 | 4984 | add_location(t, s, get_track(s, p, alloc)); |
88a420e4 | 4985 | } |
64dd6849 | 4986 | #endif /* CONFIG_DEBUG_FS */ |
6dfd1b65 | 4987 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 4988 | |
ab4d5ed5 | 4989 | #ifdef CONFIG_SYSFS |
81819f0f | 4990 | enum slab_stat_type { |
205ab99d CL |
4991 | SL_ALL, /* All slabs */ |
4992 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4993 | SL_CPU, /* Only slabs used for cpu caches */ | |
4994 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4995 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4996 | }; |
4997 | ||
205ab99d | 4998 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4999 | #define SO_PARTIAL (1 << SL_PARTIAL) |
5000 | #define SO_CPU (1 << SL_CPU) | |
5001 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 5002 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 5003 | |
62e5c4b4 | 5004 | static ssize_t show_slab_objects(struct kmem_cache *s, |
bf16d19a | 5005 | char *buf, unsigned long flags) |
81819f0f CL |
5006 | { |
5007 | unsigned long total = 0; | |
81819f0f CL |
5008 | int node; |
5009 | int x; | |
5010 | unsigned long *nodes; | |
bf16d19a | 5011 | int len = 0; |
81819f0f | 5012 | |
6396bb22 | 5013 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
5014 | if (!nodes) |
5015 | return -ENOMEM; | |
81819f0f | 5016 | |
205ab99d CL |
5017 | if (flags & SO_CPU) { |
5018 | int cpu; | |
81819f0f | 5019 | |
205ab99d | 5020 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
5021 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
5022 | cpu); | |
ec3ab083 | 5023 | int node; |
49e22585 | 5024 | struct page *page; |
dfb4f096 | 5025 | |
4db0c3c2 | 5026 | page = READ_ONCE(c->page); |
ec3ab083 CL |
5027 | if (!page) |
5028 | continue; | |
205ab99d | 5029 | |
ec3ab083 CL |
5030 | node = page_to_nid(page); |
5031 | if (flags & SO_TOTAL) | |
5032 | x = page->objects; | |
5033 | else if (flags & SO_OBJECTS) | |
5034 | x = page->inuse; | |
5035 | else | |
5036 | x = 1; | |
49e22585 | 5037 | |
ec3ab083 CL |
5038 | total += x; |
5039 | nodes[node] += x; | |
5040 | ||
a93cf07b | 5041 | page = slub_percpu_partial_read_once(c); |
49e22585 | 5042 | if (page) { |
8afb1474 LZ |
5043 | node = page_to_nid(page); |
5044 | if (flags & SO_TOTAL) | |
5045 | WARN_ON_ONCE(1); | |
5046 | else if (flags & SO_OBJECTS) | |
5047 | WARN_ON_ONCE(1); | |
5048 | else | |
5049 | x = page->pages; | |
bc6697d8 ED |
5050 | total += x; |
5051 | nodes[node] += x; | |
49e22585 | 5052 | } |
81819f0f CL |
5053 | } |
5054 | } | |
5055 | ||
e4f8e513 QC |
5056 | /* |
5057 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
5058 | * already held which will conflict with an existing lock order: | |
5059 | * | |
5060 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
5061 | * | |
5062 | * We don't really need mem_hotplug_lock (to hold off | |
5063 | * slab_mem_going_offline_callback) here because slab's memory hot | |
5064 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
5065 | */ | |
5066 | ||
ab4d5ed5 | 5067 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 5068 | if (flags & SO_ALL) { |
fa45dc25 CL |
5069 | struct kmem_cache_node *n; |
5070 | ||
5071 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 5072 | |
d0e0ac97 CG |
5073 | if (flags & SO_TOTAL) |
5074 | x = atomic_long_read(&n->total_objects); | |
5075 | else if (flags & SO_OBJECTS) | |
5076 | x = atomic_long_read(&n->total_objects) - | |
5077 | count_partial(n, count_free); | |
81819f0f | 5078 | else |
205ab99d | 5079 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
5080 | total += x; |
5081 | nodes[node] += x; | |
5082 | } | |
5083 | ||
ab4d5ed5 CL |
5084 | } else |
5085 | #endif | |
5086 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 5087 | struct kmem_cache_node *n; |
81819f0f | 5088 | |
fa45dc25 | 5089 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
5090 | if (flags & SO_TOTAL) |
5091 | x = count_partial(n, count_total); | |
5092 | else if (flags & SO_OBJECTS) | |
5093 | x = count_partial(n, count_inuse); | |
81819f0f | 5094 | else |
205ab99d | 5095 | x = n->nr_partial; |
81819f0f CL |
5096 | total += x; |
5097 | nodes[node] += x; | |
5098 | } | |
5099 | } | |
bf16d19a JP |
5100 | |
5101 | len += sysfs_emit_at(buf, len, "%lu", total); | |
81819f0f | 5102 | #ifdef CONFIG_NUMA |
bf16d19a | 5103 | for (node = 0; node < nr_node_ids; node++) { |
81819f0f | 5104 | if (nodes[node]) |
bf16d19a JP |
5105 | len += sysfs_emit_at(buf, len, " N%d=%lu", |
5106 | node, nodes[node]); | |
5107 | } | |
81819f0f | 5108 | #endif |
bf16d19a | 5109 | len += sysfs_emit_at(buf, len, "\n"); |
81819f0f | 5110 | kfree(nodes); |
bf16d19a JP |
5111 | |
5112 | return len; | |
81819f0f CL |
5113 | } |
5114 | ||
81819f0f | 5115 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) |
497888cf | 5116 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
5117 | |
5118 | struct slab_attribute { | |
5119 | struct attribute attr; | |
5120 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
5121 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
5122 | }; | |
5123 | ||
5124 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
5125 | static struct slab_attribute _name##_attr = \ |
5126 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
5127 | |
5128 | #define SLAB_ATTR(_name) \ | |
5129 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 5130 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 5131 | |
81819f0f CL |
5132 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
5133 | { | |
bf16d19a | 5134 | return sysfs_emit(buf, "%u\n", s->size); |
81819f0f CL |
5135 | } |
5136 | SLAB_ATTR_RO(slab_size); | |
5137 | ||
5138 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
5139 | { | |
bf16d19a | 5140 | return sysfs_emit(buf, "%u\n", s->align); |
81819f0f CL |
5141 | } |
5142 | SLAB_ATTR_RO(align); | |
5143 | ||
5144 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
5145 | { | |
bf16d19a | 5146 | return sysfs_emit(buf, "%u\n", s->object_size); |
81819f0f CL |
5147 | } |
5148 | SLAB_ATTR_RO(object_size); | |
5149 | ||
5150 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
5151 | { | |
bf16d19a | 5152 | return sysfs_emit(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
5153 | } |
5154 | SLAB_ATTR_RO(objs_per_slab); | |
5155 | ||
5156 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
5157 | { | |
bf16d19a | 5158 | return sysfs_emit(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 5159 | } |
32a6f409 | 5160 | SLAB_ATTR_RO(order); |
81819f0f | 5161 | |
73d342b1 DR |
5162 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
5163 | { | |
bf16d19a | 5164 | return sysfs_emit(buf, "%lu\n", s->min_partial); |
73d342b1 DR |
5165 | } |
5166 | ||
5167 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
5168 | size_t length) | |
5169 | { | |
5170 | unsigned long min; | |
5171 | int err; | |
5172 | ||
3dbb95f7 | 5173 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
5174 | if (err) |
5175 | return err; | |
5176 | ||
c0bdb232 | 5177 | set_min_partial(s, min); |
73d342b1 DR |
5178 | return length; |
5179 | } | |
5180 | SLAB_ATTR(min_partial); | |
5181 | ||
49e22585 CL |
5182 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
5183 | { | |
bf16d19a | 5184 | return sysfs_emit(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
5185 | } |
5186 | ||
5187 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
5188 | size_t length) | |
5189 | { | |
e5d9998f | 5190 | unsigned int objects; |
49e22585 CL |
5191 | int err; |
5192 | ||
e5d9998f | 5193 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5194 | if (err) |
5195 | return err; | |
345c905d | 5196 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5197 | return -EINVAL; |
49e22585 | 5198 | |
e6d0e1dc | 5199 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5200 | flush_all(s); |
5201 | return length; | |
5202 | } | |
5203 | SLAB_ATTR(cpu_partial); | |
5204 | ||
81819f0f CL |
5205 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5206 | { | |
62c70bce JP |
5207 | if (!s->ctor) |
5208 | return 0; | |
bf16d19a | 5209 | return sysfs_emit(buf, "%pS\n", s->ctor); |
81819f0f CL |
5210 | } |
5211 | SLAB_ATTR_RO(ctor); | |
5212 | ||
81819f0f CL |
5213 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5214 | { | |
bf16d19a | 5215 | return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5216 | } |
5217 | SLAB_ATTR_RO(aliases); | |
5218 | ||
81819f0f CL |
5219 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5220 | { | |
d9acf4b7 | 5221 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5222 | } |
5223 | SLAB_ATTR_RO(partial); | |
5224 | ||
5225 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5226 | { | |
d9acf4b7 | 5227 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5228 | } |
5229 | SLAB_ATTR_RO(cpu_slabs); | |
5230 | ||
5231 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5232 | { | |
205ab99d | 5233 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5234 | } |
5235 | SLAB_ATTR_RO(objects); | |
5236 | ||
205ab99d CL |
5237 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5238 | { | |
5239 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5240 | } | |
5241 | SLAB_ATTR_RO(objects_partial); | |
5242 | ||
49e22585 CL |
5243 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5244 | { | |
5245 | int objects = 0; | |
5246 | int pages = 0; | |
5247 | int cpu; | |
bf16d19a | 5248 | int len = 0; |
49e22585 CL |
5249 | |
5250 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5251 | struct page *page; |
5252 | ||
5253 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5254 | |
5255 | if (page) { | |
5256 | pages += page->pages; | |
5257 | objects += page->pobjects; | |
5258 | } | |
5259 | } | |
5260 | ||
bf16d19a | 5261 | len += sysfs_emit_at(buf, len, "%d(%d)", objects, pages); |
49e22585 CL |
5262 | |
5263 | #ifdef CONFIG_SMP | |
5264 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5265 | struct page *page; |
5266 | ||
5267 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
bf16d19a JP |
5268 | if (page) |
5269 | len += sysfs_emit_at(buf, len, " C%d=%d(%d)", | |
5270 | cpu, page->pobjects, page->pages); | |
49e22585 CL |
5271 | } |
5272 | #endif | |
bf16d19a JP |
5273 | len += sysfs_emit_at(buf, len, "\n"); |
5274 | ||
5275 | return len; | |
49e22585 CL |
5276 | } |
5277 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5278 | ||
a5a84755 CL |
5279 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5280 | { | |
bf16d19a | 5281 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); |
a5a84755 | 5282 | } |
8f58119a | 5283 | SLAB_ATTR_RO(reclaim_account); |
a5a84755 CL |
5284 | |
5285 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5286 | { | |
bf16d19a | 5287 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
a5a84755 CL |
5288 | } |
5289 | SLAB_ATTR_RO(hwcache_align); | |
5290 | ||
5291 | #ifdef CONFIG_ZONE_DMA | |
5292 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5293 | { | |
bf16d19a | 5294 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); |
a5a84755 CL |
5295 | } |
5296 | SLAB_ATTR_RO(cache_dma); | |
5297 | #endif | |
5298 | ||
8eb8284b DW |
5299 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5300 | { | |
bf16d19a | 5301 | return sysfs_emit(buf, "%u\n", s->usersize); |
8eb8284b DW |
5302 | } |
5303 | SLAB_ATTR_RO(usersize); | |
5304 | ||
a5a84755 CL |
5305 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5306 | { | |
bf16d19a | 5307 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5308 | } |
5309 | SLAB_ATTR_RO(destroy_by_rcu); | |
5310 | ||
ab4d5ed5 | 5311 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5312 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5313 | { | |
5314 | return show_slab_objects(s, buf, SO_ALL); | |
5315 | } | |
5316 | SLAB_ATTR_RO(slabs); | |
5317 | ||
205ab99d CL |
5318 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5319 | { | |
5320 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5321 | } | |
5322 | SLAB_ATTR_RO(total_objects); | |
5323 | ||
81819f0f CL |
5324 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5325 | { | |
bf16d19a | 5326 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f | 5327 | } |
060807f8 | 5328 | SLAB_ATTR_RO(sanity_checks); |
81819f0f CL |
5329 | |
5330 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5331 | { | |
bf16d19a | 5332 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE)); |
81819f0f | 5333 | } |
060807f8 | 5334 | SLAB_ATTR_RO(trace); |
81819f0f | 5335 | |
81819f0f CL |
5336 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5337 | { | |
bf16d19a | 5338 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); |
81819f0f CL |
5339 | } |
5340 | ||
ad38b5b1 | 5341 | SLAB_ATTR_RO(red_zone); |
81819f0f CL |
5342 | |
5343 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5344 | { | |
bf16d19a | 5345 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON)); |
81819f0f CL |
5346 | } |
5347 | ||
ad38b5b1 | 5348 | SLAB_ATTR_RO(poison); |
81819f0f CL |
5349 | |
5350 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5351 | { | |
bf16d19a | 5352 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); |
81819f0f CL |
5353 | } |
5354 | ||
ad38b5b1 | 5355 | SLAB_ATTR_RO(store_user); |
81819f0f | 5356 | |
53e15af0 CL |
5357 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5358 | { | |
5359 | return 0; | |
5360 | } | |
5361 | ||
5362 | static ssize_t validate_store(struct kmem_cache *s, | |
5363 | const char *buf, size_t length) | |
5364 | { | |
434e245d CL |
5365 | int ret = -EINVAL; |
5366 | ||
5367 | if (buf[0] == '1') { | |
5368 | ret = validate_slab_cache(s); | |
5369 | if (ret >= 0) | |
5370 | ret = length; | |
5371 | } | |
5372 | return ret; | |
53e15af0 CL |
5373 | } |
5374 | SLAB_ATTR(validate); | |
a5a84755 | 5375 | |
a5a84755 CL |
5376 | #endif /* CONFIG_SLUB_DEBUG */ |
5377 | ||
5378 | #ifdef CONFIG_FAILSLAB | |
5379 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5380 | { | |
bf16d19a | 5381 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); |
a5a84755 | 5382 | } |
060807f8 | 5383 | SLAB_ATTR_RO(failslab); |
ab4d5ed5 | 5384 | #endif |
53e15af0 | 5385 | |
2086d26a CL |
5386 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5387 | { | |
5388 | return 0; | |
5389 | } | |
5390 | ||
5391 | static ssize_t shrink_store(struct kmem_cache *s, | |
5392 | const char *buf, size_t length) | |
5393 | { | |
832f37f5 | 5394 | if (buf[0] == '1') |
10befea9 | 5395 | kmem_cache_shrink(s); |
832f37f5 | 5396 | else |
2086d26a CL |
5397 | return -EINVAL; |
5398 | return length; | |
5399 | } | |
5400 | SLAB_ATTR(shrink); | |
5401 | ||
81819f0f | 5402 | #ifdef CONFIG_NUMA |
9824601e | 5403 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5404 | { |
bf16d19a | 5405 | return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5406 | } |
5407 | ||
9824601e | 5408 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5409 | const char *buf, size_t length) |
5410 | { | |
eb7235eb | 5411 | unsigned int ratio; |
0121c619 CL |
5412 | int err; |
5413 | ||
eb7235eb | 5414 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5415 | if (err) |
5416 | return err; | |
eb7235eb AD |
5417 | if (ratio > 100) |
5418 | return -ERANGE; | |
0121c619 | 5419 | |
eb7235eb | 5420 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5421 | |
81819f0f CL |
5422 | return length; |
5423 | } | |
9824601e | 5424 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5425 | #endif |
5426 | ||
8ff12cfc | 5427 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5428 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5429 | { | |
5430 | unsigned long sum = 0; | |
5431 | int cpu; | |
bf16d19a | 5432 | int len = 0; |
6da2ec56 | 5433 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5434 | |
5435 | if (!data) | |
5436 | return -ENOMEM; | |
5437 | ||
5438 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5439 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5440 | |
5441 | data[cpu] = x; | |
5442 | sum += x; | |
5443 | } | |
5444 | ||
bf16d19a | 5445 | len += sysfs_emit_at(buf, len, "%lu", sum); |
8ff12cfc | 5446 | |
50ef37b9 | 5447 | #ifdef CONFIG_SMP |
8ff12cfc | 5448 | for_each_online_cpu(cpu) { |
bf16d19a JP |
5449 | if (data[cpu]) |
5450 | len += sysfs_emit_at(buf, len, " C%d=%u", | |
5451 | cpu, data[cpu]); | |
8ff12cfc | 5452 | } |
50ef37b9 | 5453 | #endif |
8ff12cfc | 5454 | kfree(data); |
bf16d19a JP |
5455 | len += sysfs_emit_at(buf, len, "\n"); |
5456 | ||
5457 | return len; | |
8ff12cfc CL |
5458 | } |
5459 | ||
78eb00cc DR |
5460 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5461 | { | |
5462 | int cpu; | |
5463 | ||
5464 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5465 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5466 | } |
5467 | ||
8ff12cfc CL |
5468 | #define STAT_ATTR(si, text) \ |
5469 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5470 | { \ | |
5471 | return show_stat(s, buf, si); \ | |
5472 | } \ | |
78eb00cc DR |
5473 | static ssize_t text##_store(struct kmem_cache *s, \ |
5474 | const char *buf, size_t length) \ | |
5475 | { \ | |
5476 | if (buf[0] != '0') \ | |
5477 | return -EINVAL; \ | |
5478 | clear_stat(s, si); \ | |
5479 | return length; \ | |
5480 | } \ | |
5481 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5482 | |
5483 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5484 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5485 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5486 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5487 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5488 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5489 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5490 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5491 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5492 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5493 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5494 | STAT_ATTR(FREE_SLAB, free_slab); |
5495 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5496 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5497 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5498 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5499 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5500 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5501 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5502 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5503 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5504 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5505 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5506 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5507 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5508 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5509 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5510 | |
06428780 | 5511 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5512 | &slab_size_attr.attr, |
5513 | &object_size_attr.attr, | |
5514 | &objs_per_slab_attr.attr, | |
5515 | &order_attr.attr, | |
73d342b1 | 5516 | &min_partial_attr.attr, |
49e22585 | 5517 | &cpu_partial_attr.attr, |
81819f0f | 5518 | &objects_attr.attr, |
205ab99d | 5519 | &objects_partial_attr.attr, |
81819f0f CL |
5520 | &partial_attr.attr, |
5521 | &cpu_slabs_attr.attr, | |
5522 | &ctor_attr.attr, | |
81819f0f CL |
5523 | &aliases_attr.attr, |
5524 | &align_attr.attr, | |
81819f0f CL |
5525 | &hwcache_align_attr.attr, |
5526 | &reclaim_account_attr.attr, | |
5527 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5528 | &shrink_attr.attr, |
49e22585 | 5529 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5530 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5531 | &total_objects_attr.attr, |
5532 | &slabs_attr.attr, | |
5533 | &sanity_checks_attr.attr, | |
5534 | &trace_attr.attr, | |
81819f0f CL |
5535 | &red_zone_attr.attr, |
5536 | &poison_attr.attr, | |
5537 | &store_user_attr.attr, | |
53e15af0 | 5538 | &validate_attr.attr, |
ab4d5ed5 | 5539 | #endif |
81819f0f CL |
5540 | #ifdef CONFIG_ZONE_DMA |
5541 | &cache_dma_attr.attr, | |
5542 | #endif | |
5543 | #ifdef CONFIG_NUMA | |
9824601e | 5544 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5545 | #endif |
5546 | #ifdef CONFIG_SLUB_STATS | |
5547 | &alloc_fastpath_attr.attr, | |
5548 | &alloc_slowpath_attr.attr, | |
5549 | &free_fastpath_attr.attr, | |
5550 | &free_slowpath_attr.attr, | |
5551 | &free_frozen_attr.attr, | |
5552 | &free_add_partial_attr.attr, | |
5553 | &free_remove_partial_attr.attr, | |
5554 | &alloc_from_partial_attr.attr, | |
5555 | &alloc_slab_attr.attr, | |
5556 | &alloc_refill_attr.attr, | |
e36a2652 | 5557 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5558 | &free_slab_attr.attr, |
5559 | &cpuslab_flush_attr.attr, | |
5560 | &deactivate_full_attr.attr, | |
5561 | &deactivate_empty_attr.attr, | |
5562 | &deactivate_to_head_attr.attr, | |
5563 | &deactivate_to_tail_attr.attr, | |
5564 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5565 | &deactivate_bypass_attr.attr, |
65c3376a | 5566 | &order_fallback_attr.attr, |
b789ef51 CL |
5567 | &cmpxchg_double_fail_attr.attr, |
5568 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5569 | &cpu_partial_alloc_attr.attr, |
5570 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5571 | &cpu_partial_node_attr.attr, |
5572 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5573 | #endif |
4c13dd3b DM |
5574 | #ifdef CONFIG_FAILSLAB |
5575 | &failslab_attr.attr, | |
5576 | #endif | |
8eb8284b | 5577 | &usersize_attr.attr, |
4c13dd3b | 5578 | |
81819f0f CL |
5579 | NULL |
5580 | }; | |
5581 | ||
1fdaaa23 | 5582 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5583 | .attrs = slab_attrs, |
5584 | }; | |
5585 | ||
5586 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5587 | struct attribute *attr, | |
5588 | char *buf) | |
5589 | { | |
5590 | struct slab_attribute *attribute; | |
5591 | struct kmem_cache *s; | |
5592 | int err; | |
5593 | ||
5594 | attribute = to_slab_attr(attr); | |
5595 | s = to_slab(kobj); | |
5596 | ||
5597 | if (!attribute->show) | |
5598 | return -EIO; | |
5599 | ||
5600 | err = attribute->show(s, buf); | |
5601 | ||
5602 | return err; | |
5603 | } | |
5604 | ||
5605 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5606 | struct attribute *attr, | |
5607 | const char *buf, size_t len) | |
5608 | { | |
5609 | struct slab_attribute *attribute; | |
5610 | struct kmem_cache *s; | |
5611 | int err; | |
5612 | ||
5613 | attribute = to_slab_attr(attr); | |
5614 | s = to_slab(kobj); | |
5615 | ||
5616 | if (!attribute->store) | |
5617 | return -EIO; | |
5618 | ||
5619 | err = attribute->store(s, buf, len); | |
81819f0f CL |
5620 | return err; |
5621 | } | |
5622 | ||
41a21285 CL |
5623 | static void kmem_cache_release(struct kobject *k) |
5624 | { | |
5625 | slab_kmem_cache_release(to_slab(k)); | |
5626 | } | |
5627 | ||
52cf25d0 | 5628 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5629 | .show = slab_attr_show, |
5630 | .store = slab_attr_store, | |
5631 | }; | |
5632 | ||
5633 | static struct kobj_type slab_ktype = { | |
5634 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5635 | .release = kmem_cache_release, |
81819f0f CL |
5636 | }; |
5637 | ||
27c3a314 | 5638 | static struct kset *slab_kset; |
81819f0f | 5639 | |
9a41707b VD |
5640 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5641 | { | |
9a41707b VD |
5642 | return slab_kset; |
5643 | } | |
5644 | ||
81819f0f CL |
5645 | #define ID_STR_LENGTH 64 |
5646 | ||
5647 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5648 | * |
5649 | * Format :[flags-]size | |
81819f0f CL |
5650 | */ |
5651 | static char *create_unique_id(struct kmem_cache *s) | |
5652 | { | |
5653 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5654 | char *p = name; | |
5655 | ||
5656 | BUG_ON(!name); | |
5657 | ||
5658 | *p++ = ':'; | |
5659 | /* | |
5660 | * First flags affecting slabcache operations. We will only | |
5661 | * get here for aliasable slabs so we do not need to support | |
5662 | * too many flags. The flags here must cover all flags that | |
5663 | * are matched during merging to guarantee that the id is | |
5664 | * unique. | |
5665 | */ | |
5666 | if (s->flags & SLAB_CACHE_DMA) | |
5667 | *p++ = 'd'; | |
6d6ea1e9 NB |
5668 | if (s->flags & SLAB_CACHE_DMA32) |
5669 | *p++ = 'D'; | |
81819f0f CL |
5670 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5671 | *p++ = 'a'; | |
becfda68 | 5672 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5673 | *p++ = 'F'; |
230e9fc2 VD |
5674 | if (s->flags & SLAB_ACCOUNT) |
5675 | *p++ = 'A'; | |
81819f0f CL |
5676 | if (p != name + 1) |
5677 | *p++ = '-'; | |
44065b2e | 5678 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5679 | |
81819f0f CL |
5680 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5681 | return name; | |
5682 | } | |
5683 | ||
5684 | static int sysfs_slab_add(struct kmem_cache *s) | |
5685 | { | |
5686 | int err; | |
5687 | const char *name; | |
1663f26d | 5688 | struct kset *kset = cache_kset(s); |
45530c44 | 5689 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5690 | |
1663f26d TH |
5691 | if (!kset) { |
5692 | kobject_init(&s->kobj, &slab_ktype); | |
5693 | return 0; | |
5694 | } | |
5695 | ||
11066386 MC |
5696 | if (!unmergeable && disable_higher_order_debug && |
5697 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5698 | unmergeable = 1; | |
5699 | ||
81819f0f CL |
5700 | if (unmergeable) { |
5701 | /* | |
5702 | * Slabcache can never be merged so we can use the name proper. | |
5703 | * This is typically the case for debug situations. In that | |
5704 | * case we can catch duplicate names easily. | |
5705 | */ | |
27c3a314 | 5706 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5707 | name = s->name; |
5708 | } else { | |
5709 | /* | |
5710 | * Create a unique name for the slab as a target | |
5711 | * for the symlinks. | |
5712 | */ | |
5713 | name = create_unique_id(s); | |
5714 | } | |
5715 | ||
1663f26d | 5716 | s->kobj.kset = kset; |
26e4f205 | 5717 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
757fed1d | 5718 | if (err) |
80da026a | 5719 | goto out; |
81819f0f CL |
5720 | |
5721 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5722 | if (err) |
5723 | goto out_del_kobj; | |
9a41707b | 5724 | |
81819f0f CL |
5725 | if (!unmergeable) { |
5726 | /* Setup first alias */ | |
5727 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5728 | } |
54b6a731 DJ |
5729 | out: |
5730 | if (!unmergeable) | |
5731 | kfree(name); | |
5732 | return err; | |
5733 | out_del_kobj: | |
5734 | kobject_del(&s->kobj); | |
54b6a731 | 5735 | goto out; |
81819f0f CL |
5736 | } |
5737 | ||
d50d82fa MP |
5738 | void sysfs_slab_unlink(struct kmem_cache *s) |
5739 | { | |
5740 | if (slab_state >= FULL) | |
5741 | kobject_del(&s->kobj); | |
5742 | } | |
5743 | ||
bf5eb3de TH |
5744 | void sysfs_slab_release(struct kmem_cache *s) |
5745 | { | |
5746 | if (slab_state >= FULL) | |
5747 | kobject_put(&s->kobj); | |
81819f0f CL |
5748 | } |
5749 | ||
5750 | /* | |
5751 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5752 | * available lest we lose that information. |
81819f0f CL |
5753 | */ |
5754 | struct saved_alias { | |
5755 | struct kmem_cache *s; | |
5756 | const char *name; | |
5757 | struct saved_alias *next; | |
5758 | }; | |
5759 | ||
5af328a5 | 5760 | static struct saved_alias *alias_list; |
81819f0f CL |
5761 | |
5762 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5763 | { | |
5764 | struct saved_alias *al; | |
5765 | ||
97d06609 | 5766 | if (slab_state == FULL) { |
81819f0f CL |
5767 | /* |
5768 | * If we have a leftover link then remove it. | |
5769 | */ | |
27c3a314 GKH |
5770 | sysfs_remove_link(&slab_kset->kobj, name); |
5771 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5772 | } |
5773 | ||
5774 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5775 | if (!al) | |
5776 | return -ENOMEM; | |
5777 | ||
5778 | al->s = s; | |
5779 | al->name = name; | |
5780 | al->next = alias_list; | |
5781 | alias_list = al; | |
5782 | return 0; | |
5783 | } | |
5784 | ||
5785 | static int __init slab_sysfs_init(void) | |
5786 | { | |
5b95a4ac | 5787 | struct kmem_cache *s; |
81819f0f CL |
5788 | int err; |
5789 | ||
18004c5d | 5790 | mutex_lock(&slab_mutex); |
2bce6485 | 5791 | |
d7660ce5 | 5792 | slab_kset = kset_create_and_add("slab", NULL, kernel_kobj); |
27c3a314 | 5793 | if (!slab_kset) { |
18004c5d | 5794 | mutex_unlock(&slab_mutex); |
f9f58285 | 5795 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5796 | return -ENOSYS; |
5797 | } | |
5798 | ||
97d06609 | 5799 | slab_state = FULL; |
26a7bd03 | 5800 | |
5b95a4ac | 5801 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5802 | err = sysfs_slab_add(s); |
5d540fb7 | 5803 | if (err) |
f9f58285 FF |
5804 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5805 | s->name); | |
26a7bd03 | 5806 | } |
81819f0f CL |
5807 | |
5808 | while (alias_list) { | |
5809 | struct saved_alias *al = alias_list; | |
5810 | ||
5811 | alias_list = alias_list->next; | |
5812 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5813 | if (err) |
f9f58285 FF |
5814 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5815 | al->name); | |
81819f0f CL |
5816 | kfree(al); |
5817 | } | |
5818 | ||
18004c5d | 5819 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5820 | return 0; |
5821 | } | |
5822 | ||
5823 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5824 | #endif /* CONFIG_SYSFS */ |
57ed3eda | 5825 | |
64dd6849 FM |
5826 | #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS) |
5827 | static int slab_debugfs_show(struct seq_file *seq, void *v) | |
5828 | { | |
5829 | ||
5830 | struct location *l; | |
5831 | unsigned int idx = *(unsigned int *)v; | |
5832 | struct loc_track *t = seq->private; | |
5833 | ||
5834 | if (idx < t->count) { | |
5835 | l = &t->loc[idx]; | |
5836 | ||
5837 | seq_printf(seq, "%7ld ", l->count); | |
5838 | ||
5839 | if (l->addr) | |
5840 | seq_printf(seq, "%pS", (void *)l->addr); | |
5841 | else | |
5842 | seq_puts(seq, "<not-available>"); | |
5843 | ||
5844 | if (l->sum_time != l->min_time) { | |
5845 | seq_printf(seq, " age=%ld/%llu/%ld", | |
5846 | l->min_time, div_u64(l->sum_time, l->count), | |
5847 | l->max_time); | |
5848 | } else | |
5849 | seq_printf(seq, " age=%ld", l->min_time); | |
5850 | ||
5851 | if (l->min_pid != l->max_pid) | |
5852 | seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid); | |
5853 | else | |
5854 | seq_printf(seq, " pid=%ld", | |
5855 | l->min_pid); | |
5856 | ||
5857 | if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus))) | |
5858 | seq_printf(seq, " cpus=%*pbl", | |
5859 | cpumask_pr_args(to_cpumask(l->cpus))); | |
5860 | ||
5861 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes)) | |
5862 | seq_printf(seq, " nodes=%*pbl", | |
5863 | nodemask_pr_args(&l->nodes)); | |
5864 | ||
5865 | seq_puts(seq, "\n"); | |
5866 | } | |
5867 | ||
5868 | if (!idx && !t->count) | |
5869 | seq_puts(seq, "No data\n"); | |
5870 | ||
5871 | return 0; | |
5872 | } | |
5873 | ||
5874 | static void slab_debugfs_stop(struct seq_file *seq, void *v) | |
5875 | { | |
5876 | } | |
5877 | ||
5878 | static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos) | |
5879 | { | |
5880 | struct loc_track *t = seq->private; | |
5881 | ||
5882 | v = ppos; | |
5883 | ++*ppos; | |
5884 | if (*ppos <= t->count) | |
5885 | return v; | |
5886 | ||
5887 | return NULL; | |
5888 | } | |
5889 | ||
5890 | static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos) | |
5891 | { | |
5892 | return ppos; | |
5893 | } | |
5894 | ||
5895 | static const struct seq_operations slab_debugfs_sops = { | |
5896 | .start = slab_debugfs_start, | |
5897 | .next = slab_debugfs_next, | |
5898 | .stop = slab_debugfs_stop, | |
5899 | .show = slab_debugfs_show, | |
5900 | }; | |
5901 | ||
5902 | static int slab_debug_trace_open(struct inode *inode, struct file *filep) | |
5903 | { | |
5904 | ||
5905 | struct kmem_cache_node *n; | |
5906 | enum track_item alloc; | |
5907 | int node; | |
5908 | struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops, | |
5909 | sizeof(struct loc_track)); | |
5910 | struct kmem_cache *s = file_inode(filep)->i_private; | |
b3fd64e1 VB |
5911 | unsigned long *obj_map; |
5912 | ||
5913 | obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL); | |
5914 | if (!obj_map) | |
5915 | return -ENOMEM; | |
64dd6849 FM |
5916 | |
5917 | if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0) | |
5918 | alloc = TRACK_ALLOC; | |
5919 | else | |
5920 | alloc = TRACK_FREE; | |
5921 | ||
b3fd64e1 VB |
5922 | if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) { |
5923 | bitmap_free(obj_map); | |
64dd6849 | 5924 | return -ENOMEM; |
b3fd64e1 | 5925 | } |
64dd6849 | 5926 | |
64dd6849 FM |
5927 | for_each_kmem_cache_node(s, node, n) { |
5928 | unsigned long flags; | |
5929 | struct page *page; | |
5930 | ||
5931 | if (!atomic_long_read(&n->nr_slabs)) | |
5932 | continue; | |
5933 | ||
5934 | spin_lock_irqsave(&n->list_lock, flags); | |
5935 | list_for_each_entry(page, &n->partial, slab_list) | |
b3fd64e1 | 5936 | process_slab(t, s, page, alloc, obj_map); |
64dd6849 | 5937 | list_for_each_entry(page, &n->full, slab_list) |
b3fd64e1 | 5938 | process_slab(t, s, page, alloc, obj_map); |
64dd6849 FM |
5939 | spin_unlock_irqrestore(&n->list_lock, flags); |
5940 | } | |
5941 | ||
b3fd64e1 | 5942 | bitmap_free(obj_map); |
64dd6849 FM |
5943 | return 0; |
5944 | } | |
5945 | ||
5946 | static int slab_debug_trace_release(struct inode *inode, struct file *file) | |
5947 | { | |
5948 | struct seq_file *seq = file->private_data; | |
5949 | struct loc_track *t = seq->private; | |
5950 | ||
5951 | free_loc_track(t); | |
5952 | return seq_release_private(inode, file); | |
5953 | } | |
5954 | ||
5955 | static const struct file_operations slab_debugfs_fops = { | |
5956 | .open = slab_debug_trace_open, | |
5957 | .read = seq_read, | |
5958 | .llseek = seq_lseek, | |
5959 | .release = slab_debug_trace_release, | |
5960 | }; | |
5961 | ||
5962 | static void debugfs_slab_add(struct kmem_cache *s) | |
5963 | { | |
5964 | struct dentry *slab_cache_dir; | |
5965 | ||
5966 | if (unlikely(!slab_debugfs_root)) | |
5967 | return; | |
5968 | ||
5969 | slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root); | |
5970 | ||
5971 | debugfs_create_file("alloc_traces", 0400, | |
5972 | slab_cache_dir, s, &slab_debugfs_fops); | |
5973 | ||
5974 | debugfs_create_file("free_traces", 0400, | |
5975 | slab_cache_dir, s, &slab_debugfs_fops); | |
5976 | } | |
5977 | ||
5978 | void debugfs_slab_release(struct kmem_cache *s) | |
5979 | { | |
5980 | debugfs_remove_recursive(debugfs_lookup(s->name, slab_debugfs_root)); | |
5981 | } | |
5982 | ||
5983 | static int __init slab_debugfs_init(void) | |
5984 | { | |
5985 | struct kmem_cache *s; | |
5986 | ||
5987 | slab_debugfs_root = debugfs_create_dir("slab", NULL); | |
5988 | ||
5989 | list_for_each_entry(s, &slab_caches, list) | |
5990 | if (s->flags & SLAB_STORE_USER) | |
5991 | debugfs_slab_add(s); | |
5992 | ||
5993 | return 0; | |
5994 | ||
5995 | } | |
5996 | __initcall(slab_debugfs_init); | |
5997 | #endif | |
57ed3eda PE |
5998 | /* |
5999 | * The /proc/slabinfo ABI | |
6000 | */ | |
5b365771 | 6001 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 6002 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 6003 | { |
57ed3eda | 6004 | unsigned long nr_slabs = 0; |
205ab99d CL |
6005 | unsigned long nr_objs = 0; |
6006 | unsigned long nr_free = 0; | |
57ed3eda | 6007 | int node; |
fa45dc25 | 6008 | struct kmem_cache_node *n; |
57ed3eda | 6009 | |
fa45dc25 | 6010 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
6011 | nr_slabs += node_nr_slabs(n); |
6012 | nr_objs += node_nr_objs(n); | |
205ab99d | 6013 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
6014 | } |
6015 | ||
0d7561c6 GC |
6016 | sinfo->active_objs = nr_objs - nr_free; |
6017 | sinfo->num_objs = nr_objs; | |
6018 | sinfo->active_slabs = nr_slabs; | |
6019 | sinfo->num_slabs = nr_slabs; | |
6020 | sinfo->objects_per_slab = oo_objects(s->oo); | |
6021 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
6022 | } |
6023 | ||
0d7561c6 | 6024 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 6025 | { |
7b3c3a50 AD |
6026 | } |
6027 | ||
b7454ad3 GC |
6028 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
6029 | size_t count, loff_t *ppos) | |
7b3c3a50 | 6030 | { |
b7454ad3 | 6031 | return -EIO; |
7b3c3a50 | 6032 | } |
5b365771 | 6033 | #endif /* CONFIG_SLUB_DEBUG */ |