]>
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> | |
c7b23b68 | 14 | #include <linux/swap.h> /* mm_account_reclaimed_pages() */ |
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> |
68ef169a | 25 | #include <linux/kmsan.h> |
81819f0f CL |
26 | #include <linux/cpu.h> |
27 | #include <linux/cpuset.h> | |
28 | #include <linux/mempolicy.h> | |
29 | #include <linux/ctype.h> | |
5cf909c5 | 30 | #include <linux/stackdepot.h> |
3ac7fe5a | 31 | #include <linux/debugobjects.h> |
81819f0f | 32 | #include <linux/kallsyms.h> |
b89fb5ef | 33 | #include <linux/kfence.h> |
b9049e23 | 34 | #include <linux/memory.h> |
f8bd2258 | 35 | #include <linux/math64.h> |
773ff60e | 36 | #include <linux/fault-inject.h> |
6011be59 | 37 | #include <linux/kmemleak.h> |
bfa71457 | 38 | #include <linux/stacktrace.h> |
4de900b4 | 39 | #include <linux/prefetch.h> |
2633d7a0 | 40 | #include <linux/memcontrol.h> |
2482ddec | 41 | #include <linux/random.h> |
1f9f78b1 | 42 | #include <kunit/test.h> |
909c6475 | 43 | #include <kunit/test-bug.h> |
553c0369 | 44 | #include <linux/sort.h> |
81819f0f | 45 | |
64dd6849 | 46 | #include <linux/debugfs.h> |
4a92379b RK |
47 | #include <trace/events/kmem.h> |
48 | ||
072bb0aa MG |
49 | #include "internal.h" |
50 | ||
81819f0f CL |
51 | /* |
52 | * Lock order: | |
18004c5d | 53 | * 1. slab_mutex (Global Mutex) |
bd0e7491 VB |
54 | * 2. node->list_lock (Spinlock) |
55 | * 3. kmem_cache->cpu_slab->lock (Local lock) | |
41bec7c3 | 56 | * 4. slab_lock(slab) (Only on some arches) |
bd0e7491 | 57 | * 5. object_map_lock (Only for debugging) |
81819f0f | 58 | * |
18004c5d | 59 | * slab_mutex |
881db7fb | 60 | * |
18004c5d | 61 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb | 62 | * and to synchronize major metadata changes to slab cache structures. |
bd0e7491 VB |
63 | * Also synchronizes memory hotplug callbacks. |
64 | * | |
65 | * slab_lock | |
66 | * | |
67 | * The slab_lock is a wrapper around the page lock, thus it is a bit | |
68 | * spinlock. | |
881db7fb | 69 | * |
41bec7c3 VB |
70 | * The slab_lock is only used on arches that do not have the ability |
71 | * to do a cmpxchg_double. It only protects: | |
72 | * | |
c2092c12 VB |
73 | * A. slab->freelist -> List of free objects in a slab |
74 | * B. slab->inuse -> Number of objects in use | |
75 | * C. slab->objects -> Number of objects in slab | |
76 | * D. slab->frozen -> frozen state | |
881db7fb | 77 | * |
bd0e7491 VB |
78 | * Frozen slabs |
79 | * | |
31bda717 CZ |
80 | * If a slab is frozen then it is exempt from list management. It is |
81 | * the cpu slab which is actively allocated from by the processor that | |
82 | * froze it and it is not on any list. The processor that froze the | |
c2092c12 | 83 | * slab is the one who can perform list operations on the slab. Other |
632b2ef0 LX |
84 | * processors may put objects onto the freelist but the processor that |
85 | * froze the slab is the only one that can retrieve the objects from the | |
c2092c12 | 86 | * slab's freelist. |
81819f0f | 87 | * |
31bda717 CZ |
88 | * CPU partial slabs |
89 | * | |
90 | * The partially empty slabs cached on the CPU partial list are used | |
91 | * for performance reasons, which speeds up the allocation process. | |
92 | * These slabs are not frozen, but are also exempt from list management, | |
93 | * by clearing the PG_workingset flag when moving out of the node | |
94 | * partial list. Please see __slab_free() for more details. | |
95 | * | |
96 | * To sum up, the current scheme is: | |
97 | * - node partial slab: PG_Workingset && !frozen | |
98 | * - cpu partial slab: !PG_Workingset && !frozen | |
99 | * - cpu slab: !PG_Workingset && frozen | |
100 | * - full slab: !PG_Workingset && !frozen | |
101 | * | |
bd0e7491 VB |
102 | * list_lock |
103 | * | |
81819f0f CL |
104 | * The list_lock protects the partial and full list on each node and |
105 | * the partial slab counter. If taken then no new slabs may be added or | |
106 | * removed from the lists nor make the number of partial slabs be modified. | |
107 | * (Note that the total number of slabs is an atomic value that may be | |
108 | * modified without taking the list lock). | |
109 | * | |
110 | * The list_lock is a centralized lock and thus we avoid taking it as | |
111 | * much as possible. As long as SLUB does not have to handle partial | |
112 | * slabs, operations can continue without any centralized lock. F.e. | |
113 | * allocating a long series of objects that fill up slabs does not require | |
114 | * the list lock. | |
bd0e7491 | 115 | * |
41bec7c3 VB |
116 | * For debug caches, all allocations are forced to go through a list_lock |
117 | * protected region to serialize against concurrent validation. | |
118 | * | |
bd0e7491 VB |
119 | * cpu_slab->lock local lock |
120 | * | |
121 | * This locks protect slowpath manipulation of all kmem_cache_cpu fields | |
122 | * except the stat counters. This is a percpu structure manipulated only by | |
123 | * the local cpu, so the lock protects against being preempted or interrupted | |
124 | * by an irq. Fast path operations rely on lockless operations instead. | |
1f04b07d TG |
125 | * |
126 | * On PREEMPT_RT, the local lock neither disables interrupts nor preemption | |
127 | * which means the lockless fastpath cannot be used as it might interfere with | |
128 | * an in-progress slow path operations. In this case the local lock is always | |
129 | * taken but it still utilizes the freelist for the common operations. | |
bd0e7491 VB |
130 | * |
131 | * lockless fastpaths | |
132 | * | |
133 | * The fast path allocation (slab_alloc_node()) and freeing (do_slab_free()) | |
134 | * are fully lockless when satisfied from the percpu slab (and when | |
135 | * cmpxchg_double is possible to use, otherwise slab_lock is taken). | |
136 | * They also don't disable preemption or migration or irqs. They rely on | |
137 | * the transaction id (tid) field to detect being preempted or moved to | |
138 | * another cpu. | |
139 | * | |
140 | * irq, preemption, migration considerations | |
141 | * | |
142 | * Interrupts are disabled as part of list_lock or local_lock operations, or | |
143 | * around the slab_lock operation, in order to make the slab allocator safe | |
144 | * to use in the context of an irq. | |
145 | * | |
146 | * In addition, preemption (or migration on PREEMPT_RT) is disabled in the | |
147 | * allocation slowpath, bulk allocation, and put_cpu_partial(), so that the | |
148 | * local cpu doesn't change in the process and e.g. the kmem_cache_cpu pointer | |
149 | * doesn't have to be revalidated in each section protected by the local lock. | |
81819f0f CL |
150 | * |
151 | * SLUB assigns one slab for allocation to each processor. | |
152 | * Allocations only occur from these slabs called cpu slabs. | |
153 | * | |
672bba3a CL |
154 | * Slabs with free elements are kept on a partial list and during regular |
155 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 156 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
157 | * We track full slabs for debugging purposes though because otherwise we |
158 | * cannot scan all objects. | |
81819f0f CL |
159 | * |
160 | * Slabs are freed when they become empty. Teardown and setup is | |
161 | * minimal so we rely on the page allocators per cpu caches for | |
162 | * fast frees and allocs. | |
163 | * | |
c2092c12 | 164 | * slab->frozen The slab is frozen and exempt from list processing. |
4b6f0750 CL |
165 | * This means that the slab is dedicated to a purpose |
166 | * such as satisfying allocations for a specific | |
167 | * processor. Objects may be freed in the slab while | |
168 | * it is frozen but slab_free will then skip the usual | |
169 | * list operations. It is up to the processor holding | |
170 | * the slab to integrate the slab into the slab lists | |
171 | * when the slab is no longer needed. | |
172 | * | |
173 | * One use of this flag is to mark slabs that are | |
174 | * used for allocations. Then such a slab becomes a cpu | |
175 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 176 | * freelist that allows lockless access to |
894b8788 CL |
177 | * free objects in addition to the regular freelist |
178 | * that requires the slab lock. | |
81819f0f | 179 | * |
aed68148 | 180 | * SLAB_DEBUG_FLAGS Slab requires special handling due to debug |
81819f0f | 181 | * options set. This moves slab handling out of |
894b8788 | 182 | * the fast path and disables lockless freelists. |
81819f0f CL |
183 | */ |
184 | ||
25c00c50 VB |
185 | /* |
186 | * We could simply use migrate_disable()/enable() but as long as it's a | |
187 | * function call even on !PREEMPT_RT, use inline preempt_disable() there. | |
188 | */ | |
189 | #ifndef CONFIG_PREEMPT_RT | |
1f04b07d TG |
190 | #define slub_get_cpu_ptr(var) get_cpu_ptr(var) |
191 | #define slub_put_cpu_ptr(var) put_cpu_ptr(var) | |
192 | #define USE_LOCKLESS_FAST_PATH() (true) | |
25c00c50 VB |
193 | #else |
194 | #define slub_get_cpu_ptr(var) \ | |
195 | ({ \ | |
196 | migrate_disable(); \ | |
197 | this_cpu_ptr(var); \ | |
198 | }) | |
199 | #define slub_put_cpu_ptr(var) \ | |
200 | do { \ | |
201 | (void)(var); \ | |
202 | migrate_enable(); \ | |
203 | } while (0) | |
1f04b07d | 204 | #define USE_LOCKLESS_FAST_PATH() (false) |
25c00c50 VB |
205 | #endif |
206 | ||
be784ba8 VB |
207 | #ifndef CONFIG_SLUB_TINY |
208 | #define __fastpath_inline __always_inline | |
209 | #else | |
210 | #define __fastpath_inline | |
211 | #endif | |
212 | ||
ca0cab65 VB |
213 | #ifdef CONFIG_SLUB_DEBUG |
214 | #ifdef CONFIG_SLUB_DEBUG_ON | |
215 | DEFINE_STATIC_KEY_TRUE(slub_debug_enabled); | |
216 | #else | |
217 | DEFINE_STATIC_KEY_FALSE(slub_debug_enabled); | |
218 | #endif | |
79270291 | 219 | #endif /* CONFIG_SLUB_DEBUG */ |
ca0cab65 | 220 | |
6edf2576 FT |
221 | /* Structure holding parameters for get_partial() call chain */ |
222 | struct partial_context { | |
6edf2576 FT |
223 | gfp_t flags; |
224 | unsigned int orig_size; | |
43c4c349 | 225 | void *object; |
6edf2576 FT |
226 | }; |
227 | ||
59052e89 VB |
228 | static inline bool kmem_cache_debug(struct kmem_cache *s) |
229 | { | |
230 | return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS); | |
af537b0a | 231 | } |
5577bd8a | 232 | |
6edf2576 FT |
233 | static inline bool slub_debug_orig_size(struct kmem_cache *s) |
234 | { | |
235 | return (kmem_cache_debug_flags(s, SLAB_STORE_USER) && | |
236 | (s->flags & SLAB_KMALLOC)); | |
237 | } | |
238 | ||
117d54df | 239 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be | 240 | { |
59052e89 | 241 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) |
d86bd1be JK |
242 | p += s->red_left_pad; |
243 | ||
244 | return p; | |
245 | } | |
246 | ||
345c905d JK |
247 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
248 | { | |
249 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
250 | return !kmem_cache_debug(s); | |
251 | #else | |
252 | return false; | |
253 | #endif | |
254 | } | |
255 | ||
81819f0f CL |
256 | /* |
257 | * Issues still to be resolved: | |
258 | * | |
81819f0f CL |
259 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
260 | * | |
81819f0f CL |
261 | * - Variable sizing of the per node arrays |
262 | */ | |
263 | ||
b789ef51 CL |
264 | /* Enable to log cmpxchg failures */ |
265 | #undef SLUB_DEBUG_CMPXCHG | |
266 | ||
5a8a3c1f | 267 | #ifndef CONFIG_SLUB_TINY |
2086d26a | 268 | /* |
dc84207d | 269 | * Minimum number of partial slabs. These will be left on the partial |
2086d26a CL |
270 | * lists even if they are empty. kmem_cache_shrink may reclaim them. |
271 | */ | |
76be8950 | 272 | #define MIN_PARTIAL 5 |
e95eed57 | 273 | |
2086d26a CL |
274 | /* |
275 | * Maximum number of desirable partial slabs. | |
276 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 277 | * sort the partial list by the number of objects in use. |
2086d26a CL |
278 | */ |
279 | #define MAX_PARTIAL 10 | |
5a8a3c1f VB |
280 | #else |
281 | #define MIN_PARTIAL 0 | |
282 | #define MAX_PARTIAL 0 | |
283 | #endif | |
2086d26a | 284 | |
becfda68 | 285 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 286 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 287 | |
149daaf3 LA |
288 | /* |
289 | * These debug flags cannot use CMPXCHG because there might be consistency | |
290 | * issues when checking or reading debug information | |
291 | */ | |
292 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
293 | SLAB_TRACE) | |
294 | ||
295 | ||
fa5ec8a1 | 296 | /* |
3de47213 | 297 | * Debugging flags that require metadata to be stored in the slab. These get |
671776b3 | 298 | * disabled when slab_debug=O is used and a cache's min order increases with |
3de47213 | 299 | * metadata. |
fa5ec8a1 | 300 | */ |
3de47213 | 301 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 302 | |
210b5c06 CG |
303 | #define OO_SHIFT 16 |
304 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
c2092c12 | 305 | #define MAX_OBJS_PER_PAGE 32767 /* since slab.objects is u15 */ |
210b5c06 | 306 | |
81819f0f | 307 | /* Internal SLUB flags */ |
d50112ed | 308 | /* Poison object */ |
cc61eb85 | 309 | #define __OBJECT_POISON __SLAB_FLAG_BIT(_SLAB_OBJECT_POISON) |
d50112ed | 310 | /* Use cmpxchg_double */ |
6801be4f PZ |
311 | |
312 | #ifdef system_has_freelist_aba | |
cc61eb85 | 313 | #define __CMPXCHG_DOUBLE __SLAB_FLAG_BIT(_SLAB_CMPXCHG_DOUBLE) |
6801be4f | 314 | #else |
cc61eb85 | 315 | #define __CMPXCHG_DOUBLE __SLAB_FLAG_UNUSED |
6801be4f | 316 | #endif |
81819f0f | 317 | |
02cbc874 CL |
318 | /* |
319 | * Tracking user of a slab. | |
320 | */ | |
d6543e39 | 321 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 322 | struct track { |
ce71e27c | 323 | unsigned long addr; /* Called from address */ |
5cf909c5 OG |
324 | #ifdef CONFIG_STACKDEPOT |
325 | depot_stack_handle_t handle; | |
d6543e39 | 326 | #endif |
02cbc874 CL |
327 | int cpu; /* Was running on cpu */ |
328 | int pid; /* Pid context */ | |
329 | unsigned long when; /* When did the operation occur */ | |
330 | }; | |
331 | ||
332 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
333 | ||
b1a413a3 | 334 | #ifdef SLAB_SUPPORTS_SYSFS |
81819f0f CL |
335 | static int sysfs_slab_add(struct kmem_cache *); |
336 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
81819f0f | 337 | #else |
0c710013 CL |
338 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
339 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
340 | { return 0; } | |
81819f0f CL |
341 | #endif |
342 | ||
64dd6849 FM |
343 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG) |
344 | static void debugfs_slab_add(struct kmem_cache *); | |
345 | #else | |
346 | static inline void debugfs_slab_add(struct kmem_cache *s) { } | |
347 | #endif | |
348 | ||
7ef08ae8 VB |
349 | enum stat_item { |
350 | ALLOC_FASTPATH, /* Allocation from cpu slab */ | |
351 | ALLOC_SLOWPATH, /* Allocation by getting a new cpu slab */ | |
352 | FREE_FASTPATH, /* Free to cpu slab */ | |
353 | FREE_SLOWPATH, /* Freeing not to cpu slab */ | |
354 | FREE_FROZEN, /* Freeing to frozen slab */ | |
355 | FREE_ADD_PARTIAL, /* Freeing moves slab to partial list */ | |
356 | FREE_REMOVE_PARTIAL, /* Freeing removes last object */ | |
357 | ALLOC_FROM_PARTIAL, /* Cpu slab acquired from node partial list */ | |
358 | ALLOC_SLAB, /* Cpu slab acquired from page allocator */ | |
359 | ALLOC_REFILL, /* Refill cpu slab from slab freelist */ | |
360 | ALLOC_NODE_MISMATCH, /* Switching cpu slab */ | |
361 | FREE_SLAB, /* Slab freed to the page allocator */ | |
362 | CPUSLAB_FLUSH, /* Abandoning of the cpu slab */ | |
363 | DEACTIVATE_FULL, /* Cpu slab was full when deactivated */ | |
364 | DEACTIVATE_EMPTY, /* Cpu slab was empty when deactivated */ | |
365 | DEACTIVATE_TO_HEAD, /* Cpu slab was moved to the head of partials */ | |
366 | DEACTIVATE_TO_TAIL, /* Cpu slab was moved to the tail of partials */ | |
367 | DEACTIVATE_REMOTE_FREES,/* Slab contained remotely freed objects */ | |
368 | DEACTIVATE_BYPASS, /* Implicit deactivation */ | |
369 | ORDER_FALLBACK, /* Number of times fallback was necessary */ | |
370 | CMPXCHG_DOUBLE_CPU_FAIL,/* Failures of this_cpu_cmpxchg_double */ | |
371 | CMPXCHG_DOUBLE_FAIL, /* Failures of slab freelist update */ | |
372 | CPU_PARTIAL_ALLOC, /* Used cpu partial on alloc */ | |
373 | CPU_PARTIAL_FREE, /* Refill cpu partial on free */ | |
374 | CPU_PARTIAL_NODE, /* Refill cpu partial from node partial */ | |
375 | CPU_PARTIAL_DRAIN, /* Drain cpu partial to node partial */ | |
376 | NR_SLUB_STAT_ITEMS | |
377 | }; | |
378 | ||
379 | #ifndef CONFIG_SLUB_TINY | |
380 | /* | |
381 | * When changing the layout, make sure freelist and tid are still compatible | |
382 | * with this_cpu_cmpxchg_double() alignment requirements. | |
383 | */ | |
384 | struct kmem_cache_cpu { | |
385 | union { | |
386 | struct { | |
387 | void **freelist; /* Pointer to next available object */ | |
388 | unsigned long tid; /* Globally unique transaction id */ | |
389 | }; | |
390 | freelist_aba_t freelist_tid; | |
391 | }; | |
392 | struct slab *slab; /* The slab from which we are allocating */ | |
393 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
c94d2224 | 394 | struct slab *partial; /* Partially allocated slabs */ |
7ef08ae8 VB |
395 | #endif |
396 | local_lock_t lock; /* Protects the fields above */ | |
397 | #ifdef CONFIG_SLUB_STATS | |
398 | unsigned int stat[NR_SLUB_STAT_ITEMS]; | |
399 | #endif | |
400 | }; | |
401 | #endif /* CONFIG_SLUB_TINY */ | |
402 | ||
4fdccdfb | 403 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
404 | { |
405 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
406 | /* |
407 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
408 | * avoid this_cpu_add()'s irq-disable overhead. | |
409 | */ | |
410 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
411 | #endif |
412 | } | |
413 | ||
6f3dd2c3 VB |
414 | static inline |
415 | void stat_add(const struct kmem_cache *s, enum stat_item si, int v) | |
416 | { | |
417 | #ifdef CONFIG_SLUB_STATS | |
418 | raw_cpu_add(s->cpu_slab->stat[si], v); | |
419 | #endif | |
420 | } | |
421 | ||
b52ef56e VB |
422 | /* |
423 | * The slab lists for all objects. | |
424 | */ | |
425 | struct kmem_cache_node { | |
426 | spinlock_t list_lock; | |
427 | unsigned long nr_partial; | |
428 | struct list_head partial; | |
429 | #ifdef CONFIG_SLUB_DEBUG | |
430 | atomic_long_t nr_slabs; | |
431 | atomic_long_t total_objects; | |
432 | struct list_head full; | |
433 | #endif | |
434 | }; | |
435 | ||
436 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
437 | { | |
438 | return s->node[node]; | |
439 | } | |
440 | ||
441 | /* | |
442 | * Iterator over all nodes. The body will be executed for each node that has | |
443 | * a kmem_cache_node structure allocated (which is true for all online nodes) | |
444 | */ | |
445 | #define for_each_kmem_cache_node(__s, __node, __n) \ | |
446 | for (__node = 0; __node < nr_node_ids; __node++) \ | |
447 | if ((__n = get_node(__s, __node))) | |
448 | ||
7e1fa93d VB |
449 | /* |
450 | * Tracks for which NUMA nodes we have kmem_cache_nodes allocated. | |
451 | * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily | |
452 | * differ during memory hotplug/hotremove operations. | |
453 | * Protected by slab_mutex. | |
454 | */ | |
455 | static nodemask_t slab_nodes; | |
456 | ||
0af8489b | 457 | #ifndef CONFIG_SLUB_TINY |
e45cc288 ML |
458 | /* |
459 | * Workqueue used for flush_cpu_slab(). | |
460 | */ | |
461 | static struct workqueue_struct *flushwq; | |
0af8489b | 462 | #endif |
e45cc288 | 463 | |
81819f0f CL |
464 | /******************************************************************** |
465 | * Core slab cache functions | |
466 | *******************************************************************/ | |
467 | ||
44f6a42d JH |
468 | /* |
469 | * freeptr_t represents a SLUB freelist pointer, which might be encoded | |
470 | * and not dereferenceable if CONFIG_SLAB_FREELIST_HARDENED is enabled. | |
471 | */ | |
472 | typedef struct { unsigned long v; } freeptr_t; | |
473 | ||
2482ddec KC |
474 | /* |
475 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
476 | * with an XOR of the address where the pointer is held and a per-cache | |
477 | * random number. | |
478 | */ | |
44f6a42d JH |
479 | static inline freeptr_t freelist_ptr_encode(const struct kmem_cache *s, |
480 | void *ptr, unsigned long ptr_addr) | |
2482ddec | 481 | { |
b06952cd VB |
482 | unsigned long encoded; |
483 | ||
2482ddec | 484 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
b06952cd | 485 | encoded = (unsigned long)ptr ^ s->random ^ swab(ptr_addr); |
44f6a42d | 486 | #else |
b06952cd | 487 | encoded = (unsigned long)ptr; |
44f6a42d | 488 | #endif |
b06952cd | 489 | return (freeptr_t){.v = encoded}; |
44f6a42d JH |
490 | } |
491 | ||
492 | static inline void *freelist_ptr_decode(const struct kmem_cache *s, | |
493 | freeptr_t ptr, unsigned long ptr_addr) | |
494 | { | |
495 | void *decoded; | |
496 | ||
497 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
b06952cd | 498 | decoded = (void *)(ptr.v ^ s->random ^ swab(ptr_addr)); |
2482ddec | 499 | #else |
44f6a42d | 500 | decoded = (void *)ptr.v; |
2482ddec | 501 | #endif |
44f6a42d | 502 | return decoded; |
2482ddec KC |
503 | } |
504 | ||
7656c72b CL |
505 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
506 | { | |
1662b6c2 VB |
507 | unsigned long ptr_addr; |
508 | freeptr_t p; | |
509 | ||
aa1ef4d7 | 510 | object = kasan_reset_tag(object); |
1662b6c2 VB |
511 | ptr_addr = (unsigned long)object + s->offset; |
512 | p = *(freeptr_t *)(ptr_addr); | |
513 | return freelist_ptr_decode(s, p, ptr_addr); | |
7656c72b CL |
514 | } |
515 | ||
0af8489b | 516 | #ifndef CONFIG_SLUB_TINY |
0ad9500e ED |
517 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
518 | { | |
04b4b006 | 519 | prefetchw(object + s->offset); |
0ad9500e | 520 | } |
0af8489b | 521 | #endif |
0ad9500e | 522 | |
68ef169a AP |
523 | /* |
524 | * When running under KMSAN, get_freepointer_safe() may return an uninitialized | |
525 | * pointer value in the case the current thread loses the race for the next | |
526 | * memory chunk in the freelist. In that case this_cpu_cmpxchg_double() in | |
527 | * slab_alloc_node() will fail, so the uninitialized value won't be used, but | |
528 | * KMSAN will still check all arguments of cmpxchg because of imperfect | |
529 | * handling of inline assembly. | |
530 | * To work around this problem, we apply __no_kmsan_checks to ensure that | |
531 | * get_freepointer_safe() returns initialized memory. | |
532 | */ | |
533 | __no_kmsan_checks | |
1393d9a1 CL |
534 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
535 | { | |
2482ddec | 536 | unsigned long freepointer_addr; |
44f6a42d | 537 | freeptr_t p; |
1393d9a1 | 538 | |
8e57f8ac | 539 | if (!debug_pagealloc_enabled_static()) |
922d566c JK |
540 | return get_freepointer(s, object); |
541 | ||
f70b0049 | 542 | object = kasan_reset_tag(object); |
2482ddec | 543 | freepointer_addr = (unsigned long)object + s->offset; |
44f6a42d JH |
544 | copy_from_kernel_nofault(&p, (freeptr_t *)freepointer_addr, sizeof(p)); |
545 | return freelist_ptr_decode(s, p, freepointer_addr); | |
1393d9a1 CL |
546 | } |
547 | ||
7656c72b CL |
548 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
549 | { | |
2482ddec KC |
550 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
551 | ||
ce6fa91b AP |
552 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
553 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
554 | #endif | |
555 | ||
aa1ef4d7 | 556 | freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr); |
44f6a42d | 557 | *(freeptr_t *)freeptr_addr = freelist_ptr_encode(s, fp, freeptr_addr); |
7656c72b CL |
558 | } |
559 | ||
8f828aa4 NB |
560 | /* |
561 | * See comment in calculate_sizes(). | |
562 | */ | |
563 | static inline bool freeptr_outside_object(struct kmem_cache *s) | |
564 | { | |
565 | return s->offset >= s->inuse; | |
566 | } | |
567 | ||
568 | /* | |
569 | * Return offset of the end of info block which is inuse + free pointer if | |
570 | * not overlapping with object. | |
571 | */ | |
572 | static inline unsigned int get_info_end(struct kmem_cache *s) | |
573 | { | |
574 | if (freeptr_outside_object(s)) | |
575 | return s->inuse + sizeof(void *); | |
576 | else | |
577 | return s->inuse; | |
578 | } | |
579 | ||
7656c72b | 580 | /* Loop over all objects in a slab */ |
224a88be | 581 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
582 | for (__p = fixup_red_left(__s, __addr); \ |
583 | __p < (__addr) + (__objects) * (__s)->size; \ | |
584 | __p += (__s)->size) | |
7656c72b | 585 | |
9736d2a9 | 586 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 587 | { |
9736d2a9 | 588 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
589 | } |
590 | ||
19af27af | 591 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 592 | unsigned int size) |
834f3d11 CL |
593 | { |
594 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 595 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
596 | }; |
597 | ||
598 | return x; | |
599 | } | |
600 | ||
19af27af | 601 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 602 | { |
210b5c06 | 603 | return x.x >> OO_SHIFT; |
834f3d11 CL |
604 | } |
605 | ||
19af27af | 606 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 607 | { |
210b5c06 | 608 | return x.x & OO_MASK; |
834f3d11 CL |
609 | } |
610 | ||
b47291ef VB |
611 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
612 | static void slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects) | |
613 | { | |
bb192ed9 | 614 | unsigned int nr_slabs; |
b47291ef VB |
615 | |
616 | s->cpu_partial = nr_objects; | |
617 | ||
618 | /* | |
619 | * We take the number of objects but actually limit the number of | |
c2092c12 VB |
620 | * slabs on the per cpu partial list, in order to limit excessive |
621 | * growth of the list. For simplicity we assume that the slabs will | |
b47291ef VB |
622 | * be half-full. |
623 | */ | |
bb192ed9 VB |
624 | nr_slabs = DIV_ROUND_UP(nr_objects * 2, oo_objects(s->oo)); |
625 | s->cpu_partial_slabs = nr_slabs; | |
b47291ef | 626 | } |
721a2f8b XS |
627 | |
628 | static inline unsigned int slub_get_cpu_partial(struct kmem_cache *s) | |
629 | { | |
630 | return s->cpu_partial_slabs; | |
631 | } | |
b47291ef VB |
632 | #else |
633 | static inline void | |
634 | slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects) | |
635 | { | |
636 | } | |
721a2f8b XS |
637 | |
638 | static inline unsigned int slub_get_cpu_partial(struct kmem_cache *s) | |
639 | { | |
640 | return 0; | |
641 | } | |
b47291ef VB |
642 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
643 | ||
881db7fb CL |
644 | /* |
645 | * Per slab locking using the pagelock | |
646 | */ | |
5875e598 | 647 | static __always_inline void slab_lock(struct slab *slab) |
881db7fb | 648 | { |
5e0debe0 | 649 | bit_spin_lock(PG_locked, &slab->__page_flags); |
881db7fb CL |
650 | } |
651 | ||
5875e598 | 652 | static __always_inline void slab_unlock(struct slab *slab) |
881db7fb | 653 | { |
5e0debe0 | 654 | bit_spin_unlock(PG_locked, &slab->__page_flags); |
881db7fb CL |
655 | } |
656 | ||
6801be4f PZ |
657 | static inline bool |
658 | __update_freelist_fast(struct slab *slab, | |
659 | void *freelist_old, unsigned long counters_old, | |
660 | void *freelist_new, unsigned long counters_new) | |
661 | { | |
662 | #ifdef system_has_freelist_aba | |
663 | freelist_aba_t old = { .freelist = freelist_old, .counter = counters_old }; | |
664 | freelist_aba_t new = { .freelist = freelist_new, .counter = counters_new }; | |
665 | ||
666 | return try_cmpxchg_freelist(&slab->freelist_counter.full, &old.full, new.full); | |
667 | #else | |
668 | return false; | |
669 | #endif | |
670 | } | |
671 | ||
672 | static inline bool | |
673 | __update_freelist_slow(struct slab *slab, | |
674 | void *freelist_old, unsigned long counters_old, | |
675 | void *freelist_new, unsigned long counters_new) | |
676 | { | |
677 | bool ret = false; | |
678 | ||
679 | slab_lock(slab); | |
680 | if (slab->freelist == freelist_old && | |
681 | slab->counters == counters_old) { | |
682 | slab->freelist = freelist_new; | |
683 | slab->counters = counters_new; | |
684 | ret = true; | |
685 | } | |
686 | slab_unlock(slab); | |
687 | ||
688 | return ret; | |
689 | } | |
690 | ||
a2b4ae8b VB |
691 | /* |
692 | * Interrupts must be disabled (for the fallback code to work right), typically | |
5875e598 VB |
693 | * by an _irqsave() lock variant. On PREEMPT_RT the preempt_disable(), which is |
694 | * part of bit_spin_lock(), is sufficient because the policy is not to allow any | |
695 | * allocation/ free operation in hardirq context. Therefore nothing can | |
696 | * interrupt the operation. | |
a2b4ae8b | 697 | */ |
6801be4f | 698 | static inline bool __slab_update_freelist(struct kmem_cache *s, struct slab *slab, |
1d07171c CL |
699 | void *freelist_old, unsigned long counters_old, |
700 | void *freelist_new, unsigned long counters_new, | |
701 | const char *n) | |
702 | { | |
6801be4f PZ |
703 | bool ret; |
704 | ||
1f04b07d | 705 | if (USE_LOCKLESS_FAST_PATH()) |
a2b4ae8b | 706 | lockdep_assert_irqs_disabled(); |
6801be4f | 707 | |
1d07171c | 708 | if (s->flags & __CMPXCHG_DOUBLE) { |
6801be4f PZ |
709 | ret = __update_freelist_fast(slab, freelist_old, counters_old, |
710 | freelist_new, counters_new); | |
711 | } else { | |
712 | ret = __update_freelist_slow(slab, freelist_old, counters_old, | |
713 | freelist_new, counters_new); | |
1d07171c | 714 | } |
6801be4f PZ |
715 | if (likely(ret)) |
716 | return true; | |
1d07171c CL |
717 | |
718 | cpu_relax(); | |
719 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
720 | ||
721 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 722 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
723 | #endif |
724 | ||
6f6528a1 | 725 | return false; |
1d07171c CL |
726 | } |
727 | ||
6801be4f | 728 | static inline bool slab_update_freelist(struct kmem_cache *s, struct slab *slab, |
b789ef51 CL |
729 | void *freelist_old, unsigned long counters_old, |
730 | void *freelist_new, unsigned long counters_new, | |
731 | const char *n) | |
732 | { | |
6801be4f PZ |
733 | bool ret; |
734 | ||
b789ef51 | 735 | if (s->flags & __CMPXCHG_DOUBLE) { |
6801be4f PZ |
736 | ret = __update_freelist_fast(slab, freelist_old, counters_old, |
737 | freelist_new, counters_new); | |
738 | } else { | |
1d07171c CL |
739 | unsigned long flags; |
740 | ||
741 | local_irq_save(flags); | |
6801be4f PZ |
742 | ret = __update_freelist_slow(slab, freelist_old, counters_old, |
743 | freelist_new, counters_new); | |
1d07171c | 744 | local_irq_restore(flags); |
b789ef51 | 745 | } |
6801be4f PZ |
746 | if (likely(ret)) |
747 | return true; | |
b789ef51 CL |
748 | |
749 | cpu_relax(); | |
750 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
751 | ||
752 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 753 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
754 | #endif |
755 | ||
6f6528a1 | 756 | return false; |
b789ef51 CL |
757 | } |
758 | ||
41ecc55b | 759 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 | 760 | static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)]; |
4ef3f5a3 | 761 | static DEFINE_SPINLOCK(object_map_lock); |
90e9f6a6 | 762 | |
b3fd64e1 | 763 | static void __fill_map(unsigned long *obj_map, struct kmem_cache *s, |
bb192ed9 | 764 | struct slab *slab) |
b3fd64e1 | 765 | { |
bb192ed9 | 766 | void *addr = slab_address(slab); |
b3fd64e1 VB |
767 | void *p; |
768 | ||
bb192ed9 | 769 | bitmap_zero(obj_map, slab->objects); |
b3fd64e1 | 770 | |
bb192ed9 | 771 | for (p = slab->freelist; p; p = get_freepointer(s, p)) |
b3fd64e1 VB |
772 | set_bit(__obj_to_index(s, addr, p), obj_map); |
773 | } | |
774 | ||
1f9f78b1 OG |
775 | #if IS_ENABLED(CONFIG_KUNIT) |
776 | static bool slab_add_kunit_errors(void) | |
777 | { | |
778 | struct kunit_resource *resource; | |
779 | ||
909c6475 | 780 | if (!kunit_get_current_test()) |
1f9f78b1 OG |
781 | return false; |
782 | ||
783 | resource = kunit_find_named_resource(current->kunit_test, "slab_errors"); | |
784 | if (!resource) | |
785 | return false; | |
786 | ||
787 | (*(int *)resource->data)++; | |
788 | kunit_put_resource(resource); | |
789 | return true; | |
790 | } | |
791 | #else | |
792 | static inline bool slab_add_kunit_errors(void) { return false; } | |
793 | #endif | |
794 | ||
870b1fbb | 795 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
796 | { |
797 | if (s->flags & SLAB_RED_ZONE) | |
798 | return s->size - s->red_left_pad; | |
799 | ||
800 | return s->size; | |
801 | } | |
802 | ||
803 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
804 | { | |
805 | if (s->flags & SLAB_RED_ZONE) | |
806 | p -= s->red_left_pad; | |
807 | ||
808 | return p; | |
809 | } | |
810 | ||
41ecc55b CL |
811 | /* |
812 | * Debug settings: | |
813 | */ | |
89d3c87e | 814 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 815 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 816 | #else |
d50112ed | 817 | static slab_flags_t slub_debug; |
f0630fff | 818 | #endif |
41ecc55b | 819 | |
e17f1dfb | 820 | static char *slub_debug_string; |
fa5ec8a1 | 821 | static int disable_higher_order_debug; |
41ecc55b | 822 | |
a79316c6 AR |
823 | /* |
824 | * slub is about to manipulate internal object metadata. This memory lies | |
825 | * outside the range of the allocated object, so accessing it would normally | |
826 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
827 | * to tell kasan that these accesses are OK. | |
828 | */ | |
829 | static inline void metadata_access_enable(void) | |
830 | { | |
831 | kasan_disable_current(); | |
832 | } | |
833 | ||
834 | static inline void metadata_access_disable(void) | |
835 | { | |
836 | kasan_enable_current(); | |
837 | } | |
838 | ||
81819f0f CL |
839 | /* |
840 | * Object debugging | |
841 | */ | |
d86bd1be JK |
842 | |
843 | /* Verify that a pointer has an address that is valid within a slab page */ | |
844 | static inline int check_valid_pointer(struct kmem_cache *s, | |
bb192ed9 | 845 | struct slab *slab, void *object) |
d86bd1be JK |
846 | { |
847 | void *base; | |
848 | ||
849 | if (!object) | |
850 | return 1; | |
851 | ||
bb192ed9 | 852 | base = slab_address(slab); |
338cfaad | 853 | object = kasan_reset_tag(object); |
d86bd1be | 854 | object = restore_red_left(s, object); |
bb192ed9 | 855 | if (object < base || object >= base + slab->objects * s->size || |
d86bd1be JK |
856 | (object - base) % s->size) { |
857 | return 0; | |
858 | } | |
859 | ||
860 | return 1; | |
861 | } | |
862 | ||
aa2efd5e DT |
863 | static void print_section(char *level, char *text, u8 *addr, |
864 | unsigned int length) | |
81819f0f | 865 | { |
a79316c6 | 866 | metadata_access_enable(); |
340caf17 KYL |
867 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, |
868 | 16, 1, kasan_reset_tag((void *)addr), length, 1); | |
a79316c6 | 869 | metadata_access_disable(); |
81819f0f CL |
870 | } |
871 | ||
81819f0f CL |
872 | static struct track *get_track(struct kmem_cache *s, void *object, |
873 | enum track_item alloc) | |
874 | { | |
875 | struct track *p; | |
876 | ||
cbfc35a4 | 877 | p = object + get_info_end(s); |
81819f0f | 878 | |
aa1ef4d7 | 879 | return kasan_reset_tag(p + alloc); |
81819f0f CL |
880 | } |
881 | ||
5cf909c5 | 882 | #ifdef CONFIG_STACKDEPOT |
c4cf6785 SAS |
883 | static noinline depot_stack_handle_t set_track_prepare(void) |
884 | { | |
885 | depot_stack_handle_t handle; | |
5cf909c5 | 886 | unsigned long entries[TRACK_ADDRS_COUNT]; |
0cd1a029 | 887 | unsigned int nr_entries; |
ae14c63a | 888 | |
5cf909c5 | 889 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3); |
c4cf6785 SAS |
890 | handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT); |
891 | ||
892 | return handle; | |
893 | } | |
894 | #else | |
895 | static inline depot_stack_handle_t set_track_prepare(void) | |
896 | { | |
897 | return 0; | |
898 | } | |
d6543e39 | 899 | #endif |
5cf909c5 | 900 | |
c4cf6785 SAS |
901 | static void set_track_update(struct kmem_cache *s, void *object, |
902 | enum track_item alloc, unsigned long addr, | |
903 | depot_stack_handle_t handle) | |
904 | { | |
905 | struct track *p = get_track(s, object, alloc); | |
906 | ||
907 | #ifdef CONFIG_STACKDEPOT | |
908 | p->handle = handle; | |
909 | #endif | |
0cd1a029 VB |
910 | p->addr = addr; |
911 | p->cpu = smp_processor_id(); | |
912 | p->pid = current->pid; | |
913 | p->when = jiffies; | |
81819f0f CL |
914 | } |
915 | ||
c4cf6785 SAS |
916 | static __always_inline void set_track(struct kmem_cache *s, void *object, |
917 | enum track_item alloc, unsigned long addr) | |
918 | { | |
919 | depot_stack_handle_t handle = set_track_prepare(); | |
920 | ||
921 | set_track_update(s, object, alloc, addr, handle); | |
922 | } | |
923 | ||
81819f0f CL |
924 | static void init_tracking(struct kmem_cache *s, void *object) |
925 | { | |
0cd1a029 VB |
926 | struct track *p; |
927 | ||
24922684 CL |
928 | if (!(s->flags & SLAB_STORE_USER)) |
929 | return; | |
930 | ||
0cd1a029 VB |
931 | p = get_track(s, object, TRACK_ALLOC); |
932 | memset(p, 0, 2*sizeof(struct track)); | |
81819f0f CL |
933 | } |
934 | ||
86609d33 | 935 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f | 936 | { |
5cf909c5 OG |
937 | depot_stack_handle_t handle __maybe_unused; |
938 | ||
81819f0f CL |
939 | if (!t->addr) |
940 | return; | |
941 | ||
96b94abc | 942 | pr_err("%s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 943 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
5cf909c5 OG |
944 | #ifdef CONFIG_STACKDEPOT |
945 | handle = READ_ONCE(t->handle); | |
946 | if (handle) | |
947 | stack_depot_print(handle); | |
948 | else | |
949 | pr_err("object allocation/free stack trace missing\n"); | |
d6543e39 | 950 | #endif |
24922684 CL |
951 | } |
952 | ||
e42f174e | 953 | void print_tracking(struct kmem_cache *s, void *object) |
24922684 | 954 | { |
86609d33 | 955 | unsigned long pr_time = jiffies; |
24922684 CL |
956 | if (!(s->flags & SLAB_STORE_USER)) |
957 | return; | |
958 | ||
86609d33 CP |
959 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
960 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
961 | } |
962 | ||
fb012e27 | 963 | static void print_slab_info(const struct slab *slab) |
24922684 | 964 | { |
fb012e27 | 965 | struct folio *folio = (struct folio *)slab_folio(slab); |
24922684 | 966 | |
fb012e27 MWO |
967 | pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%pGp\n", |
968 | slab, slab->objects, slab->inuse, slab->freelist, | |
969 | folio_flags(folio, 0)); | |
24922684 CL |
970 | } |
971 | ||
6edf2576 FT |
972 | /* |
973 | * kmalloc caches has fixed sizes (mostly power of 2), and kmalloc() API | |
974 | * family will round up the real request size to these fixed ones, so | |
975 | * there could be an extra area than what is requested. Save the original | |
976 | * request size in the meta data area, for better debug and sanity check. | |
977 | */ | |
978 | static inline void set_orig_size(struct kmem_cache *s, | |
979 | void *object, unsigned int orig_size) | |
980 | { | |
981 | void *p = kasan_reset_tag(object); | |
2d552463 | 982 | unsigned int kasan_meta_size; |
6edf2576 FT |
983 | |
984 | if (!slub_debug_orig_size(s)) | |
985 | return; | |
986 | ||
946fa0db | 987 | /* |
2d552463 AK |
988 | * KASAN can save its free meta data inside of the object at offset 0. |
989 | * If this meta data size is larger than 'orig_size', it will overlap | |
990 | * the data redzone in [orig_size+1, object_size]. Thus, we adjust | |
991 | * 'orig_size' to be as at least as big as KASAN's meta data. | |
946fa0db | 992 | */ |
2d552463 AK |
993 | kasan_meta_size = kasan_metadata_size(s, true); |
994 | if (kasan_meta_size > orig_size) | |
995 | orig_size = kasan_meta_size; | |
946fa0db | 996 | |
6edf2576 FT |
997 | p += get_info_end(s); |
998 | p += sizeof(struct track) * 2; | |
999 | ||
1000 | *(unsigned int *)p = orig_size; | |
1001 | } | |
1002 | ||
1003 | static inline unsigned int get_orig_size(struct kmem_cache *s, void *object) | |
1004 | { | |
1005 | void *p = kasan_reset_tag(object); | |
1006 | ||
1007 | if (!slub_debug_orig_size(s)) | |
1008 | return s->object_size; | |
1009 | ||
1010 | p += get_info_end(s); | |
1011 | p += sizeof(struct track) * 2; | |
1012 | ||
1013 | return *(unsigned int *)p; | |
1014 | } | |
1015 | ||
946fa0db FT |
1016 | void skip_orig_size_check(struct kmem_cache *s, const void *object) |
1017 | { | |
1018 | set_orig_size(s, (void *)object, s->object_size); | |
1019 | } | |
1020 | ||
24922684 CL |
1021 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) |
1022 | { | |
ecc42fbe | 1023 | struct va_format vaf; |
24922684 | 1024 | va_list args; |
24922684 CL |
1025 | |
1026 | va_start(args, fmt); | |
ecc42fbe FF |
1027 | vaf.fmt = fmt; |
1028 | vaf.va = &args; | |
f9f58285 | 1029 | pr_err("=============================================================================\n"); |
ecc42fbe | 1030 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 1031 | pr_err("-----------------------------------------------------------------------------\n\n"); |
ecc42fbe | 1032 | va_end(args); |
81819f0f CL |
1033 | } |
1034 | ||
582d1212 | 1035 | __printf(2, 3) |
24922684 CL |
1036 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
1037 | { | |
ecc42fbe | 1038 | struct va_format vaf; |
24922684 | 1039 | va_list args; |
24922684 | 1040 | |
1f9f78b1 OG |
1041 | if (slab_add_kunit_errors()) |
1042 | return; | |
1043 | ||
24922684 | 1044 | va_start(args, fmt); |
ecc42fbe FF |
1045 | vaf.fmt = fmt; |
1046 | vaf.va = &args; | |
1047 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 1048 | va_end(args); |
24922684 CL |
1049 | } |
1050 | ||
bb192ed9 | 1051 | static void print_trailer(struct kmem_cache *s, struct slab *slab, u8 *p) |
81819f0f CL |
1052 | { |
1053 | unsigned int off; /* Offset of last byte */ | |
bb192ed9 | 1054 | u8 *addr = slab_address(slab); |
24922684 CL |
1055 | |
1056 | print_tracking(s, p); | |
1057 | ||
bb192ed9 | 1058 | print_slab_info(slab); |
24922684 | 1059 | |
96b94abc | 1060 | pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n", |
f9f58285 | 1061 | p, p - addr, get_freepointer(s, p)); |
24922684 | 1062 | |
d86bd1be | 1063 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 1064 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
aa2efd5e | 1065 | s->red_left_pad); |
d86bd1be | 1066 | else if (p > addr + 16) |
aa2efd5e | 1067 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 1068 | |
8669dbab | 1069 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 1070 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 1071 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 1072 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 1073 | s->inuse - s->object_size); |
81819f0f | 1074 | |
cbfc35a4 | 1075 | off = get_info_end(s); |
81819f0f | 1076 | |
24922684 | 1077 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 1078 | off += 2 * sizeof(struct track); |
81819f0f | 1079 | |
6edf2576 FT |
1080 | if (slub_debug_orig_size(s)) |
1081 | off += sizeof(unsigned int); | |
1082 | ||
5d1ba310 | 1083 | off += kasan_metadata_size(s, false); |
80a9201a | 1084 | |
d86bd1be | 1085 | if (off != size_from_object(s)) |
81819f0f | 1086 | /* Beginning of the filler is the free pointer */ |
8669dbab | 1087 | print_section(KERN_ERR, "Padding ", p + off, |
aa2efd5e | 1088 | size_from_object(s) - off); |
24922684 CL |
1089 | |
1090 | dump_stack(); | |
81819f0f CL |
1091 | } |
1092 | ||
bb192ed9 | 1093 | static void object_err(struct kmem_cache *s, struct slab *slab, |
81819f0f CL |
1094 | u8 *object, char *reason) |
1095 | { | |
1f9f78b1 OG |
1096 | if (slab_add_kunit_errors()) |
1097 | return; | |
1098 | ||
3dc50637 | 1099 | slab_bug(s, "%s", reason); |
bb192ed9 | 1100 | print_trailer(s, slab, object); |
65ebdeef | 1101 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
1102 | } |
1103 | ||
bb192ed9 | 1104 | static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab, |
ae16d059 VB |
1105 | void **freelist, void *nextfree) |
1106 | { | |
1107 | if ((s->flags & SLAB_CONSISTENCY_CHECKS) && | |
bb192ed9 VB |
1108 | !check_valid_pointer(s, slab, nextfree) && freelist) { |
1109 | object_err(s, slab, *freelist, "Freechain corrupt"); | |
ae16d059 VB |
1110 | *freelist = NULL; |
1111 | slab_fix(s, "Isolate corrupted freechain"); | |
1112 | return true; | |
1113 | } | |
1114 | ||
1115 | return false; | |
1116 | } | |
1117 | ||
bb192ed9 | 1118 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct slab *slab, |
d0e0ac97 | 1119 | const char *fmt, ...) |
81819f0f CL |
1120 | { |
1121 | va_list args; | |
1122 | char buf[100]; | |
1123 | ||
1f9f78b1 OG |
1124 | if (slab_add_kunit_errors()) |
1125 | return; | |
1126 | ||
24922684 CL |
1127 | va_start(args, fmt); |
1128 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 1129 | va_end(args); |
3dc50637 | 1130 | slab_bug(s, "%s", buf); |
bb192ed9 | 1131 | print_slab_info(slab); |
81819f0f | 1132 | dump_stack(); |
65ebdeef | 1133 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
1134 | } |
1135 | ||
f7cb1933 | 1136 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f | 1137 | { |
aa1ef4d7 | 1138 | u8 *p = kasan_reset_tag(object); |
946fa0db | 1139 | unsigned int poison_size = s->object_size; |
81819f0f | 1140 | |
946fa0db | 1141 | if (s->flags & SLAB_RED_ZONE) { |
d86bd1be JK |
1142 | memset(p - s->red_left_pad, val, s->red_left_pad); |
1143 | ||
946fa0db FT |
1144 | if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) { |
1145 | /* | |
1146 | * Redzone the extra allocated space by kmalloc than | |
1147 | * requested, and the poison size will be limited to | |
1148 | * the original request size accordingly. | |
1149 | */ | |
1150 | poison_size = get_orig_size(s, object); | |
1151 | } | |
1152 | } | |
1153 | ||
81819f0f | 1154 | if (s->flags & __OBJECT_POISON) { |
946fa0db FT |
1155 | memset(p, POISON_FREE, poison_size - 1); |
1156 | p[poison_size - 1] = POISON_END; | |
81819f0f CL |
1157 | } |
1158 | ||
1159 | if (s->flags & SLAB_RED_ZONE) | |
946fa0db | 1160 | memset(p + poison_size, val, s->inuse - poison_size); |
81819f0f CL |
1161 | } |
1162 | ||
24922684 CL |
1163 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
1164 | void *from, void *to) | |
1165 | { | |
582d1212 | 1166 | slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data); |
24922684 CL |
1167 | memset(from, data, to - from); |
1168 | } | |
1169 | ||
bb192ed9 | 1170 | static int check_bytes_and_report(struct kmem_cache *s, struct slab *slab, |
24922684 | 1171 | u8 *object, char *what, |
06428780 | 1172 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
1173 | { |
1174 | u8 *fault; | |
1175 | u8 *end; | |
bb192ed9 | 1176 | u8 *addr = slab_address(slab); |
24922684 | 1177 | |
a79316c6 | 1178 | metadata_access_enable(); |
aa1ef4d7 | 1179 | fault = memchr_inv(kasan_reset_tag(start), value, bytes); |
a79316c6 | 1180 | metadata_access_disable(); |
24922684 CL |
1181 | if (!fault) |
1182 | return 1; | |
1183 | ||
1184 | end = start + bytes; | |
1185 | while (end > fault && end[-1] == value) | |
1186 | end--; | |
1187 | ||
1f9f78b1 OG |
1188 | if (slab_add_kunit_errors()) |
1189 | goto skip_bug_print; | |
1190 | ||
24922684 | 1191 | slab_bug(s, "%s overwritten", what); |
96b94abc | 1192 | pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", |
e1b70dd1 MC |
1193 | fault, end - 1, fault - addr, |
1194 | fault[0], value); | |
bb192ed9 | 1195 | print_trailer(s, slab, object); |
65ebdeef | 1196 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
24922684 | 1197 | |
1f9f78b1 | 1198 | skip_bug_print: |
24922684 CL |
1199 | restore_bytes(s, what, value, fault, end); |
1200 | return 0; | |
81819f0f CL |
1201 | } |
1202 | ||
81819f0f CL |
1203 | /* |
1204 | * Object layout: | |
1205 | * | |
1206 | * object address | |
1207 | * Bytes of the object to be managed. | |
1208 | * If the freepointer may overlay the object then the free | |
cbfc35a4 | 1209 | * pointer is at the middle of the object. |
672bba3a | 1210 | * |
81819f0f CL |
1211 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
1212 | * 0xa5 (POISON_END) | |
1213 | * | |
3b0efdfa | 1214 | * object + s->object_size |
81819f0f | 1215 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 1216 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 1217 | * object_size == inuse. |
672bba3a | 1218 | * |
81819f0f CL |
1219 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
1220 | * 0xcc (RED_ACTIVE) for objects in use. | |
1221 | * | |
1222 | * object + s->inuse | |
672bba3a CL |
1223 | * Meta data starts here. |
1224 | * | |
81819f0f CL |
1225 | * A. Free pointer (if we cannot overwrite object on free) |
1226 | * B. Tracking data for SLAB_STORE_USER | |
6edf2576 FT |
1227 | * C. Original request size for kmalloc object (SLAB_STORE_USER enabled) |
1228 | * D. Padding to reach required alignment boundary or at minimum | |
6446faa2 | 1229 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
1230 | * before the word boundary. |
1231 | * | |
1232 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
1233 | * |
1234 | * object + s->size | |
672bba3a | 1235 | * Nothing is used beyond s->size. |
81819f0f | 1236 | * |
3b0efdfa | 1237 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 1238 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
1239 | * may be used with merged slabcaches. |
1240 | */ | |
1241 | ||
bb192ed9 | 1242 | static int check_pad_bytes(struct kmem_cache *s, struct slab *slab, u8 *p) |
81819f0f | 1243 | { |
cbfc35a4 | 1244 | unsigned long off = get_info_end(s); /* The end of info */ |
81819f0f | 1245 | |
6edf2576 | 1246 | if (s->flags & SLAB_STORE_USER) { |
81819f0f CL |
1247 | /* We also have user information there */ |
1248 | off += 2 * sizeof(struct track); | |
1249 | ||
6edf2576 FT |
1250 | if (s->flags & SLAB_KMALLOC) |
1251 | off += sizeof(unsigned int); | |
1252 | } | |
1253 | ||
5d1ba310 | 1254 | off += kasan_metadata_size(s, false); |
80a9201a | 1255 | |
d86bd1be | 1256 | if (size_from_object(s) == off) |
81819f0f CL |
1257 | return 1; |
1258 | ||
bb192ed9 | 1259 | return check_bytes_and_report(s, slab, p, "Object padding", |
d86bd1be | 1260 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
1261 | } |
1262 | ||
39b26464 | 1263 | /* Check the pad bytes at the end of a slab page */ |
a204e6d6 | 1264 | static void slab_pad_check(struct kmem_cache *s, struct slab *slab) |
81819f0f | 1265 | { |
24922684 CL |
1266 | u8 *start; |
1267 | u8 *fault; | |
1268 | u8 *end; | |
5d682681 | 1269 | u8 *pad; |
24922684 CL |
1270 | int length; |
1271 | int remainder; | |
81819f0f CL |
1272 | |
1273 | if (!(s->flags & SLAB_POISON)) | |
a204e6d6 | 1274 | return; |
81819f0f | 1275 | |
bb192ed9 VB |
1276 | start = slab_address(slab); |
1277 | length = slab_size(slab); | |
39b26464 CL |
1278 | end = start + length; |
1279 | remainder = length % s->size; | |
81819f0f | 1280 | if (!remainder) |
a204e6d6 | 1281 | return; |
81819f0f | 1282 | |
5d682681 | 1283 | pad = end - remainder; |
a79316c6 | 1284 | metadata_access_enable(); |
aa1ef4d7 | 1285 | fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder); |
a79316c6 | 1286 | metadata_access_disable(); |
24922684 | 1287 | if (!fault) |
a204e6d6 | 1288 | return; |
24922684 CL |
1289 | while (end > fault && end[-1] == POISON_INUSE) |
1290 | end--; | |
1291 | ||
bb192ed9 | 1292 | slab_err(s, slab, "Padding overwritten. 0x%p-0x%p @offset=%tu", |
e1b70dd1 | 1293 | fault, end - 1, fault - start); |
5d682681 | 1294 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 1295 | |
5d682681 | 1296 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
81819f0f CL |
1297 | } |
1298 | ||
bb192ed9 | 1299 | static int check_object(struct kmem_cache *s, struct slab *slab, |
f7cb1933 | 1300 | void *object, u8 val) |
81819f0f CL |
1301 | { |
1302 | u8 *p = object; | |
3b0efdfa | 1303 | u8 *endobject = object + s->object_size; |
2d552463 | 1304 | unsigned int orig_size, kasan_meta_size; |
81819f0f CL |
1305 | |
1306 | if (s->flags & SLAB_RED_ZONE) { | |
bb192ed9 | 1307 | if (!check_bytes_and_report(s, slab, object, "Left Redzone", |
d86bd1be JK |
1308 | object - s->red_left_pad, val, s->red_left_pad)) |
1309 | return 0; | |
1310 | ||
bb192ed9 | 1311 | if (!check_bytes_and_report(s, slab, object, "Right Redzone", |
3b0efdfa | 1312 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 1313 | return 0; |
946fa0db FT |
1314 | |
1315 | if (slub_debug_orig_size(s) && val == SLUB_RED_ACTIVE) { | |
1316 | orig_size = get_orig_size(s, object); | |
1317 | ||
1318 | if (s->object_size > orig_size && | |
1319 | !check_bytes_and_report(s, slab, object, | |
1320 | "kmalloc Redzone", p + orig_size, | |
1321 | val, s->object_size - orig_size)) { | |
1322 | return 0; | |
1323 | } | |
1324 | } | |
81819f0f | 1325 | } else { |
3b0efdfa | 1326 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
bb192ed9 | 1327 | check_bytes_and_report(s, slab, p, "Alignment padding", |
d0e0ac97 CG |
1328 | endobject, POISON_INUSE, |
1329 | s->inuse - s->object_size); | |
3adbefee | 1330 | } |
81819f0f CL |
1331 | } |
1332 | ||
1333 | if (s->flags & SLAB_POISON) { | |
2d552463 AK |
1334 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON)) { |
1335 | /* | |
1336 | * KASAN can save its free meta data inside of the | |
1337 | * object at offset 0. Thus, skip checking the part of | |
1338 | * the redzone that overlaps with the meta data. | |
1339 | */ | |
1340 | kasan_meta_size = kasan_metadata_size(s, true); | |
1341 | if (kasan_meta_size < s->object_size - 1 && | |
1342 | !check_bytes_and_report(s, slab, p, "Poison", | |
1343 | p + kasan_meta_size, POISON_FREE, | |
1344 | s->object_size - kasan_meta_size - 1)) | |
1345 | return 0; | |
1346 | if (kasan_meta_size < s->object_size && | |
1347 | !check_bytes_and_report(s, slab, p, "End Poison", | |
1348 | p + s->object_size - 1, POISON_END, 1)) | |
1349 | return 0; | |
1350 | } | |
81819f0f CL |
1351 | /* |
1352 | * check_pad_bytes cleans up on its own. | |
1353 | */ | |
bb192ed9 | 1354 | check_pad_bytes(s, slab, p); |
81819f0f CL |
1355 | } |
1356 | ||
cbfc35a4 | 1357 | if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE) |
81819f0f CL |
1358 | /* |
1359 | * Object and freepointer overlap. Cannot check | |
1360 | * freepointer while object is allocated. | |
1361 | */ | |
1362 | return 1; | |
1363 | ||
1364 | /* Check free pointer validity */ | |
bb192ed9 VB |
1365 | if (!check_valid_pointer(s, slab, get_freepointer(s, p))) { |
1366 | object_err(s, slab, p, "Freepointer corrupt"); | |
81819f0f | 1367 | /* |
9f6c708e | 1368 | * No choice but to zap it and thus lose the remainder |
81819f0f | 1369 | * of the free objects in this slab. May cause |
672bba3a | 1370 | * another error because the object count is now wrong. |
81819f0f | 1371 | */ |
a973e9dd | 1372 | set_freepointer(s, p, NULL); |
81819f0f CL |
1373 | return 0; |
1374 | } | |
1375 | return 1; | |
1376 | } | |
1377 | ||
bb192ed9 | 1378 | static int check_slab(struct kmem_cache *s, struct slab *slab) |
81819f0f | 1379 | { |
39b26464 CL |
1380 | int maxobj; |
1381 | ||
bb192ed9 VB |
1382 | if (!folio_test_slab(slab_folio(slab))) { |
1383 | slab_err(s, slab, "Not a valid slab page"); | |
81819f0f CL |
1384 | return 0; |
1385 | } | |
39b26464 | 1386 | |
bb192ed9 VB |
1387 | maxobj = order_objects(slab_order(slab), s->size); |
1388 | if (slab->objects > maxobj) { | |
1389 | slab_err(s, slab, "objects %u > max %u", | |
1390 | slab->objects, maxobj); | |
39b26464 CL |
1391 | return 0; |
1392 | } | |
bb192ed9 VB |
1393 | if (slab->inuse > slab->objects) { |
1394 | slab_err(s, slab, "inuse %u > max %u", | |
1395 | slab->inuse, slab->objects); | |
81819f0f CL |
1396 | return 0; |
1397 | } | |
1398 | /* Slab_pad_check fixes things up after itself */ | |
bb192ed9 | 1399 | slab_pad_check(s, slab); |
81819f0f CL |
1400 | return 1; |
1401 | } | |
1402 | ||
1403 | /* | |
c2092c12 | 1404 | * Determine if a certain object in a slab is on the freelist. Must hold the |
672bba3a | 1405 | * slab lock to guarantee that the chains are in a consistent state. |
81819f0f | 1406 | */ |
bb192ed9 | 1407 | static int on_freelist(struct kmem_cache *s, struct slab *slab, void *search) |
81819f0f CL |
1408 | { |
1409 | int nr = 0; | |
881db7fb | 1410 | void *fp; |
81819f0f | 1411 | void *object = NULL; |
f6edde9c | 1412 | int max_objects; |
81819f0f | 1413 | |
bb192ed9 VB |
1414 | fp = slab->freelist; |
1415 | while (fp && nr <= slab->objects) { | |
81819f0f CL |
1416 | if (fp == search) |
1417 | return 1; | |
bb192ed9 | 1418 | if (!check_valid_pointer(s, slab, fp)) { |
81819f0f | 1419 | if (object) { |
bb192ed9 | 1420 | object_err(s, slab, object, |
81819f0f | 1421 | "Freechain corrupt"); |
a973e9dd | 1422 | set_freepointer(s, object, NULL); |
81819f0f | 1423 | } else { |
bb192ed9 VB |
1424 | slab_err(s, slab, "Freepointer corrupt"); |
1425 | slab->freelist = NULL; | |
1426 | slab->inuse = slab->objects; | |
24922684 | 1427 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
1428 | return 0; |
1429 | } | |
1430 | break; | |
1431 | } | |
1432 | object = fp; | |
1433 | fp = get_freepointer(s, object); | |
1434 | nr++; | |
1435 | } | |
1436 | ||
bb192ed9 | 1437 | max_objects = order_objects(slab_order(slab), s->size); |
210b5c06 CG |
1438 | if (max_objects > MAX_OBJS_PER_PAGE) |
1439 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be | 1440 | |
bb192ed9 VB |
1441 | if (slab->objects != max_objects) { |
1442 | slab_err(s, slab, "Wrong number of objects. Found %d but should be %d", | |
1443 | slab->objects, max_objects); | |
1444 | slab->objects = max_objects; | |
582d1212 | 1445 | slab_fix(s, "Number of objects adjusted"); |
224a88be | 1446 | } |
bb192ed9 VB |
1447 | if (slab->inuse != slab->objects - nr) { |
1448 | slab_err(s, slab, "Wrong object count. Counter is %d but counted were %d", | |
1449 | slab->inuse, slab->objects - nr); | |
1450 | slab->inuse = slab->objects - nr; | |
582d1212 | 1451 | slab_fix(s, "Object count adjusted"); |
81819f0f CL |
1452 | } |
1453 | return search == NULL; | |
1454 | } | |
1455 | ||
bb192ed9 | 1456 | static void trace(struct kmem_cache *s, struct slab *slab, void *object, |
0121c619 | 1457 | int alloc) |
3ec09742 CL |
1458 | { |
1459 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1460 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1461 | s->name, |
1462 | alloc ? "alloc" : "free", | |
bb192ed9 VB |
1463 | object, slab->inuse, |
1464 | slab->freelist); | |
3ec09742 CL |
1465 | |
1466 | if (!alloc) | |
aa2efd5e | 1467 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1468 | s->object_size); |
3ec09742 CL |
1469 | |
1470 | dump_stack(); | |
1471 | } | |
1472 | } | |
1473 | ||
643b1138 | 1474 | /* |
672bba3a | 1475 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1476 | */ |
5cc6eee8 | 1477 | static void add_full(struct kmem_cache *s, |
bb192ed9 | 1478 | struct kmem_cache_node *n, struct slab *slab) |
643b1138 | 1479 | { |
5cc6eee8 CL |
1480 | if (!(s->flags & SLAB_STORE_USER)) |
1481 | return; | |
1482 | ||
255d0884 | 1483 | lockdep_assert_held(&n->list_lock); |
bb192ed9 | 1484 | list_add(&slab->slab_list, &n->full); |
643b1138 CL |
1485 | } |
1486 | ||
bb192ed9 | 1487 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct slab *slab) |
643b1138 | 1488 | { |
643b1138 CL |
1489 | if (!(s->flags & SLAB_STORE_USER)) |
1490 | return; | |
1491 | ||
255d0884 | 1492 | lockdep_assert_held(&n->list_lock); |
bb192ed9 | 1493 | list_del(&slab->slab_list); |
643b1138 CL |
1494 | } |
1495 | ||
26c02cf0 AB |
1496 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1497 | { | |
1498 | return atomic_long_read(&n->nr_slabs); | |
1499 | } | |
1500 | ||
205ab99d | 1501 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1502 | { |
1503 | struct kmem_cache_node *n = get_node(s, node); | |
1504 | ||
3dd549a5 CZ |
1505 | atomic_long_inc(&n->nr_slabs); |
1506 | atomic_long_add(objects, &n->total_objects); | |
0f389ec6 | 1507 | } |
205ab99d | 1508 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1509 | { |
1510 | struct kmem_cache_node *n = get_node(s, node); | |
1511 | ||
1512 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1513 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1514 | } |
1515 | ||
1516 | /* Object debug checks for alloc/free paths */ | |
c0f81a94 | 1517 | static void setup_object_debug(struct kmem_cache *s, void *object) |
3ec09742 | 1518 | { |
8fc8d666 | 1519 | if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)) |
3ec09742 CL |
1520 | return; |
1521 | ||
f7cb1933 | 1522 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1523 | init_tracking(s, object); |
1524 | } | |
1525 | ||
a50b854e | 1526 | static |
bb192ed9 | 1527 | void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr) |
a7101224 | 1528 | { |
8fc8d666 | 1529 | if (!kmem_cache_debug_flags(s, SLAB_POISON)) |
a7101224 AK |
1530 | return; |
1531 | ||
1532 | metadata_access_enable(); | |
bb192ed9 | 1533 | memset(kasan_reset_tag(addr), POISON_INUSE, slab_size(slab)); |
a7101224 AK |
1534 | metadata_access_disable(); |
1535 | } | |
1536 | ||
becfda68 | 1537 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
bb192ed9 | 1538 | struct slab *slab, void *object) |
81819f0f | 1539 | { |
bb192ed9 | 1540 | if (!check_slab(s, slab)) |
becfda68 | 1541 | return 0; |
81819f0f | 1542 | |
bb192ed9 VB |
1543 | if (!check_valid_pointer(s, slab, object)) { |
1544 | object_err(s, slab, object, "Freelist Pointer check fails"); | |
becfda68 | 1545 | return 0; |
81819f0f CL |
1546 | } |
1547 | ||
bb192ed9 | 1548 | if (!check_object(s, slab, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1549 | return 0; |
1550 | ||
1551 | return 1; | |
1552 | } | |
1553 | ||
fa9b88e4 | 1554 | static noinline bool alloc_debug_processing(struct kmem_cache *s, |
6edf2576 | 1555 | struct slab *slab, void *object, int orig_size) |
becfda68 LA |
1556 | { |
1557 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
bb192ed9 | 1558 | if (!alloc_consistency_checks(s, slab, object)) |
becfda68 LA |
1559 | goto bad; |
1560 | } | |
81819f0f | 1561 | |
c7323a5a | 1562 | /* Success. Perform special debug activities for allocs */ |
bb192ed9 | 1563 | trace(s, slab, object, 1); |
6edf2576 | 1564 | set_orig_size(s, object, orig_size); |
f7cb1933 | 1565 | init_object(s, object, SLUB_RED_ACTIVE); |
fa9b88e4 | 1566 | return true; |
3ec09742 | 1567 | |
81819f0f | 1568 | bad: |
bb192ed9 | 1569 | if (folio_test_slab(slab_folio(slab))) { |
81819f0f CL |
1570 | /* |
1571 | * If this is a slab page then lets do the best we can | |
1572 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1573 | * as used avoids touching the remaining objects. |
81819f0f | 1574 | */ |
24922684 | 1575 | slab_fix(s, "Marking all objects used"); |
bb192ed9 VB |
1576 | slab->inuse = slab->objects; |
1577 | slab->freelist = NULL; | |
81819f0f | 1578 | } |
fa9b88e4 | 1579 | return false; |
81819f0f CL |
1580 | } |
1581 | ||
becfda68 | 1582 | static inline int free_consistency_checks(struct kmem_cache *s, |
bb192ed9 | 1583 | struct slab *slab, void *object, unsigned long addr) |
81819f0f | 1584 | { |
bb192ed9 VB |
1585 | if (!check_valid_pointer(s, slab, object)) { |
1586 | slab_err(s, slab, "Invalid object pointer 0x%p", object); | |
becfda68 | 1587 | return 0; |
81819f0f CL |
1588 | } |
1589 | ||
bb192ed9 VB |
1590 | if (on_freelist(s, slab, object)) { |
1591 | object_err(s, slab, object, "Object already free"); | |
becfda68 | 1592 | return 0; |
81819f0f CL |
1593 | } |
1594 | ||
bb192ed9 | 1595 | if (!check_object(s, slab, object, SLUB_RED_ACTIVE)) |
becfda68 | 1596 | return 0; |
81819f0f | 1597 | |
bb192ed9 VB |
1598 | if (unlikely(s != slab->slab_cache)) { |
1599 | if (!folio_test_slab(slab_folio(slab))) { | |
1600 | slab_err(s, slab, "Attempt to free object(0x%p) outside of slab", | |
756a025f | 1601 | object); |
bb192ed9 | 1602 | } else if (!slab->slab_cache) { |
f9f58285 FF |
1603 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1604 | object); | |
70d71228 | 1605 | dump_stack(); |
06428780 | 1606 | } else |
bb192ed9 | 1607 | object_err(s, slab, object, |
24922684 | 1608 | "page slab pointer corrupt."); |
becfda68 LA |
1609 | return 0; |
1610 | } | |
1611 | return 1; | |
1612 | } | |
1613 | ||
e17f1dfb | 1614 | /* |
671776b3 | 1615 | * Parse a block of slab_debug options. Blocks are delimited by ';' |
e17f1dfb VB |
1616 | * |
1617 | * @str: start of block | |
1618 | * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified | |
1619 | * @slabs: return start of list of slabs, or NULL when there's no list | |
1620 | * @init: assume this is initial parsing and not per-kmem-create parsing | |
1621 | * | |
1622 | * returns the start of next block if there's any, or NULL | |
1623 | */ | |
1624 | static char * | |
1625 | parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init) | |
41ecc55b | 1626 | { |
e17f1dfb | 1627 | bool higher_order_disable = false; |
f0630fff | 1628 | |
e17f1dfb VB |
1629 | /* Skip any completely empty blocks */ |
1630 | while (*str && *str == ';') | |
1631 | str++; | |
1632 | ||
1633 | if (*str == ',') { | |
f0630fff CL |
1634 | /* |
1635 | * No options but restriction on slabs. This means full | |
1636 | * debugging for slabs matching a pattern. | |
1637 | */ | |
e17f1dfb | 1638 | *flags = DEBUG_DEFAULT_FLAGS; |
f0630fff | 1639 | goto check_slabs; |
e17f1dfb VB |
1640 | } |
1641 | *flags = 0; | |
f0630fff | 1642 | |
e17f1dfb VB |
1643 | /* Determine which debug features should be switched on */ |
1644 | for (; *str && *str != ',' && *str != ';'; str++) { | |
f0630fff | 1645 | switch (tolower(*str)) { |
e17f1dfb VB |
1646 | case '-': |
1647 | *flags = 0; | |
1648 | break; | |
f0630fff | 1649 | case 'f': |
e17f1dfb | 1650 | *flags |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1651 | break; |
1652 | case 'z': | |
e17f1dfb | 1653 | *flags |= SLAB_RED_ZONE; |
f0630fff CL |
1654 | break; |
1655 | case 'p': | |
e17f1dfb | 1656 | *flags |= SLAB_POISON; |
f0630fff CL |
1657 | break; |
1658 | case 'u': | |
e17f1dfb | 1659 | *flags |= SLAB_STORE_USER; |
f0630fff CL |
1660 | break; |
1661 | case 't': | |
e17f1dfb | 1662 | *flags |= SLAB_TRACE; |
f0630fff | 1663 | break; |
4c13dd3b | 1664 | case 'a': |
e17f1dfb | 1665 | *flags |= SLAB_FAILSLAB; |
4c13dd3b | 1666 | break; |
08303a73 CA |
1667 | case 'o': |
1668 | /* | |
1669 | * Avoid enabling debugging on caches if its minimum | |
1670 | * order would increase as a result. | |
1671 | */ | |
e17f1dfb | 1672 | higher_order_disable = true; |
08303a73 | 1673 | break; |
f0630fff | 1674 | default: |
e17f1dfb | 1675 | if (init) |
671776b3 | 1676 | pr_err("slab_debug option '%c' unknown. skipped\n", *str); |
f0630fff | 1677 | } |
41ecc55b | 1678 | } |
f0630fff | 1679 | check_slabs: |
41ecc55b | 1680 | if (*str == ',') |
e17f1dfb VB |
1681 | *slabs = ++str; |
1682 | else | |
1683 | *slabs = NULL; | |
1684 | ||
1685 | /* Skip over the slab list */ | |
1686 | while (*str && *str != ';') | |
1687 | str++; | |
1688 | ||
1689 | /* Skip any completely empty blocks */ | |
1690 | while (*str && *str == ';') | |
1691 | str++; | |
1692 | ||
1693 | if (init && higher_order_disable) | |
1694 | disable_higher_order_debug = 1; | |
1695 | ||
1696 | if (*str) | |
1697 | return str; | |
1698 | else | |
1699 | return NULL; | |
1700 | } | |
1701 | ||
1702 | static int __init setup_slub_debug(char *str) | |
1703 | { | |
1704 | slab_flags_t flags; | |
a7f1d485 | 1705 | slab_flags_t global_flags; |
e17f1dfb VB |
1706 | char *saved_str; |
1707 | char *slab_list; | |
1708 | bool global_slub_debug_changed = false; | |
1709 | bool slab_list_specified = false; | |
1710 | ||
a7f1d485 | 1711 | global_flags = DEBUG_DEFAULT_FLAGS; |
e17f1dfb VB |
1712 | if (*str++ != '=' || !*str) |
1713 | /* | |
1714 | * No options specified. Switch on full debugging. | |
1715 | */ | |
1716 | goto out; | |
1717 | ||
1718 | saved_str = str; | |
1719 | while (str) { | |
1720 | str = parse_slub_debug_flags(str, &flags, &slab_list, true); | |
1721 | ||
1722 | if (!slab_list) { | |
a7f1d485 | 1723 | global_flags = flags; |
e17f1dfb VB |
1724 | global_slub_debug_changed = true; |
1725 | } else { | |
1726 | slab_list_specified = true; | |
5cf909c5 | 1727 | if (flags & SLAB_STORE_USER) |
1c0310ad | 1728 | stack_depot_request_early_init(); |
e17f1dfb VB |
1729 | } |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * For backwards compatibility, a single list of flags with list of | |
a7f1d485 | 1734 | * slabs means debugging is only changed for those slabs, so the global |
671776b3 | 1735 | * slab_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending |
a7f1d485 | 1736 | * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as |
e17f1dfb VB |
1737 | * long as there is no option specifying flags without a slab list. |
1738 | */ | |
1739 | if (slab_list_specified) { | |
1740 | if (!global_slub_debug_changed) | |
a7f1d485 | 1741 | global_flags = slub_debug; |
e17f1dfb VB |
1742 | slub_debug_string = saved_str; |
1743 | } | |
f0630fff | 1744 | out: |
a7f1d485 | 1745 | slub_debug = global_flags; |
5cf909c5 | 1746 | if (slub_debug & SLAB_STORE_USER) |
1c0310ad | 1747 | stack_depot_request_early_init(); |
ca0cab65 VB |
1748 | if (slub_debug != 0 || slub_debug_string) |
1749 | static_branch_enable(&slub_debug_enabled); | |
02ac47d0 SB |
1750 | else |
1751 | static_branch_disable(&slub_debug_enabled); | |
6471384a AP |
1752 | if ((static_branch_unlikely(&init_on_alloc) || |
1753 | static_branch_unlikely(&init_on_free)) && | |
1754 | (slub_debug & SLAB_POISON)) | |
1755 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
41ecc55b CL |
1756 | return 1; |
1757 | } | |
1758 | ||
671776b3 XS |
1759 | __setup("slab_debug", setup_slub_debug); |
1760 | __setup_param("slub_debug", slub_debug, setup_slub_debug, 0); | |
41ecc55b | 1761 | |
c5fd3ca0 AT |
1762 | /* |
1763 | * kmem_cache_flags - apply debugging options to the cache | |
c5fd3ca0 AT |
1764 | * @flags: flags to set |
1765 | * @name: name of the cache | |
c5fd3ca0 AT |
1766 | * |
1767 | * Debug option(s) are applied to @flags. In addition to the debug | |
1768 | * option(s), if a slab name (or multiple) is specified i.e. | |
671776b3 | 1769 | * slab_debug=<Debug-Options>,<slab name1>,<slab name2> ... |
c5fd3ca0 AT |
1770 | * then only the select slabs will receive the debug option(s). |
1771 | */ | |
303cd693 | 1772 | slab_flags_t kmem_cache_flags(slab_flags_t flags, const char *name) |
41ecc55b | 1773 | { |
c5fd3ca0 AT |
1774 | char *iter; |
1775 | size_t len; | |
e17f1dfb VB |
1776 | char *next_block; |
1777 | slab_flags_t block_flags; | |
ca220593 JB |
1778 | slab_flags_t slub_debug_local = slub_debug; |
1779 | ||
a285909f HY |
1780 | if (flags & SLAB_NO_USER_FLAGS) |
1781 | return flags; | |
1782 | ||
ca220593 JB |
1783 | /* |
1784 | * If the slab cache is for debugging (e.g. kmemleak) then | |
1785 | * don't store user (stack trace) information by default, | |
1786 | * but let the user enable it via the command line below. | |
1787 | */ | |
1788 | if (flags & SLAB_NOLEAKTRACE) | |
1789 | slub_debug_local &= ~SLAB_STORE_USER; | |
c5fd3ca0 | 1790 | |
c5fd3ca0 | 1791 | len = strlen(name); |
e17f1dfb VB |
1792 | next_block = slub_debug_string; |
1793 | /* Go through all blocks of debug options, see if any matches our slab's name */ | |
1794 | while (next_block) { | |
1795 | next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false); | |
1796 | if (!iter) | |
1797 | continue; | |
1798 | /* Found a block that has a slab list, search it */ | |
1799 | while (*iter) { | |
1800 | char *end, *glob; | |
1801 | size_t cmplen; | |
1802 | ||
1803 | end = strchrnul(iter, ','); | |
1804 | if (next_block && next_block < end) | |
1805 | end = next_block - 1; | |
1806 | ||
1807 | glob = strnchr(iter, end - iter, '*'); | |
1808 | if (glob) | |
1809 | cmplen = glob - iter; | |
1810 | else | |
1811 | cmplen = max_t(size_t, len, (end - iter)); | |
c5fd3ca0 | 1812 | |
e17f1dfb VB |
1813 | if (!strncmp(name, iter, cmplen)) { |
1814 | flags |= block_flags; | |
1815 | return flags; | |
1816 | } | |
c5fd3ca0 | 1817 | |
e17f1dfb VB |
1818 | if (!*end || *end == ';') |
1819 | break; | |
1820 | iter = end + 1; | |
c5fd3ca0 | 1821 | } |
c5fd3ca0 | 1822 | } |
ba0268a8 | 1823 | |
ca220593 | 1824 | return flags | slub_debug_local; |
41ecc55b | 1825 | } |
b4a64718 | 1826 | #else /* !CONFIG_SLUB_DEBUG */ |
c0f81a94 | 1827 | static inline void setup_object_debug(struct kmem_cache *s, void *object) {} |
a50b854e | 1828 | static inline |
bb192ed9 | 1829 | void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr) {} |
41ecc55b | 1830 | |
fa9b88e4 VB |
1831 | static inline bool alloc_debug_processing(struct kmem_cache *s, |
1832 | struct slab *slab, void *object, int orig_size) { return true; } | |
41ecc55b | 1833 | |
fa9b88e4 VB |
1834 | static inline bool free_debug_processing(struct kmem_cache *s, |
1835 | struct slab *slab, void *head, void *tail, int *bulk_cnt, | |
1836 | unsigned long addr, depot_stack_handle_t handle) { return true; } | |
41ecc55b | 1837 | |
a204e6d6 | 1838 | static inline void slab_pad_check(struct kmem_cache *s, struct slab *slab) {} |
bb192ed9 | 1839 | static inline int check_object(struct kmem_cache *s, struct slab *slab, |
f7cb1933 | 1840 | void *object, u8 val) { return 1; } |
fa9b88e4 | 1841 | static inline depot_stack_handle_t set_track_prepare(void) { return 0; } |
c7323a5a VB |
1842 | static inline void set_track(struct kmem_cache *s, void *object, |
1843 | enum track_item alloc, unsigned long addr) {} | |
5cc6eee8 | 1844 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
bb192ed9 | 1845 | struct slab *slab) {} |
c65c1877 | 1846 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
bb192ed9 | 1847 | struct slab *slab) {} |
303cd693 | 1848 | slab_flags_t kmem_cache_flags(slab_flags_t flags, const char *name) |
ba0268a8 CL |
1849 | { |
1850 | return flags; | |
1851 | } | |
41ecc55b | 1852 | #define slub_debug 0 |
0f389ec6 | 1853 | |
fdaa45e9 IM |
1854 | #define disable_higher_order_debug 0 |
1855 | ||
26c02cf0 AB |
1856 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1857 | { return 0; } | |
205ab99d CL |
1858 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1859 | int objects) {} | |
1860 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1861 | int objects) {} | |
7d550c56 | 1862 | |
0af8489b | 1863 | #ifndef CONFIG_SLUB_TINY |
bb192ed9 | 1864 | static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab, |
dc07a728 | 1865 | void **freelist, void *nextfree) |
52f23478 DZ |
1866 | { |
1867 | return false; | |
1868 | } | |
0af8489b | 1869 | #endif |
02e72cc6 AR |
1870 | #endif /* CONFIG_SLUB_DEBUG */ |
1871 | ||
21c690a3 SB |
1872 | #ifdef CONFIG_SLAB_OBJ_EXT |
1873 | ||
239d6c96 SB |
1874 | #ifdef CONFIG_MEM_ALLOC_PROFILING_DEBUG |
1875 | ||
1876 | static inline void mark_objexts_empty(struct slabobj_ext *obj_exts) | |
0bedcc66 | 1877 | { |
239d6c96 SB |
1878 | struct slabobj_ext *slab_exts; |
1879 | struct slab *obj_exts_slab; | |
1880 | ||
1881 | obj_exts_slab = virt_to_slab(obj_exts); | |
1882 | slab_exts = slab_obj_exts(obj_exts_slab); | |
1883 | if (slab_exts) { | |
1884 | unsigned int offs = obj_to_index(obj_exts_slab->slab_cache, | |
1885 | obj_exts_slab, obj_exts); | |
1886 | /* codetag should be NULL */ | |
1887 | WARN_ON(slab_exts[offs].ref.ct); | |
1888 | set_codetag_empty(&slab_exts[offs].ref); | |
1889 | } | |
0bedcc66 VB |
1890 | } |
1891 | ||
09c46563 | 1892 | static inline void mark_failed_objexts_alloc(struct slab *slab) |
0bedcc66 | 1893 | { |
09c46563 | 1894 | slab->obj_exts = OBJEXTS_ALLOC_FAIL; |
0bedcc66 VB |
1895 | } |
1896 | ||
09c46563 SB |
1897 | static inline void handle_failed_objexts_alloc(unsigned long obj_exts, |
1898 | struct slabobj_ext *vec, unsigned int objects) | |
0bedcc66 VB |
1899 | { |
1900 | /* | |
09c46563 SB |
1901 | * If vector previously failed to allocate then we have live |
1902 | * objects with no tag reference. Mark all references in this | |
1903 | * vector as empty to avoid warnings later on. | |
0bedcc66 | 1904 | */ |
09c46563 SB |
1905 | if (obj_exts & OBJEXTS_ALLOC_FAIL) { |
1906 | unsigned int i; | |
1907 | ||
1908 | for (i = 0; i < objects; i++) | |
1909 | set_codetag_empty(&vec[i].ref); | |
1910 | } | |
0bedcc66 VB |
1911 | } |
1912 | ||
239d6c96 SB |
1913 | #else /* CONFIG_MEM_ALLOC_PROFILING_DEBUG */ |
1914 | ||
1915 | static inline void mark_objexts_empty(struct slabobj_ext *obj_exts) {} | |
09c46563 SB |
1916 | static inline void mark_failed_objexts_alloc(struct slab *slab) {} |
1917 | static inline void handle_failed_objexts_alloc(unsigned long obj_exts, | |
1918 | struct slabobj_ext *vec, unsigned int objects) {} | |
239d6c96 SB |
1919 | |
1920 | #endif /* CONFIG_MEM_ALLOC_PROFILING_DEBUG */ | |
1921 | ||
0bedcc66 | 1922 | /* |
21c690a3 SB |
1923 | * The allocated objcg pointers array is not accounted directly. |
1924 | * Moreover, it should not come from DMA buffer and is not readily | |
1925 | * reclaimable. So those GFP bits should be masked off. | |
0bedcc66 | 1926 | */ |
21c690a3 SB |
1927 | #define OBJCGS_CLEAR_MASK (__GFP_DMA | __GFP_RECLAIMABLE | \ |
1928 | __GFP_ACCOUNT | __GFP_NOFAIL) | |
1929 | ||
e6100a45 VB |
1930 | int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s, |
1931 | gfp_t gfp, bool new_slab) | |
21c690a3 SB |
1932 | { |
1933 | unsigned int objects = objs_per_slab(s, slab); | |
09c46563 SB |
1934 | unsigned long new_exts; |
1935 | unsigned long old_exts; | |
1936 | struct slabobj_ext *vec; | |
21c690a3 SB |
1937 | |
1938 | gfp &= ~OBJCGS_CLEAR_MASK; | |
768c33be SB |
1939 | /* Prevent recursive extension vector allocation */ |
1940 | gfp |= __GFP_NO_OBJ_EXT; | |
21c690a3 SB |
1941 | vec = kcalloc_node(objects, sizeof(struct slabobj_ext), gfp, |
1942 | slab_nid(slab)); | |
09c46563 SB |
1943 | if (!vec) { |
1944 | /* Mark vectors which failed to allocate */ | |
1945 | if (new_slab) | |
1946 | mark_failed_objexts_alloc(slab); | |
1947 | ||
21c690a3 | 1948 | return -ENOMEM; |
09c46563 | 1949 | } |
21c690a3 | 1950 | |
09c46563 | 1951 | new_exts = (unsigned long)vec; |
21c690a3 | 1952 | #ifdef CONFIG_MEMCG |
09c46563 | 1953 | new_exts |= MEMCG_DATA_OBJEXTS; |
21c690a3 | 1954 | #endif |
09c46563 SB |
1955 | old_exts = slab->obj_exts; |
1956 | handle_failed_objexts_alloc(old_exts, vec, objects); | |
21c690a3 SB |
1957 | if (new_slab) { |
1958 | /* | |
1959 | * If the slab is brand new and nobody can yet access its | |
1960 | * obj_exts, no synchronization is required and obj_exts can | |
1961 | * be simply assigned. | |
1962 | */ | |
09c46563 SB |
1963 | slab->obj_exts = new_exts; |
1964 | } else if (cmpxchg(&slab->obj_exts, old_exts, new_exts) != old_exts) { | |
21c690a3 SB |
1965 | /* |
1966 | * If the slab is already in use, somebody can allocate and | |
1967 | * assign slabobj_exts in parallel. In this case the existing | |
1968 | * objcg vector should be reused. | |
1969 | */ | |
239d6c96 | 1970 | mark_objexts_empty(vec); |
21c690a3 SB |
1971 | kfree(vec); |
1972 | return 0; | |
1973 | } | |
1974 | ||
1975 | kmemleak_not_leak(vec); | |
1976 | return 0; | |
1977 | } | |
1978 | ||
1979 | static inline void free_slab_obj_exts(struct slab *slab) | |
0bedcc66 | 1980 | { |
21c690a3 SB |
1981 | struct slabobj_ext *obj_exts; |
1982 | ||
1983 | obj_exts = slab_obj_exts(slab); | |
1984 | if (!obj_exts) | |
1985 | return; | |
1986 | ||
0bedcc66 | 1987 | /* |
239d6c96 SB |
1988 | * obj_exts was created with __GFP_NO_OBJ_EXT flag, therefore its |
1989 | * corresponding extension will be NULL. alloc_tag_sub() will throw a | |
1990 | * warning if slab has extensions but the extension of an object is | |
1991 | * NULL, therefore replace NULL with CODETAG_EMPTY to indicate that | |
1992 | * the extension for obj_exts is expected to be NULL. | |
0bedcc66 | 1993 | */ |
239d6c96 | 1994 | mark_objexts_empty(obj_exts); |
21c690a3 SB |
1995 | kfree(obj_exts); |
1996 | slab->obj_exts = 0; | |
1997 | } | |
4b873696 SB |
1998 | |
1999 | static inline bool need_slab_obj_ext(void) | |
2000 | { | |
2001 | if (mem_alloc_profiling_enabled()) | |
0bedcc66 VB |
2002 | return true; |
2003 | ||
4b873696 SB |
2004 | /* |
2005 | * CONFIG_MEMCG_KMEM creates vector of obj_cgroup objects conditionally | |
2006 | * inside memcg_slab_post_alloc_hook. No other users for now. | |
2007 | */ | |
2008 | return false; | |
2009 | } | |
0bedcc66 | 2010 | |
4b873696 SB |
2011 | static inline struct slabobj_ext * |
2012 | prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p) | |
2013 | { | |
2014 | struct slab *slab; | |
0bedcc66 | 2015 | |
4b873696 SB |
2016 | if (!p) |
2017 | return NULL; | |
0bedcc66 | 2018 | |
3b89ec41 | 2019 | if (s->flags & (SLAB_NO_OBJ_EXT | SLAB_NOLEAKTRACE)) |
4b873696 | 2020 | return NULL; |
0bedcc66 | 2021 | |
4b873696 SB |
2022 | if (flags & __GFP_NO_OBJ_EXT) |
2023 | return NULL; | |
2024 | ||
2025 | slab = virt_to_slab(p); | |
2026 | if (!slab_obj_exts(slab) && | |
2027 | WARN(alloc_slab_obj_exts(slab, s, flags, false), | |
2028 | "%s, %s: Failed to create slab extension vector!\n", | |
2029 | __func__, s->name)) | |
2030 | return NULL; | |
2031 | ||
2032 | return slab_obj_exts(slab) + obj_to_index(s, slab, p); | |
0bedcc66 VB |
2033 | } |
2034 | ||
4b873696 SB |
2035 | static inline void |
2036 | alloc_tagging_slab_free_hook(struct kmem_cache *s, struct slab *slab, void **p, | |
2037 | int objects) | |
3450a0e5 | 2038 | { |
4b873696 SB |
2039 | #ifdef CONFIG_MEM_ALLOC_PROFILING |
2040 | struct slabobj_ext *obj_exts; | |
2041 | int i; | |
3450a0e5 | 2042 | |
4b873696 SB |
2043 | if (!mem_alloc_profiling_enabled()) |
2044 | return; | |
3450a0e5 | 2045 | |
4b873696 SB |
2046 | obj_exts = slab_obj_exts(slab); |
2047 | if (!obj_exts) | |
2048 | return; | |
2049 | ||
2050 | for (i = 0; i < objects; i++) { | |
2051 | unsigned int off = obj_to_index(s, slab, p[i]); | |
2052 | ||
2053 | alloc_tag_sub(&obj_exts[off].ref, s->size); | |
2054 | } | |
2055 | #endif | |
3450a0e5 VB |
2056 | } |
2057 | ||
21c690a3 | 2058 | #else /* CONFIG_SLAB_OBJ_EXT */ |
4b873696 | 2059 | |
21c690a3 SB |
2060 | static int alloc_slab_obj_exts(struct slab *slab, struct kmem_cache *s, |
2061 | gfp_t gfp, bool new_slab) | |
0bedcc66 | 2062 | { |
21c690a3 SB |
2063 | return 0; |
2064 | } | |
0bedcc66 | 2065 | |
21c690a3 | 2066 | static inline void free_slab_obj_exts(struct slab *slab) |
0bedcc66 | 2067 | { |
0bedcc66 | 2068 | } |
0bedcc66 | 2069 | |
4b873696 SB |
2070 | static inline bool need_slab_obj_ext(void) |
2071 | { | |
2072 | return false; | |
2073 | } | |
0bedcc66 | 2074 | |
4b873696 SB |
2075 | static inline struct slabobj_ext * |
2076 | prepare_slab_obj_exts_hook(struct kmem_cache *s, gfp_t flags, void *p) | |
2077 | { | |
2078 | return NULL; | |
2079 | } | |
0bedcc66 | 2080 | |
4b873696 SB |
2081 | static inline void |
2082 | alloc_tagging_slab_free_hook(struct kmem_cache *s, struct slab *slab, void **p, | |
2083 | int objects) | |
2084 | { | |
0bedcc66 VB |
2085 | } |
2086 | ||
21c690a3 | 2087 | #endif /* CONFIG_SLAB_OBJ_EXT */ |
0bedcc66 | 2088 | |
21c690a3 | 2089 | #ifdef CONFIG_MEMCG_KMEM |
0bedcc66 | 2090 | |
9f9796b4 VB |
2091 | static void memcg_alloc_abort_single(struct kmem_cache *s, void *object); |
2092 | ||
3450a0e5 | 2093 | static __fastpath_inline |
9f9796b4 | 2094 | bool memcg_slab_post_alloc_hook(struct kmem_cache *s, struct list_lru *lru, |
3450a0e5 VB |
2095 | gfp_t flags, size_t size, void **p) |
2096 | { | |
9f9796b4 | 2097 | if (likely(!memcg_kmem_online())) |
3450a0e5 | 2098 | return true; |
3450a0e5 | 2099 | |
3450a0e5 VB |
2100 | if (likely(!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))) |
2101 | return true; | |
0bedcc66 | 2102 | |
9f9796b4 VB |
2103 | if (likely(__memcg_slab_post_alloc_hook(s, lru, flags, size, p))) |
2104 | return true; | |
0bedcc66 | 2105 | |
9f9796b4 VB |
2106 | if (likely(size == 1)) { |
2107 | memcg_alloc_abort_single(s, *p); | |
2108 | *p = NULL; | |
2109 | } else { | |
2110 | kmem_cache_free_bulk(s, size, p); | |
0bedcc66 | 2111 | } |
3450a0e5 | 2112 | |
9f9796b4 | 2113 | return false; |
0bedcc66 | 2114 | } |
ecf9a253 VB |
2115 | |
2116 | static __fastpath_inline | |
2117 | void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, void **p, | |
2118 | int objects) | |
2119 | { | |
21c690a3 | 2120 | struct slabobj_ext *obj_exts; |
ecf9a253 VB |
2121 | |
2122 | if (!memcg_kmem_online()) | |
2123 | return; | |
2124 | ||
21c690a3 SB |
2125 | obj_exts = slab_obj_exts(slab); |
2126 | if (likely(!obj_exts)) | |
ecf9a253 VB |
2127 | return; |
2128 | ||
21c690a3 | 2129 | __memcg_slab_free_hook(s, slab, p, objects, obj_exts); |
520a688a | 2130 | } |
0bedcc66 | 2131 | #else /* CONFIG_MEMCG_KMEM */ |
9f9796b4 VB |
2132 | static inline bool memcg_slab_post_alloc_hook(struct kmem_cache *s, |
2133 | struct list_lru *lru, | |
0bedcc66 VB |
2134 | gfp_t flags, size_t size, |
2135 | void **p) | |
2136 | { | |
9f9796b4 | 2137 | return true; |
0bedcc66 VB |
2138 | } |
2139 | ||
2140 | static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, | |
2141 | void **p, int objects) | |
2142 | { | |
2143 | } | |
2144 | #endif /* CONFIG_MEMCG_KMEM */ | |
2145 | ||
02e72cc6 AR |
2146 | /* |
2147 | * Hooks for other subsystems that check memory allocations. In a typical | |
2148 | * production configuration these hooks all should produce no code at all. | |
284f17ac VB |
2149 | * |
2150 | * Returns true if freeing of the object can proceed, false if its reuse | |
782f8906 | 2151 | * was delayed by KASAN quarantine, or it was returned to KFENCE. |
02e72cc6 | 2152 | */ |
284f17ac VB |
2153 | static __always_inline |
2154 | bool slab_free_hook(struct kmem_cache *s, void *x, bool init) | |
d56791b3 RB |
2155 | { |
2156 | kmemleak_free_recursive(x, s->flags); | |
68ef169a | 2157 | kmsan_slab_free(s, x); |
7d550c56 | 2158 | |
84048039 | 2159 | debug_check_no_locks_freed(x, s->object_size); |
02e72cc6 | 2160 | |
02e72cc6 AR |
2161 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
2162 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 2163 | |
cfbe1636 ME |
2164 | /* Use KCSAN to help debug racy use-after-free. */ |
2165 | if (!(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
2166 | __kcsan_check_access(x, s->object_size, | |
2167 | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT); | |
2168 | ||
782f8906 VB |
2169 | if (kfence_free(x)) |
2170 | return false; | |
2171 | ||
d57a964e AK |
2172 | /* |
2173 | * As memory initialization might be integrated into KASAN, | |
2174 | * kasan_slab_free and initialization memset's must be | |
2175 | * kept together to avoid discrepancies in behavior. | |
2176 | * | |
2177 | * The initialization memset's clear the object and the metadata, | |
2178 | * but don't touch the SLAB redzone. | |
8f828aa4 NB |
2179 | * |
2180 | * The object's freepointer is also avoided if stored outside the | |
2181 | * object. | |
d57a964e | 2182 | */ |
ecf9a253 | 2183 | if (unlikely(init)) { |
d57a964e | 2184 | int rsize; |
8f828aa4 | 2185 | unsigned int inuse; |
d57a964e | 2186 | |
8f828aa4 | 2187 | inuse = get_info_end(s); |
d57a964e AK |
2188 | if (!kasan_has_integrated_init()) |
2189 | memset(kasan_reset_tag(x), 0, s->object_size); | |
2190 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0; | |
8f828aa4 NB |
2191 | memset((char *)kasan_reset_tag(x) + inuse, 0, |
2192 | s->size - inuse - rsize); | |
d57a964e AK |
2193 | } |
2194 | /* KASAN might put x into memory quarantine, delaying its reuse. */ | |
284f17ac | 2195 | return !kasan_slab_free(s, x, init); |
02e72cc6 | 2196 | } |
205ab99d | 2197 | |
9ea9cd8e KO |
2198 | static __fastpath_inline |
2199 | bool slab_free_freelist_hook(struct kmem_cache *s, void **head, void **tail, | |
2200 | int *cnt) | |
81084651 | 2201 | { |
6471384a AP |
2202 | |
2203 | void *object; | |
2204 | void *next = *head; | |
284f17ac | 2205 | void *old_tail = *tail; |
782f8906 | 2206 | bool init; |
6471384a | 2207 | |
b89fb5ef | 2208 | if (is_kfence_address(next)) { |
d57a964e | 2209 | slab_free_hook(s, next, false); |
782f8906 | 2210 | return false; |
b89fb5ef AP |
2211 | } |
2212 | ||
aea4df4c LA |
2213 | /* Head and tail of the reconstructed freelist */ |
2214 | *head = NULL; | |
2215 | *tail = NULL; | |
1b7e816f | 2216 | |
782f8906 VB |
2217 | init = slab_want_init_on_free(s); |
2218 | ||
aea4df4c LA |
2219 | do { |
2220 | object = next; | |
2221 | next = get_freepointer(s, object); | |
2222 | ||
c3895391 | 2223 | /* If object's reuse doesn't have to be delayed */ |
782f8906 | 2224 | if (likely(slab_free_hook(s, object, init))) { |
c3895391 AK |
2225 | /* Move object to the new freelist */ |
2226 | set_freepointer(s, object, *head); | |
2227 | *head = object; | |
2228 | if (!*tail) | |
2229 | *tail = object; | |
899447f6 ML |
2230 | } else { |
2231 | /* | |
2232 | * Adjust the reconstructed freelist depth | |
2233 | * accordingly if object's reuse is delayed. | |
2234 | */ | |
2235 | --(*cnt); | |
c3895391 AK |
2236 | } |
2237 | } while (object != old_tail); | |
2238 | ||
c3895391 | 2239 | return *head != NULL; |
81084651 JDB |
2240 | } |
2241 | ||
c0f81a94 | 2242 | static void *setup_object(struct kmem_cache *s, void *object) |
588f8ba9 | 2243 | { |
c0f81a94 | 2244 | setup_object_debug(s, object); |
4d176711 | 2245 | object = kasan_init_slab_obj(s, object); |
588f8ba9 | 2246 | if (unlikely(s->ctor)) { |
1ce9a052 | 2247 | kasan_unpoison_new_object(s, object); |
588f8ba9 | 2248 | s->ctor(object); |
1ce9a052 | 2249 | kasan_poison_new_object(s, object); |
588f8ba9 | 2250 | } |
4d176711 | 2251 | return object; |
588f8ba9 TG |
2252 | } |
2253 | ||
81819f0f CL |
2254 | /* |
2255 | * Slab allocation and freeing | |
2256 | */ | |
a485e1da XS |
2257 | static inline struct slab *alloc_slab_page(gfp_t flags, int node, |
2258 | struct kmem_cache_order_objects oo) | |
65c3376a | 2259 | { |
45387b8c VB |
2260 | struct folio *folio; |
2261 | struct slab *slab; | |
19af27af | 2262 | unsigned int order = oo_order(oo); |
65c3376a | 2263 | |
8014c46a | 2264 | folio = (struct folio *)alloc_pages_node(node, flags, order); |
45387b8c VB |
2265 | if (!folio) |
2266 | return NULL; | |
2267 | ||
2268 | slab = folio_slab(folio); | |
2269 | __folio_set_slab(folio); | |
8b881763 VB |
2270 | /* Make the flag visible before any changes to folio->mapping */ |
2271 | smp_wmb(); | |
02d65d6f | 2272 | if (folio_is_pfmemalloc(folio)) |
45387b8c VB |
2273 | slab_set_pfmemalloc(slab); |
2274 | ||
2275 | return slab; | |
65c3376a CL |
2276 | } |
2277 | ||
210e7a43 TG |
2278 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
2279 | /* Pre-initialize the random sequence cache */ | |
2280 | static int init_cache_random_seq(struct kmem_cache *s) | |
2281 | { | |
19af27af | 2282 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 2283 | int err; |
210e7a43 | 2284 | |
a810007a SR |
2285 | /* Bailout if already initialised */ |
2286 | if (s->random_seq) | |
2287 | return 0; | |
2288 | ||
210e7a43 TG |
2289 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
2290 | if (err) { | |
2291 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
2292 | s->name); | |
2293 | return err; | |
2294 | } | |
2295 | ||
2296 | /* Transform to an offset on the set of pages */ | |
2297 | if (s->random_seq) { | |
19af27af AD |
2298 | unsigned int i; |
2299 | ||
210e7a43 TG |
2300 | for (i = 0; i < count; i++) |
2301 | s->random_seq[i] *= s->size; | |
2302 | } | |
2303 | return 0; | |
2304 | } | |
2305 | ||
2306 | /* Initialize each random sequence freelist per cache */ | |
2307 | static void __init init_freelist_randomization(void) | |
2308 | { | |
2309 | struct kmem_cache *s; | |
2310 | ||
2311 | mutex_lock(&slab_mutex); | |
2312 | ||
2313 | list_for_each_entry(s, &slab_caches, list) | |
2314 | init_cache_random_seq(s); | |
2315 | ||
2316 | mutex_unlock(&slab_mutex); | |
2317 | } | |
2318 | ||
2319 | /* Get the next entry on the pre-computed freelist randomized */ | |
c63349fc | 2320 | static void *next_freelist_entry(struct kmem_cache *s, |
210e7a43 TG |
2321 | unsigned long *pos, void *start, |
2322 | unsigned long page_limit, | |
2323 | unsigned long freelist_count) | |
2324 | { | |
2325 | unsigned int idx; | |
2326 | ||
2327 | /* | |
2328 | * If the target page allocation failed, the number of objects on the | |
2329 | * page might be smaller than the usual size defined by the cache. | |
2330 | */ | |
2331 | do { | |
2332 | idx = s->random_seq[*pos]; | |
2333 | *pos += 1; | |
2334 | if (*pos >= freelist_count) | |
2335 | *pos = 0; | |
2336 | } while (unlikely(idx >= page_limit)); | |
2337 | ||
2338 | return (char *)start + idx; | |
2339 | } | |
2340 | ||
2341 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
bb192ed9 | 2342 | static bool shuffle_freelist(struct kmem_cache *s, struct slab *slab) |
210e7a43 TG |
2343 | { |
2344 | void *start; | |
2345 | void *cur; | |
2346 | void *next; | |
2347 | unsigned long idx, pos, page_limit, freelist_count; | |
2348 | ||
bb192ed9 | 2349 | if (slab->objects < 2 || !s->random_seq) |
210e7a43 TG |
2350 | return false; |
2351 | ||
2352 | freelist_count = oo_objects(s->oo); | |
8032bf12 | 2353 | pos = get_random_u32_below(freelist_count); |
210e7a43 | 2354 | |
bb192ed9 VB |
2355 | page_limit = slab->objects * s->size; |
2356 | start = fixup_red_left(s, slab_address(slab)); | |
210e7a43 TG |
2357 | |
2358 | /* First entry is used as the base of the freelist */ | |
c63349fc | 2359 | cur = next_freelist_entry(s, &pos, start, page_limit, freelist_count); |
c0f81a94 | 2360 | cur = setup_object(s, cur); |
bb192ed9 | 2361 | slab->freelist = cur; |
210e7a43 | 2362 | |
bb192ed9 | 2363 | for (idx = 1; idx < slab->objects; idx++) { |
c63349fc | 2364 | next = next_freelist_entry(s, &pos, start, page_limit, |
210e7a43 | 2365 | freelist_count); |
c0f81a94 | 2366 | next = setup_object(s, next); |
210e7a43 TG |
2367 | set_freepointer(s, cur, next); |
2368 | cur = next; | |
2369 | } | |
210e7a43 TG |
2370 | set_freepointer(s, cur, NULL); |
2371 | ||
2372 | return true; | |
2373 | } | |
2374 | #else | |
2375 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
2376 | { | |
2377 | return 0; | |
2378 | } | |
2379 | static inline void init_freelist_randomization(void) { } | |
bb192ed9 | 2380 | static inline bool shuffle_freelist(struct kmem_cache *s, struct slab *slab) |
210e7a43 TG |
2381 | { |
2382 | return false; | |
2383 | } | |
2384 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
2385 | ||
0bedcc66 VB |
2386 | static __always_inline void account_slab(struct slab *slab, int order, |
2387 | struct kmem_cache *s, gfp_t gfp) | |
2388 | { | |
2389 | if (memcg_kmem_online() && (s->flags & SLAB_ACCOUNT)) | |
21c690a3 | 2390 | alloc_slab_obj_exts(slab, s, gfp, true); |
0bedcc66 VB |
2391 | |
2392 | mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s), | |
2393 | PAGE_SIZE << order); | |
2394 | } | |
2395 | ||
2396 | static __always_inline void unaccount_slab(struct slab *slab, int order, | |
2397 | struct kmem_cache *s) | |
2398 | { | |
4b873696 | 2399 | if (memcg_kmem_online() || need_slab_obj_ext()) |
21c690a3 | 2400 | free_slab_obj_exts(slab); |
0bedcc66 VB |
2401 | |
2402 | mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s), | |
2403 | -(PAGE_SIZE << order)); | |
2404 | } | |
2405 | ||
bb192ed9 | 2406 | static struct slab *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
81819f0f | 2407 | { |
bb192ed9 | 2408 | struct slab *slab; |
834f3d11 | 2409 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 2410 | gfp_t alloc_gfp; |
4d176711 | 2411 | void *start, *p, *next; |
a50b854e | 2412 | int idx; |
210e7a43 | 2413 | bool shuffle; |
81819f0f | 2414 | |
7e0528da CL |
2415 | flags &= gfp_allowed_mask; |
2416 | ||
b7a49f0d | 2417 | flags |= s->allocflags; |
e12ba74d | 2418 | |
ba52270d PE |
2419 | /* |
2420 | * Let the initial higher-order allocation fail under memory pressure | |
2421 | * so we fall-back to the minimum order allocation. | |
2422 | */ | |
2423 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 2424 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
27c08f75 | 2425 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_RECLAIM; |
ba52270d | 2426 | |
a485e1da | 2427 | slab = alloc_slab_page(alloc_gfp, node, oo); |
bb192ed9 | 2428 | if (unlikely(!slab)) { |
65c3376a | 2429 | oo = s->min; |
80c3a998 | 2430 | alloc_gfp = flags; |
65c3376a CL |
2431 | /* |
2432 | * Allocation may have failed due to fragmentation. | |
2433 | * Try a lower order alloc if possible | |
2434 | */ | |
a485e1da | 2435 | slab = alloc_slab_page(alloc_gfp, node, oo); |
bb192ed9 | 2436 | if (unlikely(!slab)) |
c7323a5a | 2437 | return NULL; |
588f8ba9 | 2438 | stat(s, ORDER_FALLBACK); |
65c3376a | 2439 | } |
5a896d9e | 2440 | |
bb192ed9 | 2441 | slab->objects = oo_objects(oo); |
c7323a5a VB |
2442 | slab->inuse = 0; |
2443 | slab->frozen = 0; | |
81819f0f | 2444 | |
bb192ed9 | 2445 | account_slab(slab, oo_order(oo), s, flags); |
1f3147b4 | 2446 | |
bb192ed9 | 2447 | slab->slab_cache = s; |
81819f0f | 2448 | |
6e48a966 | 2449 | kasan_poison_slab(slab); |
81819f0f | 2450 | |
bb192ed9 | 2451 | start = slab_address(slab); |
81819f0f | 2452 | |
bb192ed9 | 2453 | setup_slab_debug(s, slab, start); |
0316bec2 | 2454 | |
bb192ed9 | 2455 | shuffle = shuffle_freelist(s, slab); |
210e7a43 TG |
2456 | |
2457 | if (!shuffle) { | |
4d176711 | 2458 | start = fixup_red_left(s, start); |
c0f81a94 | 2459 | start = setup_object(s, start); |
bb192ed9 VB |
2460 | slab->freelist = start; |
2461 | for (idx = 0, p = start; idx < slab->objects - 1; idx++) { | |
18e50661 | 2462 | next = p + s->size; |
c0f81a94 | 2463 | next = setup_object(s, next); |
18e50661 AK |
2464 | set_freepointer(s, p, next); |
2465 | p = next; | |
2466 | } | |
2467 | set_freepointer(s, p, NULL); | |
81819f0f | 2468 | } |
81819f0f | 2469 | |
bb192ed9 | 2470 | return slab; |
81819f0f CL |
2471 | } |
2472 | ||
bb192ed9 | 2473 | static struct slab *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
588f8ba9 | 2474 | { |
44405099 LL |
2475 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
2476 | flags = kmalloc_fix_flags(flags); | |
588f8ba9 | 2477 | |
53a0de06 VB |
2478 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2479 | ||
588f8ba9 TG |
2480 | return allocate_slab(s, |
2481 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
2482 | } | |
2483 | ||
4020b4a2 | 2484 | static void __free_slab(struct kmem_cache *s, struct slab *slab) |
81819f0f | 2485 | { |
4020b4a2 VB |
2486 | struct folio *folio = slab_folio(slab); |
2487 | int order = folio_order(folio); | |
834f3d11 | 2488 | int pages = 1 << order; |
81819f0f | 2489 | |
4020b4a2 | 2490 | __slab_clear_pfmemalloc(slab); |
4020b4a2 | 2491 | folio->mapping = NULL; |
8b881763 VB |
2492 | /* Make the mapping reset visible before clearing the flag */ |
2493 | smp_wmb(); | |
2494 | __folio_clear_slab(folio); | |
c7b23b68 | 2495 | mm_account_reclaimed_pages(pages); |
4020b4a2 | 2496 | unaccount_slab(slab, order, s); |
c034c6a4 | 2497 | __free_pages(&folio->page, order); |
81819f0f CL |
2498 | } |
2499 | ||
2500 | static void rcu_free_slab(struct rcu_head *h) | |
2501 | { | |
bb192ed9 | 2502 | struct slab *slab = container_of(h, struct slab, rcu_head); |
da9a638c | 2503 | |
bb192ed9 | 2504 | __free_slab(slab->slab_cache, slab); |
81819f0f CL |
2505 | } |
2506 | ||
bb192ed9 | 2507 | static void free_slab(struct kmem_cache *s, struct slab *slab) |
81819f0f | 2508 | { |
bc29d5bd VB |
2509 | if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) { |
2510 | void *p; | |
2511 | ||
2512 | slab_pad_check(s, slab); | |
2513 | for_each_object(p, s, slab_address(slab), slab->objects) | |
2514 | check_object(s, slab, p, SLUB_RED_INACTIVE); | |
2515 | } | |
2516 | ||
2517 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
bb192ed9 | 2518 | call_rcu(&slab->rcu_head, rcu_free_slab); |
bc29d5bd | 2519 | else |
bb192ed9 | 2520 | __free_slab(s, slab); |
81819f0f CL |
2521 | } |
2522 | ||
bb192ed9 | 2523 | static void discard_slab(struct kmem_cache *s, struct slab *slab) |
81819f0f | 2524 | { |
bb192ed9 VB |
2525 | dec_slabs_node(s, slab_nid(slab), slab->objects); |
2526 | free_slab(s, slab); | |
81819f0f CL |
2527 | } |
2528 | ||
8a399e2f CZ |
2529 | /* |
2530 | * SLUB reuses PG_workingset bit to keep track of whether it's on | |
2531 | * the per-node partial list. | |
2532 | */ | |
2533 | static inline bool slab_test_node_partial(const struct slab *slab) | |
2534 | { | |
2535 | return folio_test_workingset((struct folio *)slab_folio(slab)); | |
2536 | } | |
2537 | ||
2538 | static inline void slab_set_node_partial(struct slab *slab) | |
2539 | { | |
2540 | set_bit(PG_workingset, folio_flags(slab_folio(slab), 0)); | |
2541 | } | |
2542 | ||
2543 | static inline void slab_clear_node_partial(struct slab *slab) | |
2544 | { | |
2545 | clear_bit(PG_workingset, folio_flags(slab_folio(slab), 0)); | |
2546 | } | |
2547 | ||
81819f0f | 2548 | /* |
5cc6eee8 | 2549 | * Management of partially allocated slabs. |
81819f0f | 2550 | */ |
1e4dd946 | 2551 | static inline void |
bb192ed9 | 2552 | __add_partial(struct kmem_cache_node *n, struct slab *slab, int tail) |
81819f0f | 2553 | { |
e95eed57 | 2554 | n->nr_partial++; |
136333d1 | 2555 | if (tail == DEACTIVATE_TO_TAIL) |
bb192ed9 | 2556 | list_add_tail(&slab->slab_list, &n->partial); |
7c2e132c | 2557 | else |
bb192ed9 | 2558 | list_add(&slab->slab_list, &n->partial); |
8a399e2f | 2559 | slab_set_node_partial(slab); |
81819f0f CL |
2560 | } |
2561 | ||
1e4dd946 | 2562 | static inline void add_partial(struct kmem_cache_node *n, |
bb192ed9 | 2563 | struct slab *slab, int tail) |
62e346a8 | 2564 | { |
c65c1877 | 2565 | lockdep_assert_held(&n->list_lock); |
bb192ed9 | 2566 | __add_partial(n, slab, tail); |
1e4dd946 | 2567 | } |
c65c1877 | 2568 | |
1e4dd946 | 2569 | static inline void remove_partial(struct kmem_cache_node *n, |
bb192ed9 | 2570 | struct slab *slab) |
1e4dd946 SR |
2571 | { |
2572 | lockdep_assert_held(&n->list_lock); | |
bb192ed9 | 2573 | list_del(&slab->slab_list); |
8a399e2f | 2574 | slab_clear_node_partial(slab); |
52b4b950 | 2575 | n->nr_partial--; |
1e4dd946 SR |
2576 | } |
2577 | ||
c7323a5a | 2578 | /* |
8cd3fa42 | 2579 | * Called only for kmem_cache_debug() caches instead of remove_partial(), with a |
c7323a5a VB |
2580 | * slab from the n->partial list. Remove only a single object from the slab, do |
2581 | * the alloc_debug_processing() checks and leave the slab on the list, or move | |
2582 | * it to full list if it was the last free object. | |
2583 | */ | |
2584 | static void *alloc_single_from_partial(struct kmem_cache *s, | |
6edf2576 | 2585 | struct kmem_cache_node *n, struct slab *slab, int orig_size) |
c7323a5a VB |
2586 | { |
2587 | void *object; | |
2588 | ||
2589 | lockdep_assert_held(&n->list_lock); | |
2590 | ||
2591 | object = slab->freelist; | |
2592 | slab->freelist = get_freepointer(s, object); | |
2593 | slab->inuse++; | |
2594 | ||
6edf2576 | 2595 | if (!alloc_debug_processing(s, slab, object, orig_size)) { |
c7323a5a VB |
2596 | remove_partial(n, slab); |
2597 | return NULL; | |
2598 | } | |
2599 | ||
2600 | if (slab->inuse == slab->objects) { | |
2601 | remove_partial(n, slab); | |
2602 | add_full(s, n, slab); | |
2603 | } | |
2604 | ||
2605 | return object; | |
2606 | } | |
2607 | ||
2608 | /* | |
2609 | * Called only for kmem_cache_debug() caches to allocate from a freshly | |
2610 | * allocated slab. Allocate a single object instead of whole freelist | |
2611 | * and put the slab to the partial (or full) list. | |
2612 | */ | |
2613 | static void *alloc_single_from_new_slab(struct kmem_cache *s, | |
6edf2576 | 2614 | struct slab *slab, int orig_size) |
c7323a5a VB |
2615 | { |
2616 | int nid = slab_nid(slab); | |
2617 | struct kmem_cache_node *n = get_node(s, nid); | |
2618 | unsigned long flags; | |
2619 | void *object; | |
2620 | ||
2621 | ||
2622 | object = slab->freelist; | |
2623 | slab->freelist = get_freepointer(s, object); | |
2624 | slab->inuse = 1; | |
2625 | ||
6edf2576 | 2626 | if (!alloc_debug_processing(s, slab, object, orig_size)) |
c7323a5a VB |
2627 | /* |
2628 | * It's not really expected that this would fail on a | |
2629 | * freshly allocated slab, but a concurrent memory | |
2630 | * corruption in theory could cause that. | |
2631 | */ | |
2632 | return NULL; | |
2633 | ||
2634 | spin_lock_irqsave(&n->list_lock, flags); | |
2635 | ||
2636 | if (slab->inuse == slab->objects) | |
2637 | add_full(s, n, slab); | |
2638 | else | |
2639 | add_partial(n, slab, DEACTIVATE_TO_HEAD); | |
2640 | ||
2641 | inc_slabs_node(s, nid, slab->objects); | |
2642 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2643 | ||
2644 | return object; | |
2645 | } | |
2646 | ||
e0a043aa | 2647 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
bb192ed9 | 2648 | static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain); |
e0a043aa | 2649 | #else |
bb192ed9 | 2650 | static inline void put_cpu_partial(struct kmem_cache *s, struct slab *slab, |
e0a043aa VB |
2651 | int drain) { } |
2652 | #endif | |
01b34d16 | 2653 | static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags); |
49e22585 | 2654 | |
81819f0f | 2655 | /* |
672bba3a | 2656 | * Try to allocate a partial slab from a specific node. |
81819f0f | 2657 | */ |
43c4c349 CZ |
2658 | static struct slab *get_partial_node(struct kmem_cache *s, |
2659 | struct kmem_cache_node *n, | |
2660 | struct partial_context *pc) | |
81819f0f | 2661 | { |
43c4c349 | 2662 | struct slab *slab, *slab2, *partial = NULL; |
4b1f449d | 2663 | unsigned long flags; |
bb192ed9 | 2664 | unsigned int partial_slabs = 0; |
81819f0f CL |
2665 | |
2666 | /* | |
2667 | * Racy check. If we mistakenly see no partial slabs then we | |
2668 | * just allocate an empty slab. If we mistakenly try to get a | |
70b6d25e | 2669 | * partial slab and there is none available then get_partial() |
672bba3a | 2670 | * will return NULL. |
81819f0f CL |
2671 | */ |
2672 | if (!n || !n->nr_partial) | |
2673 | return NULL; | |
2674 | ||
4b1f449d | 2675 | spin_lock_irqsave(&n->list_lock, flags); |
bb192ed9 | 2676 | list_for_each_entry_safe(slab, slab2, &n->partial, slab_list) { |
6edf2576 | 2677 | if (!pfmemalloc_match(slab, pc->flags)) |
8ba00bb6 JK |
2678 | continue; |
2679 | ||
0af8489b | 2680 | if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) { |
8cd3fa42 | 2681 | void *object = alloc_single_from_partial(s, n, slab, |
6edf2576 | 2682 | pc->orig_size); |
43c4c349 CZ |
2683 | if (object) { |
2684 | partial = slab; | |
2685 | pc->object = object; | |
c7323a5a | 2686 | break; |
43c4c349 | 2687 | } |
c7323a5a VB |
2688 | continue; |
2689 | } | |
2690 | ||
8cd3fa42 | 2691 | remove_partial(n, slab); |
49e22585 | 2692 | |
43c4c349 CZ |
2693 | if (!partial) { |
2694 | partial = slab; | |
49e22585 | 2695 | stat(s, ALLOC_FROM_PARTIAL); |
ff99b18f XS |
2696 | |
2697 | if ((slub_get_cpu_partial(s) == 0)) { | |
2698 | break; | |
2699 | } | |
49e22585 | 2700 | } else { |
bb192ed9 | 2701 | put_cpu_partial(s, slab, 0); |
8028dcea | 2702 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 2703 | |
ff99b18f XS |
2704 | if (++partial_slabs > slub_get_cpu_partial(s) / 2) { |
2705 | break; | |
2706 | } | |
2707 | } | |
497b66f2 | 2708 | } |
4b1f449d | 2709 | spin_unlock_irqrestore(&n->list_lock, flags); |
43c4c349 | 2710 | return partial; |
81819f0f CL |
2711 | } |
2712 | ||
2713 | /* | |
c2092c12 | 2714 | * Get a slab from somewhere. Search in increasing NUMA distances. |
81819f0f | 2715 | */ |
43c4c349 CZ |
2716 | static struct slab *get_any_partial(struct kmem_cache *s, |
2717 | struct partial_context *pc) | |
81819f0f CL |
2718 | { |
2719 | #ifdef CONFIG_NUMA | |
2720 | struct zonelist *zonelist; | |
dd1a239f | 2721 | struct zoneref *z; |
54a6eb5c | 2722 | struct zone *zone; |
6edf2576 | 2723 | enum zone_type highest_zoneidx = gfp_zone(pc->flags); |
43c4c349 | 2724 | struct slab *slab; |
cc9a6c87 | 2725 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
2726 | |
2727 | /* | |
672bba3a CL |
2728 | * The defrag ratio allows a configuration of the tradeoffs between |
2729 | * inter node defragmentation and node local allocations. A lower | |
2730 | * defrag_ratio increases the tendency to do local allocations | |
2731 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 2732 | * |
672bba3a CL |
2733 | * If the defrag_ratio is set to 0 then kmalloc() always |
2734 | * returns node local objects. If the ratio is higher then kmalloc() | |
2735 | * may return off node objects because partial slabs are obtained | |
2736 | * from other nodes and filled up. | |
81819f0f | 2737 | * |
43efd3ea LP |
2738 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
2739 | * (which makes defrag_ratio = 1000) then every (well almost) | |
2740 | * allocation will first attempt to defrag slab caches on other nodes. | |
2741 | * This means scanning over all nodes to look for partial slabs which | |
2742 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 2743 | * with available objects. |
81819f0f | 2744 | */ |
9824601e CL |
2745 | if (!s->remote_node_defrag_ratio || |
2746 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
2747 | return NULL; |
2748 | ||
cc9a6c87 | 2749 | do { |
d26914d1 | 2750 | cpuset_mems_cookie = read_mems_allowed_begin(); |
6edf2576 | 2751 | zonelist = node_zonelist(mempolicy_slab_node(), pc->flags); |
97a225e6 | 2752 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
cc9a6c87 MG |
2753 | struct kmem_cache_node *n; |
2754 | ||
2755 | n = get_node(s, zone_to_nid(zone)); | |
2756 | ||
6edf2576 | 2757 | if (n && cpuset_zone_allowed(zone, pc->flags) && |
cc9a6c87 | 2758 | n->nr_partial > s->min_partial) { |
43c4c349 CZ |
2759 | slab = get_partial_node(s, n, pc); |
2760 | if (slab) { | |
cc9a6c87 | 2761 | /* |
d26914d1 MG |
2762 | * Don't check read_mems_allowed_retry() |
2763 | * here - if mems_allowed was updated in | |
2764 | * parallel, that was a harmless race | |
2765 | * between allocation and the cpuset | |
2766 | * update | |
cc9a6c87 | 2767 | */ |
43c4c349 | 2768 | return slab; |
cc9a6c87 | 2769 | } |
c0ff7453 | 2770 | } |
81819f0f | 2771 | } |
d26914d1 | 2772 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 2773 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
2774 | return NULL; |
2775 | } | |
2776 | ||
2777 | /* | |
c2092c12 | 2778 | * Get a partial slab, lock it and return it. |
81819f0f | 2779 | */ |
43c4c349 CZ |
2780 | static struct slab *get_partial(struct kmem_cache *s, int node, |
2781 | struct partial_context *pc) | |
81819f0f | 2782 | { |
43c4c349 | 2783 | struct slab *slab; |
a561ce00 JK |
2784 | int searchnode = node; |
2785 | ||
2786 | if (node == NUMA_NO_NODE) | |
2787 | searchnode = numa_mem_id(); | |
81819f0f | 2788 | |
43c4c349 | 2789 | slab = get_partial_node(s, get_node(s, searchnode), pc); |
9198ffbd | 2790 | if (slab || (node != NUMA_NO_NODE && (pc->flags & __GFP_THISNODE))) |
43c4c349 | 2791 | return slab; |
81819f0f | 2792 | |
6edf2576 | 2793 | return get_any_partial(s, pc); |
81819f0f CL |
2794 | } |
2795 | ||
0af8489b VB |
2796 | #ifndef CONFIG_SLUB_TINY |
2797 | ||
923717cb | 2798 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2799 | /* |
0d645ed1 | 2800 | * Calculate the next globally unique transaction for disambiguation |
8a5ec0ba CL |
2801 | * during cmpxchg. The transactions start with the cpu number and are then |
2802 | * incremented by CONFIG_NR_CPUS. | |
2803 | */ | |
2804 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
2805 | #else | |
2806 | /* | |
2807 | * No preemption supported therefore also no need to check for | |
2808 | * different cpus. | |
2809 | */ | |
2810 | #define TID_STEP 1 | |
0af8489b | 2811 | #endif /* CONFIG_PREEMPTION */ |
8a5ec0ba CL |
2812 | |
2813 | static inline unsigned long next_tid(unsigned long tid) | |
2814 | { | |
2815 | return tid + TID_STEP; | |
2816 | } | |
2817 | ||
9d5f0be0 | 2818 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2819 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2820 | { | |
2821 | return tid % TID_STEP; | |
2822 | } | |
2823 | ||
2824 | static inline unsigned long tid_to_event(unsigned long tid) | |
2825 | { | |
2826 | return tid / TID_STEP; | |
2827 | } | |
9d5f0be0 | 2828 | #endif |
8a5ec0ba CL |
2829 | |
2830 | static inline unsigned int init_tid(int cpu) | |
2831 | { | |
2832 | return cpu; | |
2833 | } | |
2834 | ||
2835 | static inline void note_cmpxchg_failure(const char *n, | |
2836 | const struct kmem_cache *s, unsigned long tid) | |
2837 | { | |
2838 | #ifdef SLUB_DEBUG_CMPXCHG | |
2839 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2840 | ||
f9f58285 | 2841 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2842 | |
923717cb | 2843 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2844 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2845 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2846 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2847 | else | |
2848 | #endif | |
2849 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2850 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2851 | tid_to_event(tid), tid_to_event(actual_tid)); |
2852 | else | |
f9f58285 | 2853 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2854 | actual_tid, tid, next_tid(tid)); |
2855 | #endif | |
4fdccdfb | 2856 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2857 | } |
2858 | ||
788e1aad | 2859 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2860 | { |
8a5ec0ba | 2861 | int cpu; |
bd0e7491 | 2862 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2863 | |
bd0e7491 VB |
2864 | for_each_possible_cpu(cpu) { |
2865 | c = per_cpu_ptr(s->cpu_slab, cpu); | |
2866 | local_lock_init(&c->lock); | |
2867 | c->tid = init_tid(cpu); | |
2868 | } | |
8a5ec0ba | 2869 | } |
2cfb7455 | 2870 | |
81819f0f | 2871 | /* |
c2092c12 | 2872 | * Finishes removing the cpu slab. Merges cpu's freelist with slab's freelist, |
a019d201 VB |
2873 | * unfreezes the slabs and puts it on the proper list. |
2874 | * Assumes the slab has been already safely taken away from kmem_cache_cpu | |
2875 | * by the caller. | |
81819f0f | 2876 | */ |
bb192ed9 | 2877 | static void deactivate_slab(struct kmem_cache *s, struct slab *slab, |
a019d201 | 2878 | void *freelist) |
81819f0f | 2879 | { |
bb192ed9 | 2880 | struct kmem_cache_node *n = get_node(s, slab_nid(slab)); |
6d3a16d0 | 2881 | int free_delta = 0; |
d930ff03 | 2882 | void *nextfree, *freelist_iter, *freelist_tail; |
136333d1 | 2883 | int tail = DEACTIVATE_TO_HEAD; |
3406e91b | 2884 | unsigned long flags = 0; |
bb192ed9 VB |
2885 | struct slab new; |
2886 | struct slab old; | |
2cfb7455 | 2887 | |
844776cb | 2888 | if (READ_ONCE(slab->freelist)) { |
84e554e6 | 2889 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2890 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2891 | } |
2892 | ||
894b8788 | 2893 | /* |
d930ff03 VB |
2894 | * Stage one: Count the objects on cpu's freelist as free_delta and |
2895 | * remember the last object in freelist_tail for later splicing. | |
2cfb7455 | 2896 | */ |
d930ff03 VB |
2897 | freelist_tail = NULL; |
2898 | freelist_iter = freelist; | |
2899 | while (freelist_iter) { | |
2900 | nextfree = get_freepointer(s, freelist_iter); | |
2cfb7455 | 2901 | |
52f23478 DZ |
2902 | /* |
2903 | * If 'nextfree' is invalid, it is possible that the object at | |
d930ff03 VB |
2904 | * 'freelist_iter' is already corrupted. So isolate all objects |
2905 | * starting at 'freelist_iter' by skipping them. | |
52f23478 | 2906 | */ |
bb192ed9 | 2907 | if (freelist_corrupted(s, slab, &freelist_iter, nextfree)) |
52f23478 DZ |
2908 | break; |
2909 | ||
d930ff03 VB |
2910 | freelist_tail = freelist_iter; |
2911 | free_delta++; | |
2cfb7455 | 2912 | |
d930ff03 | 2913 | freelist_iter = nextfree; |
2cfb7455 CL |
2914 | } |
2915 | ||
894b8788 | 2916 | /* |
c2092c12 VB |
2917 | * Stage two: Unfreeze the slab while splicing the per-cpu |
2918 | * freelist to the head of slab's freelist. | |
894b8788 | 2919 | */ |
00eb60c2 CZ |
2920 | do { |
2921 | old.freelist = READ_ONCE(slab->freelist); | |
2922 | old.counters = READ_ONCE(slab->counters); | |
2923 | VM_BUG_ON(!old.frozen); | |
2924 | ||
2925 | /* Determine target state of the slab */ | |
2926 | new.counters = old.counters; | |
2927 | new.frozen = 0; | |
2928 | if (freelist_tail) { | |
2929 | new.inuse -= free_delta; | |
2930 | set_freepointer(s, freelist_tail, old.freelist); | |
2931 | new.freelist = freelist; | |
2932 | } else { | |
2933 | new.freelist = old.freelist; | |
2934 | } | |
2935 | } while (!slab_update_freelist(s, slab, | |
2936 | old.freelist, old.counters, | |
2937 | new.freelist, new.counters, | |
2938 | "unfreezing slab")); | |
2cfb7455 | 2939 | |
00eb60c2 CZ |
2940 | /* |
2941 | * Stage three: Manipulate the slab list based on the updated state. | |
2942 | */ | |
6d3a16d0 | 2943 | if (!new.inuse && n->nr_partial >= s->min_partial) { |
00eb60c2 CZ |
2944 | stat(s, DEACTIVATE_EMPTY); |
2945 | discard_slab(s, slab); | |
2946 | stat(s, FREE_SLAB); | |
6d3a16d0 | 2947 | } else if (new.freelist) { |
6d3a16d0 | 2948 | spin_lock_irqsave(&n->list_lock, flags); |
6d3a16d0 HY |
2949 | add_partial(n, slab, tail); |
2950 | spin_unlock_irqrestore(&n->list_lock, flags); | |
88349a28 | 2951 | stat(s, tail); |
00eb60c2 | 2952 | } else { |
6d3a16d0 | 2953 | stat(s, DEACTIVATE_FULL); |
894b8788 | 2954 | } |
81819f0f CL |
2955 | } |
2956 | ||
345c905d | 2957 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
21316fdc | 2958 | static void __put_partials(struct kmem_cache *s, struct slab *partial_slab) |
fc1455f4 | 2959 | { |
43d77867 | 2960 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
bb192ed9 | 2961 | struct slab *slab, *slab_to_discard = NULL; |
7cf9f3ba | 2962 | unsigned long flags = 0; |
49e22585 | 2963 | |
bb192ed9 | 2964 | while (partial_slab) { |
bb192ed9 VB |
2965 | slab = partial_slab; |
2966 | partial_slab = slab->next; | |
43d77867 | 2967 | |
bb192ed9 | 2968 | n2 = get_node(s, slab_nid(slab)); |
43d77867 JK |
2969 | if (n != n2) { |
2970 | if (n) | |
7cf9f3ba | 2971 | spin_unlock_irqrestore(&n->list_lock, flags); |
43d77867 JK |
2972 | |
2973 | n = n2; | |
7cf9f3ba | 2974 | spin_lock_irqsave(&n->list_lock, flags); |
43d77867 | 2975 | } |
49e22585 | 2976 | |
8cd3fa42 | 2977 | if (unlikely(!slab->inuse && n->nr_partial >= s->min_partial)) { |
bb192ed9 VB |
2978 | slab->next = slab_to_discard; |
2979 | slab_to_discard = slab; | |
43d77867 | 2980 | } else { |
bb192ed9 | 2981 | add_partial(n, slab, DEACTIVATE_TO_TAIL); |
43d77867 | 2982 | stat(s, FREE_ADD_PARTIAL); |
49e22585 CL |
2983 | } |
2984 | } | |
2985 | ||
2986 | if (n) | |
7cf9f3ba | 2987 | spin_unlock_irqrestore(&n->list_lock, flags); |
8de06a6f | 2988 | |
bb192ed9 VB |
2989 | while (slab_to_discard) { |
2990 | slab = slab_to_discard; | |
2991 | slab_to_discard = slab_to_discard->next; | |
9ada1934 SL |
2992 | |
2993 | stat(s, DEACTIVATE_EMPTY); | |
bb192ed9 | 2994 | discard_slab(s, slab); |
9ada1934 SL |
2995 | stat(s, FREE_SLAB); |
2996 | } | |
fc1455f4 | 2997 | } |
f3ab8b6b | 2998 | |
fc1455f4 | 2999 | /* |
21316fdc | 3000 | * Put all the cpu partial slabs to the node partial list. |
fc1455f4 | 3001 | */ |
21316fdc | 3002 | static void put_partials(struct kmem_cache *s) |
fc1455f4 | 3003 | { |
bb192ed9 | 3004 | struct slab *partial_slab; |
fc1455f4 VB |
3005 | unsigned long flags; |
3006 | ||
bd0e7491 | 3007 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 3008 | partial_slab = this_cpu_read(s->cpu_slab->partial); |
fc1455f4 | 3009 | this_cpu_write(s->cpu_slab->partial, NULL); |
bd0e7491 | 3010 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
fc1455f4 | 3011 | |
bb192ed9 | 3012 | if (partial_slab) |
21316fdc | 3013 | __put_partials(s, partial_slab); |
fc1455f4 VB |
3014 | } |
3015 | ||
21316fdc CZ |
3016 | static void put_partials_cpu(struct kmem_cache *s, |
3017 | struct kmem_cache_cpu *c) | |
fc1455f4 | 3018 | { |
bb192ed9 | 3019 | struct slab *partial_slab; |
fc1455f4 | 3020 | |
bb192ed9 | 3021 | partial_slab = slub_percpu_partial(c); |
fc1455f4 VB |
3022 | c->partial = NULL; |
3023 | ||
bb192ed9 | 3024 | if (partial_slab) |
21316fdc | 3025 | __put_partials(s, partial_slab); |
49e22585 CL |
3026 | } |
3027 | ||
3028 | /* | |
31bda717 | 3029 | * Put a slab into a partial slab slot if available. |
49e22585 CL |
3030 | * |
3031 | * If we did not find a slot then simply move all the partials to the | |
3032 | * per node partial list. | |
3033 | */ | |
bb192ed9 | 3034 | static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain) |
49e22585 | 3035 | { |
bb192ed9 | 3036 | struct slab *oldslab; |
21316fdc | 3037 | struct slab *slab_to_put = NULL; |
e0a043aa | 3038 | unsigned long flags; |
bb192ed9 | 3039 | int slabs = 0; |
49e22585 | 3040 | |
bd0e7491 | 3041 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
49e22585 | 3042 | |
bb192ed9 | 3043 | oldslab = this_cpu_read(s->cpu_slab->partial); |
e0a043aa | 3044 | |
bb192ed9 VB |
3045 | if (oldslab) { |
3046 | if (drain && oldslab->slabs >= s->cpu_partial_slabs) { | |
e0a043aa VB |
3047 | /* |
3048 | * Partial array is full. Move the existing set to the | |
3049 | * per node partial list. Postpone the actual unfreezing | |
3050 | * outside of the critical section. | |
3051 | */ | |
21316fdc | 3052 | slab_to_put = oldslab; |
bb192ed9 | 3053 | oldslab = NULL; |
e0a043aa | 3054 | } else { |
bb192ed9 | 3055 | slabs = oldslab->slabs; |
49e22585 | 3056 | } |
e0a043aa | 3057 | } |
49e22585 | 3058 | |
bb192ed9 | 3059 | slabs++; |
49e22585 | 3060 | |
bb192ed9 VB |
3061 | slab->slabs = slabs; |
3062 | slab->next = oldslab; | |
49e22585 | 3063 | |
bb192ed9 | 3064 | this_cpu_write(s->cpu_slab->partial, slab); |
e0a043aa | 3065 | |
bd0e7491 | 3066 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
e0a043aa | 3067 | |
21316fdc CZ |
3068 | if (slab_to_put) { |
3069 | __put_partials(s, slab_to_put); | |
e0a043aa VB |
3070 | stat(s, CPU_PARTIAL_DRAIN); |
3071 | } | |
49e22585 CL |
3072 | } |
3073 | ||
e0a043aa VB |
3074 | #else /* CONFIG_SLUB_CPU_PARTIAL */ |
3075 | ||
21316fdc CZ |
3076 | static inline void put_partials(struct kmem_cache *s) { } |
3077 | static inline void put_partials_cpu(struct kmem_cache *s, | |
3078 | struct kmem_cache_cpu *c) { } | |
e0a043aa VB |
3079 | |
3080 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ | |
3081 | ||
dfb4f096 | 3082 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 3083 | { |
5a836bf6 | 3084 | unsigned long flags; |
bb192ed9 | 3085 | struct slab *slab; |
5a836bf6 SAS |
3086 | void *freelist; |
3087 | ||
bd0e7491 | 3088 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
5a836bf6 | 3089 | |
bb192ed9 | 3090 | slab = c->slab; |
5a836bf6 | 3091 | freelist = c->freelist; |
c17dda40 | 3092 | |
bb192ed9 | 3093 | c->slab = NULL; |
a019d201 | 3094 | c->freelist = NULL; |
c17dda40 | 3095 | c->tid = next_tid(c->tid); |
a019d201 | 3096 | |
bd0e7491 | 3097 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
a019d201 | 3098 | |
bb192ed9 VB |
3099 | if (slab) { |
3100 | deactivate_slab(s, slab, freelist); | |
5a836bf6 SAS |
3101 | stat(s, CPUSLAB_FLUSH); |
3102 | } | |
81819f0f CL |
3103 | } |
3104 | ||
0c710013 | 3105 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 3106 | { |
9dfc6e68 | 3107 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
08beb547 | 3108 | void *freelist = c->freelist; |
bb192ed9 | 3109 | struct slab *slab = c->slab; |
81819f0f | 3110 | |
bb192ed9 | 3111 | c->slab = NULL; |
08beb547 VB |
3112 | c->freelist = NULL; |
3113 | c->tid = next_tid(c->tid); | |
3114 | ||
bb192ed9 VB |
3115 | if (slab) { |
3116 | deactivate_slab(s, slab, freelist); | |
08beb547 VB |
3117 | stat(s, CPUSLAB_FLUSH); |
3118 | } | |
49e22585 | 3119 | |
21316fdc | 3120 | put_partials_cpu(s, c); |
81819f0f CL |
3121 | } |
3122 | ||
5a836bf6 SAS |
3123 | struct slub_flush_work { |
3124 | struct work_struct work; | |
3125 | struct kmem_cache *s; | |
3126 | bool skip; | |
3127 | }; | |
3128 | ||
fc1455f4 VB |
3129 | /* |
3130 | * Flush cpu slab. | |
3131 | * | |
5a836bf6 | 3132 | * Called from CPU work handler with migration disabled. |
fc1455f4 | 3133 | */ |
5a836bf6 | 3134 | static void flush_cpu_slab(struct work_struct *w) |
81819f0f | 3135 | { |
5a836bf6 SAS |
3136 | struct kmem_cache *s; |
3137 | struct kmem_cache_cpu *c; | |
3138 | struct slub_flush_work *sfw; | |
3139 | ||
3140 | sfw = container_of(w, struct slub_flush_work, work); | |
3141 | ||
3142 | s = sfw->s; | |
3143 | c = this_cpu_ptr(s->cpu_slab); | |
fc1455f4 | 3144 | |
bb192ed9 | 3145 | if (c->slab) |
fc1455f4 | 3146 | flush_slab(s, c); |
81819f0f | 3147 | |
21316fdc | 3148 | put_partials(s); |
81819f0f CL |
3149 | } |
3150 | ||
5a836bf6 | 3151 | static bool has_cpu_slab(int cpu, struct kmem_cache *s) |
a8364d55 | 3152 | { |
a8364d55 GBY |
3153 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
3154 | ||
bb192ed9 | 3155 | return c->slab || slub_percpu_partial(c); |
a8364d55 GBY |
3156 | } |
3157 | ||
5a836bf6 SAS |
3158 | static DEFINE_MUTEX(flush_lock); |
3159 | static DEFINE_PER_CPU(struct slub_flush_work, slub_flush); | |
3160 | ||
3161 | static void flush_all_cpus_locked(struct kmem_cache *s) | |
3162 | { | |
3163 | struct slub_flush_work *sfw; | |
3164 | unsigned int cpu; | |
3165 | ||
3166 | lockdep_assert_cpus_held(); | |
3167 | mutex_lock(&flush_lock); | |
3168 | ||
3169 | for_each_online_cpu(cpu) { | |
3170 | sfw = &per_cpu(slub_flush, cpu); | |
3171 | if (!has_cpu_slab(cpu, s)) { | |
3172 | sfw->skip = true; | |
3173 | continue; | |
3174 | } | |
3175 | INIT_WORK(&sfw->work, flush_cpu_slab); | |
3176 | sfw->skip = false; | |
3177 | sfw->s = s; | |
e45cc288 | 3178 | queue_work_on(cpu, flushwq, &sfw->work); |
5a836bf6 SAS |
3179 | } |
3180 | ||
3181 | for_each_online_cpu(cpu) { | |
3182 | sfw = &per_cpu(slub_flush, cpu); | |
3183 | if (sfw->skip) | |
3184 | continue; | |
3185 | flush_work(&sfw->work); | |
3186 | } | |
3187 | ||
3188 | mutex_unlock(&flush_lock); | |
3189 | } | |
3190 | ||
81819f0f CL |
3191 | static void flush_all(struct kmem_cache *s) |
3192 | { | |
5a836bf6 SAS |
3193 | cpus_read_lock(); |
3194 | flush_all_cpus_locked(s); | |
3195 | cpus_read_unlock(); | |
81819f0f CL |
3196 | } |
3197 | ||
a96a87bf SAS |
3198 | /* |
3199 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
3200 | * necessary. | |
3201 | */ | |
3202 | static int slub_cpu_dead(unsigned int cpu) | |
3203 | { | |
3204 | struct kmem_cache *s; | |
a96a87bf SAS |
3205 | |
3206 | mutex_lock(&slab_mutex); | |
0e7ac738 | 3207 | list_for_each_entry(s, &slab_caches, list) |
a96a87bf | 3208 | __flush_cpu_slab(s, cpu); |
a96a87bf SAS |
3209 | mutex_unlock(&slab_mutex); |
3210 | return 0; | |
3211 | } | |
3212 | ||
0af8489b VB |
3213 | #else /* CONFIG_SLUB_TINY */ |
3214 | static inline void flush_all_cpus_locked(struct kmem_cache *s) { } | |
3215 | static inline void flush_all(struct kmem_cache *s) { } | |
3216 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) { } | |
3217 | static inline int slub_cpu_dead(unsigned int cpu) { return 0; } | |
3218 | #endif /* CONFIG_SLUB_TINY */ | |
3219 | ||
dfb4f096 CL |
3220 | /* |
3221 | * Check if the objects in a per cpu structure fit numa | |
3222 | * locality expectations. | |
3223 | */ | |
bb192ed9 | 3224 | static inline int node_match(struct slab *slab, int node) |
dfb4f096 CL |
3225 | { |
3226 | #ifdef CONFIG_NUMA | |
bb192ed9 | 3227 | if (node != NUMA_NO_NODE && slab_nid(slab) != node) |
dfb4f096 CL |
3228 | return 0; |
3229 | #endif | |
3230 | return 1; | |
3231 | } | |
3232 | ||
9a02d699 | 3233 | #ifdef CONFIG_SLUB_DEBUG |
bb192ed9 | 3234 | static int count_free(struct slab *slab) |
781b2ba6 | 3235 | { |
bb192ed9 | 3236 | return slab->objects - slab->inuse; |
781b2ba6 PE |
3237 | } |
3238 | ||
9a02d699 DR |
3239 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
3240 | { | |
3241 | return atomic_long_read(&n->total_objects); | |
3242 | } | |
a579b056 VB |
3243 | |
3244 | /* Supports checking bulk free of a constructed freelist */ | |
fa9b88e4 VB |
3245 | static inline bool free_debug_processing(struct kmem_cache *s, |
3246 | struct slab *slab, void *head, void *tail, int *bulk_cnt, | |
3247 | unsigned long addr, depot_stack_handle_t handle) | |
a579b056 | 3248 | { |
fa9b88e4 | 3249 | bool checks_ok = false; |
a579b056 VB |
3250 | void *object = head; |
3251 | int cnt = 0; | |
a579b056 VB |
3252 | |
3253 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
3254 | if (!check_slab(s, slab)) | |
3255 | goto out; | |
3256 | } | |
3257 | ||
fa9b88e4 | 3258 | if (slab->inuse < *bulk_cnt) { |
c7323a5a | 3259 | slab_err(s, slab, "Slab has %d allocated objects but %d are to be freed\n", |
fa9b88e4 | 3260 | slab->inuse, *bulk_cnt); |
c7323a5a VB |
3261 | goto out; |
3262 | } | |
3263 | ||
a579b056 | 3264 | next_object: |
c7323a5a | 3265 | |
fa9b88e4 | 3266 | if (++cnt > *bulk_cnt) |
c7323a5a | 3267 | goto out_cnt; |
a579b056 VB |
3268 | |
3269 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
3270 | if (!free_consistency_checks(s, slab, object, addr)) | |
3271 | goto out; | |
3272 | } | |
3273 | ||
3274 | if (s->flags & SLAB_STORE_USER) | |
3275 | set_track_update(s, object, TRACK_FREE, addr, handle); | |
3276 | trace(s, slab, object, 0); | |
3277 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ | |
3278 | init_object(s, object, SLUB_RED_INACTIVE); | |
3279 | ||
3280 | /* Reached end of constructed freelist yet? */ | |
3281 | if (object != tail) { | |
3282 | object = get_freepointer(s, object); | |
3283 | goto next_object; | |
3284 | } | |
c7323a5a | 3285 | checks_ok = true; |
a579b056 | 3286 | |
c7323a5a | 3287 | out_cnt: |
fa9b88e4 | 3288 | if (cnt != *bulk_cnt) { |
c7323a5a | 3289 | slab_err(s, slab, "Bulk free expected %d objects but found %d\n", |
fa9b88e4 VB |
3290 | *bulk_cnt, cnt); |
3291 | *bulk_cnt = cnt; | |
c7323a5a VB |
3292 | } |
3293 | ||
fa9b88e4 | 3294 | out: |
c7323a5a VB |
3295 | |
3296 | if (!checks_ok) | |
a579b056 | 3297 | slab_fix(s, "Object at 0x%p not freed", object); |
c7323a5a | 3298 | |
fa9b88e4 | 3299 | return checks_ok; |
a579b056 | 3300 | } |
9a02d699 DR |
3301 | #endif /* CONFIG_SLUB_DEBUG */ |
3302 | ||
b1a413a3 | 3303 | #if defined(CONFIG_SLUB_DEBUG) || defined(SLAB_SUPPORTS_SYSFS) |
781b2ba6 | 3304 | static unsigned long count_partial(struct kmem_cache_node *n, |
bb192ed9 | 3305 | int (*get_count)(struct slab *)) |
781b2ba6 PE |
3306 | { |
3307 | unsigned long flags; | |
3308 | unsigned long x = 0; | |
bb192ed9 | 3309 | struct slab *slab; |
781b2ba6 PE |
3310 | |
3311 | spin_lock_irqsave(&n->list_lock, flags); | |
bb192ed9 VB |
3312 | list_for_each_entry(slab, &n->partial, slab_list) |
3313 | x += get_count(slab); | |
781b2ba6 PE |
3314 | spin_unlock_irqrestore(&n->list_lock, flags); |
3315 | return x; | |
3316 | } | |
b1a413a3 | 3317 | #endif /* CONFIG_SLUB_DEBUG || SLAB_SUPPORTS_SYSFS */ |
26c02cf0 | 3318 | |
56d5a2b9 | 3319 | #ifdef CONFIG_SLUB_DEBUG |
046f4c69 JW |
3320 | #define MAX_PARTIAL_TO_SCAN 10000 |
3321 | ||
3322 | static unsigned long count_partial_free_approx(struct kmem_cache_node *n) | |
3323 | { | |
3324 | unsigned long flags; | |
3325 | unsigned long x = 0; | |
3326 | struct slab *slab; | |
3327 | ||
3328 | spin_lock_irqsave(&n->list_lock, flags); | |
3329 | if (n->nr_partial <= MAX_PARTIAL_TO_SCAN) { | |
3330 | list_for_each_entry(slab, &n->partial, slab_list) | |
3331 | x += slab->objects - slab->inuse; | |
3332 | } else { | |
3333 | /* | |
3334 | * For a long list, approximate the total count of objects in | |
3335 | * it to meet the limit on the number of slabs to scan. | |
3336 | * Scan from both the list's head and tail for better accuracy. | |
3337 | */ | |
3338 | unsigned long scanned = 0; | |
3339 | ||
3340 | list_for_each_entry(slab, &n->partial, slab_list) { | |
3341 | x += slab->objects - slab->inuse; | |
3342 | if (++scanned == MAX_PARTIAL_TO_SCAN / 2) | |
3343 | break; | |
3344 | } | |
3345 | list_for_each_entry_reverse(slab, &n->partial, slab_list) { | |
3346 | x += slab->objects - slab->inuse; | |
3347 | if (++scanned == MAX_PARTIAL_TO_SCAN) | |
3348 | break; | |
3349 | } | |
3350 | x = mult_frac(x, n->nr_partial, scanned); | |
3351 | x = min(x, node_nr_objs(n)); | |
3352 | } | |
3353 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3354 | return x; | |
3355 | } | |
3356 | ||
781b2ba6 PE |
3357 | static noinline void |
3358 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
3359 | { | |
9a02d699 DR |
3360 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
3361 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 3362 | int node; |
fa45dc25 | 3363 | struct kmem_cache_node *n; |
781b2ba6 | 3364 | |
9a02d699 DR |
3365 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
3366 | return; | |
3367 | ||
5b3810e5 VB |
3368 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
3369 | nid, gfpflags, &gfpflags); | |
19af27af | 3370 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
3371 | s->name, s->object_size, s->size, oo_order(s->oo), |
3372 | oo_order(s->min)); | |
781b2ba6 | 3373 | |
3b0efdfa | 3374 | if (oo_order(s->min) > get_order(s->object_size)) |
671776b3 | 3375 | pr_warn(" %s debugging increased min order, use slab_debug=O to disable.\n", |
f9f58285 | 3376 | s->name); |
fa5ec8a1 | 3377 | |
fa45dc25 | 3378 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
3379 | unsigned long nr_slabs; |
3380 | unsigned long nr_objs; | |
3381 | unsigned long nr_free; | |
3382 | ||
b3d8a8e8 | 3383 | nr_free = count_partial_free_approx(n); |
26c02cf0 AB |
3384 | nr_slabs = node_nr_slabs(n); |
3385 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 3386 | |
f9f58285 | 3387 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
3388 | node, nr_slabs, nr_objs, nr_free); |
3389 | } | |
3390 | } | |
56d5a2b9 VB |
3391 | #else /* CONFIG_SLUB_DEBUG */ |
3392 | static inline void | |
3393 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) { } | |
3394 | #endif | |
781b2ba6 | 3395 | |
01b34d16 | 3396 | static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags) |
072bb0aa | 3397 | { |
01b34d16 | 3398 | if (unlikely(slab_test_pfmemalloc(slab))) |
0b303fb4 VB |
3399 | return gfp_pfmemalloc_allowed(gfpflags); |
3400 | ||
3401 | return true; | |
3402 | } | |
3403 | ||
0af8489b | 3404 | #ifndef CONFIG_SLUB_TINY |
6801be4f PZ |
3405 | static inline bool |
3406 | __update_cpu_freelist_fast(struct kmem_cache *s, | |
3407 | void *freelist_old, void *freelist_new, | |
3408 | unsigned long tid) | |
3409 | { | |
3410 | freelist_aba_t old = { .freelist = freelist_old, .counter = tid }; | |
3411 | freelist_aba_t new = { .freelist = freelist_new, .counter = next_tid(tid) }; | |
3412 | ||
3413 | return this_cpu_try_cmpxchg_freelist(s->cpu_slab->freelist_tid.full, | |
3414 | &old.full, new.full); | |
3415 | } | |
3416 | ||
213eeb9f | 3417 | /* |
c2092c12 VB |
3418 | * Check the slab->freelist and either transfer the freelist to the |
3419 | * per cpu freelist or deactivate the slab. | |
213eeb9f | 3420 | * |
c2092c12 | 3421 | * The slab is still frozen if the return value is not NULL. |
213eeb9f | 3422 | * |
c2092c12 | 3423 | * If this function returns NULL then the slab has been unfrozen. |
213eeb9f | 3424 | */ |
bb192ed9 | 3425 | static inline void *get_freelist(struct kmem_cache *s, struct slab *slab) |
213eeb9f | 3426 | { |
bb192ed9 | 3427 | struct slab new; |
213eeb9f CL |
3428 | unsigned long counters; |
3429 | void *freelist; | |
3430 | ||
bd0e7491 VB |
3431 | lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock)); |
3432 | ||
213eeb9f | 3433 | do { |
bb192ed9 VB |
3434 | freelist = slab->freelist; |
3435 | counters = slab->counters; | |
6faa6833 | 3436 | |
213eeb9f | 3437 | new.counters = counters; |
213eeb9f | 3438 | |
bb192ed9 | 3439 | new.inuse = slab->objects; |
213eeb9f CL |
3440 | new.frozen = freelist != NULL; |
3441 | ||
6801be4f | 3442 | } while (!__slab_update_freelist(s, slab, |
213eeb9f CL |
3443 | freelist, counters, |
3444 | NULL, new.counters, | |
3445 | "get_freelist")); | |
3446 | ||
3447 | return freelist; | |
3448 | } | |
3449 | ||
213094b5 CZ |
3450 | /* |
3451 | * Freeze the partial slab and return the pointer to the freelist. | |
3452 | */ | |
3453 | static inline void *freeze_slab(struct kmem_cache *s, struct slab *slab) | |
3454 | { | |
3455 | struct slab new; | |
3456 | unsigned long counters; | |
3457 | void *freelist; | |
3458 | ||
3459 | do { | |
3460 | freelist = slab->freelist; | |
3461 | counters = slab->counters; | |
3462 | ||
3463 | new.counters = counters; | |
3464 | VM_BUG_ON(new.frozen); | |
3465 | ||
3466 | new.inuse = slab->objects; | |
3467 | new.frozen = 1; | |
3468 | ||
3469 | } while (!slab_update_freelist(s, slab, | |
3470 | freelist, counters, | |
3471 | NULL, new.counters, | |
3472 | "freeze_slab")); | |
3473 | ||
3474 | return freelist; | |
3475 | } | |
3476 | ||
81819f0f | 3477 | /* |
894b8788 CL |
3478 | * Slow path. The lockless freelist is empty or we need to perform |
3479 | * debugging duties. | |
3480 | * | |
894b8788 CL |
3481 | * Processing is still very fast if new objects have been freed to the |
3482 | * regular freelist. In that case we simply take over the regular freelist | |
3483 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 3484 | * |
894b8788 CL |
3485 | * If that is not working then we fall back to the partial lists. We take the |
3486 | * first element of the freelist as the object to allocate now and move the | |
3487 | * rest of the freelist to the lockless freelist. | |
81819f0f | 3488 | * |
894b8788 | 3489 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
3490 | * we need to allocate a new slab. This is the slowest path since it involves |
3491 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 | 3492 | * |
e500059b | 3493 | * Version of __slab_alloc to use when we know that preemption is |
a380a3c7 | 3494 | * already disabled (which is the case for bulk allocation). |
81819f0f | 3495 | */ |
a380a3c7 | 3496 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
6edf2576 | 3497 | unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size) |
81819f0f | 3498 | { |
6faa6833 | 3499 | void *freelist; |
bb192ed9 | 3500 | struct slab *slab; |
e500059b | 3501 | unsigned long flags; |
6edf2576 | 3502 | struct partial_context pc; |
9198ffbd | 3503 | bool try_thisnode = true; |
81819f0f | 3504 | |
9f986d99 AW |
3505 | stat(s, ALLOC_SLOWPATH); |
3506 | ||
c2092c12 | 3507 | reread_slab: |
0b303fb4 | 3508 | |
bb192ed9 VB |
3509 | slab = READ_ONCE(c->slab); |
3510 | if (!slab) { | |
0715e6c5 VB |
3511 | /* |
3512 | * if the node is not online or has no normal memory, just | |
3513 | * ignore the node constraint | |
3514 | */ | |
3515 | if (unlikely(node != NUMA_NO_NODE && | |
7e1fa93d | 3516 | !node_isset(node, slab_nodes))) |
0715e6c5 | 3517 | node = NUMA_NO_NODE; |
81819f0f | 3518 | goto new_slab; |
0715e6c5 | 3519 | } |
6faa6833 | 3520 | |
bb192ed9 | 3521 | if (unlikely(!node_match(slab, node))) { |
0715e6c5 VB |
3522 | /* |
3523 | * same as above but node_match() being false already | |
3524 | * implies node != NUMA_NO_NODE | |
3525 | */ | |
7e1fa93d | 3526 | if (!node_isset(node, slab_nodes)) { |
0715e6c5 | 3527 | node = NUMA_NO_NODE; |
0715e6c5 | 3528 | } else { |
a561ce00 | 3529 | stat(s, ALLOC_NODE_MISMATCH); |
0b303fb4 | 3530 | goto deactivate_slab; |
a561ce00 | 3531 | } |
fc59c053 | 3532 | } |
6446faa2 | 3533 | |
072bb0aa MG |
3534 | /* |
3535 | * By rights, we should be searching for a slab page that was | |
3536 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
3537 | * information when the page leaves the per-cpu allocator | |
3538 | */ | |
bb192ed9 | 3539 | if (unlikely(!pfmemalloc_match(slab, gfpflags))) |
0b303fb4 | 3540 | goto deactivate_slab; |
072bb0aa | 3541 | |
c2092c12 | 3542 | /* must check again c->slab in case we got preempted and it changed */ |
bd0e7491 | 3543 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 3544 | if (unlikely(slab != c->slab)) { |
bd0e7491 | 3545 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
c2092c12 | 3546 | goto reread_slab; |
0b303fb4 | 3547 | } |
6faa6833 CL |
3548 | freelist = c->freelist; |
3549 | if (freelist) | |
73736e03 | 3550 | goto load_freelist; |
03e404af | 3551 | |
bb192ed9 | 3552 | freelist = get_freelist(s, slab); |
6446faa2 | 3553 | |
6faa6833 | 3554 | if (!freelist) { |
bb192ed9 | 3555 | c->slab = NULL; |
eeaa345e | 3556 | c->tid = next_tid(c->tid); |
bd0e7491 | 3557 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
03e404af | 3558 | stat(s, DEACTIVATE_BYPASS); |
fc59c053 | 3559 | goto new_slab; |
03e404af | 3560 | } |
6446faa2 | 3561 | |
84e554e6 | 3562 | stat(s, ALLOC_REFILL); |
6446faa2 | 3563 | |
894b8788 | 3564 | load_freelist: |
0b303fb4 | 3565 | |
bd0e7491 | 3566 | lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock)); |
0b303fb4 | 3567 | |
507effea CL |
3568 | /* |
3569 | * freelist is pointing to the list of objects to be used. | |
c2092c12 VB |
3570 | * slab is pointing to the slab from which the objects are obtained. |
3571 | * That slab must be frozen for per cpu allocations to work. | |
507effea | 3572 | */ |
bb192ed9 | 3573 | VM_BUG_ON(!c->slab->frozen); |
6faa6833 | 3574 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 3575 | c->tid = next_tid(c->tid); |
bd0e7491 | 3576 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
6faa6833 | 3577 | return freelist; |
81819f0f | 3578 | |
0b303fb4 VB |
3579 | deactivate_slab: |
3580 | ||
bd0e7491 | 3581 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 3582 | if (slab != c->slab) { |
bd0e7491 | 3583 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
c2092c12 | 3584 | goto reread_slab; |
0b303fb4 | 3585 | } |
a019d201 | 3586 | freelist = c->freelist; |
bb192ed9 | 3587 | c->slab = NULL; |
a019d201 | 3588 | c->freelist = NULL; |
eeaa345e | 3589 | c->tid = next_tid(c->tid); |
bd0e7491 | 3590 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
bb192ed9 | 3591 | deactivate_slab(s, slab, freelist); |
0b303fb4 | 3592 | |
81819f0f | 3593 | new_slab: |
2cfb7455 | 3594 | |
8cd3fa42 CZ |
3595 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
3596 | while (slub_percpu_partial(c)) { | |
bd0e7491 | 3597 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 3598 | if (unlikely(c->slab)) { |
bd0e7491 | 3599 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
c2092c12 | 3600 | goto reread_slab; |
fa417ab7 | 3601 | } |
4b1f449d | 3602 | if (unlikely(!slub_percpu_partial(c))) { |
bd0e7491 | 3603 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
25c00c50 VB |
3604 | /* we were preempted and partial list got empty */ |
3605 | goto new_objects; | |
4b1f449d | 3606 | } |
fa417ab7 | 3607 | |
8cd3fa42 | 3608 | slab = slub_percpu_partial(c); |
bb192ed9 | 3609 | slub_set_percpu_partial(c, slab); |
8cd3fa42 | 3610 | |
90b1e566 CZ |
3611 | if (likely(node_match(slab, node) && |
3612 | pfmemalloc_match(slab, gfpflags))) { | |
3613 | c->slab = slab; | |
3614 | freelist = get_freelist(s, slab); | |
3615 | VM_BUG_ON(!freelist); | |
3616 | stat(s, CPU_PARTIAL_ALLOC); | |
3617 | goto load_freelist; | |
8cd3fa42 CZ |
3618 | } |
3619 | ||
90b1e566 CZ |
3620 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
3621 | ||
3622 | slab->next = NULL; | |
3623 | __put_partials(s, slab); | |
81819f0f | 3624 | } |
8cd3fa42 | 3625 | #endif |
81819f0f | 3626 | |
fa417ab7 VB |
3627 | new_objects: |
3628 | ||
6edf2576 | 3629 | pc.flags = gfpflags; |
9198ffbd CJ |
3630 | /* |
3631 | * When a preferred node is indicated but no __GFP_THISNODE | |
3632 | * | |
3633 | * 1) try to get a partial slab from target node only by having | |
3634 | * __GFP_THISNODE in pc.flags for get_partial() | |
3635 | * 2) if 1) failed, try to allocate a new slab from target node with | |
3636 | * GPF_NOWAIT | __GFP_THISNODE opportunistically | |
3637 | * 3) if 2) failed, retry with original gfpflags which will allow | |
3638 | * get_partial() try partial lists of other nodes before potentially | |
3639 | * allocating new page from other nodes | |
3640 | */ | |
3641 | if (unlikely(node != NUMA_NO_NODE && !(gfpflags & __GFP_THISNODE) | |
3642 | && try_thisnode)) | |
3643 | pc.flags = GFP_NOWAIT | __GFP_THISNODE; | |
3644 | ||
6edf2576 | 3645 | pc.orig_size = orig_size; |
43c4c349 CZ |
3646 | slab = get_partial(s, node, &pc); |
3647 | if (slab) { | |
24c6a097 | 3648 | if (kmem_cache_debug(s)) { |
8cd3fa42 | 3649 | freelist = pc.object; |
24c6a097 CZ |
3650 | /* |
3651 | * For debug caches here we had to go through | |
3652 | * alloc_single_from_partial() so just store the | |
3653 | * tracking info and return the object. | |
3654 | */ | |
3655 | if (s->flags & SLAB_STORE_USER) | |
3656 | set_track(s, freelist, TRACK_ALLOC, addr); | |
3657 | ||
3658 | return freelist; | |
3659 | } | |
3660 | ||
8cd3fa42 | 3661 | freelist = freeze_slab(s, slab); |
24c6a097 CZ |
3662 | goto retry_load_slab; |
3663 | } | |
2a904905 | 3664 | |
25c00c50 | 3665 | slub_put_cpu_ptr(s->cpu_slab); |
9198ffbd | 3666 | slab = new_slab(s, pc.flags, node); |
25c00c50 | 3667 | c = slub_get_cpu_ptr(s->cpu_slab); |
01ad8a7b | 3668 | |
bb192ed9 | 3669 | if (unlikely(!slab)) { |
9198ffbd CJ |
3670 | if (node != NUMA_NO_NODE && !(gfpflags & __GFP_THISNODE) |
3671 | && try_thisnode) { | |
3672 | try_thisnode = false; | |
3673 | goto new_objects; | |
3674 | } | |
9a02d699 | 3675 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 3676 | return NULL; |
81819f0f | 3677 | } |
2cfb7455 | 3678 | |
c7323a5a VB |
3679 | stat(s, ALLOC_SLAB); |
3680 | ||
3681 | if (kmem_cache_debug(s)) { | |
6edf2576 | 3682 | freelist = alloc_single_from_new_slab(s, slab, orig_size); |
c7323a5a VB |
3683 | |
3684 | if (unlikely(!freelist)) | |
3685 | goto new_objects; | |
3686 | ||
3687 | if (s->flags & SLAB_STORE_USER) | |
3688 | set_track(s, freelist, TRACK_ALLOC, addr); | |
3689 | ||
3690 | return freelist; | |
3691 | } | |
3692 | ||
53a0de06 | 3693 | /* |
c2092c12 | 3694 | * No other reference to the slab yet so we can |
53a0de06 VB |
3695 | * muck around with it freely without cmpxchg |
3696 | */ | |
bb192ed9 VB |
3697 | freelist = slab->freelist; |
3698 | slab->freelist = NULL; | |
c7323a5a VB |
3699 | slab->inuse = slab->objects; |
3700 | slab->frozen = 1; | |
53a0de06 | 3701 | |
c7323a5a | 3702 | inc_slabs_node(s, slab_nid(slab), slab->objects); |
53a0de06 | 3703 | |
c7323a5a | 3704 | if (unlikely(!pfmemalloc_match(slab, gfpflags))) { |
1572df7c VB |
3705 | /* |
3706 | * For !pfmemalloc_match() case we don't load freelist so that | |
3707 | * we don't make further mismatched allocations easier. | |
3708 | */ | |
c7323a5a VB |
3709 | deactivate_slab(s, slab, get_freepointer(s, freelist)); |
3710 | return freelist; | |
3711 | } | |
1572df7c | 3712 | |
c2092c12 | 3713 | retry_load_slab: |
cfdf836e | 3714 | |
bd0e7491 | 3715 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 3716 | if (unlikely(c->slab)) { |
cfdf836e | 3717 | void *flush_freelist = c->freelist; |
bb192ed9 | 3718 | struct slab *flush_slab = c->slab; |
cfdf836e | 3719 | |
bb192ed9 | 3720 | c->slab = NULL; |
cfdf836e VB |
3721 | c->freelist = NULL; |
3722 | c->tid = next_tid(c->tid); | |
3723 | ||
bd0e7491 | 3724 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
cfdf836e | 3725 | |
bb192ed9 | 3726 | deactivate_slab(s, flush_slab, flush_freelist); |
cfdf836e VB |
3727 | |
3728 | stat(s, CPUSLAB_FLUSH); | |
3729 | ||
c2092c12 | 3730 | goto retry_load_slab; |
cfdf836e | 3731 | } |
bb192ed9 | 3732 | c->slab = slab; |
3f2b77e3 | 3733 | |
1572df7c | 3734 | goto load_freelist; |
894b8788 CL |
3735 | } |
3736 | ||
a380a3c7 | 3737 | /* |
e500059b VB |
3738 | * A wrapper for ___slab_alloc() for contexts where preemption is not yet |
3739 | * disabled. Compensates for possible cpu changes by refetching the per cpu area | |
3740 | * pointer. | |
a380a3c7 CL |
3741 | */ |
3742 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
6edf2576 | 3743 | unsigned long addr, struct kmem_cache_cpu *c, unsigned int orig_size) |
a380a3c7 CL |
3744 | { |
3745 | void *p; | |
a380a3c7 | 3746 | |
e500059b | 3747 | #ifdef CONFIG_PREEMPT_COUNT |
a380a3c7 CL |
3748 | /* |
3749 | * We may have been preempted and rescheduled on a different | |
e500059b | 3750 | * cpu before disabling preemption. Need to reload cpu area |
a380a3c7 CL |
3751 | * pointer. |
3752 | */ | |
25c00c50 | 3753 | c = slub_get_cpu_ptr(s->cpu_slab); |
a380a3c7 CL |
3754 | #endif |
3755 | ||
6edf2576 | 3756 | p = ___slab_alloc(s, gfpflags, node, addr, c, orig_size); |
e500059b | 3757 | #ifdef CONFIG_PREEMPT_COUNT |
25c00c50 | 3758 | slub_put_cpu_ptr(s->cpu_slab); |
e500059b | 3759 | #endif |
a380a3c7 CL |
3760 | return p; |
3761 | } | |
3762 | ||
56d5a2b9 | 3763 | static __always_inline void *__slab_alloc_node(struct kmem_cache *s, |
b89fb5ef | 3764 | gfp_t gfpflags, int node, unsigned long addr, size_t orig_size) |
894b8788 | 3765 | { |
dfb4f096 | 3766 | struct kmem_cache_cpu *c; |
bb192ed9 | 3767 | struct slab *slab; |
8a5ec0ba | 3768 | unsigned long tid; |
56d5a2b9 | 3769 | void *object; |
b89fb5ef | 3770 | |
8a5ec0ba | 3771 | redo: |
8a5ec0ba CL |
3772 | /* |
3773 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
3774 | * enabled. We may switch back and forth between cpus while | |
3775 | * reading from one cpu area. That does not matter as long | |
3776 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 3777 | * |
9b4bc85a VB |
3778 | * We must guarantee that tid and kmem_cache_cpu are retrieved on the |
3779 | * same cpu. We read first the kmem_cache_cpu pointer and use it to read | |
3780 | * the tid. If we are preempted and switched to another cpu between the | |
3781 | * two reads, it's OK as the two are still associated with the same cpu | |
3782 | * and cmpxchg later will validate the cpu. | |
8a5ec0ba | 3783 | */ |
9b4bc85a VB |
3784 | c = raw_cpu_ptr(s->cpu_slab); |
3785 | tid = READ_ONCE(c->tid); | |
9aabf810 JK |
3786 | |
3787 | /* | |
3788 | * Irqless object alloc/free algorithm used here depends on sequence | |
3789 | * of fetching cpu_slab's data. tid should be fetched before anything | |
c2092c12 | 3790 | * on c to guarantee that object and slab associated with previous tid |
9aabf810 | 3791 | * won't be used with current tid. If we fetch tid first, object and |
c2092c12 | 3792 | * slab could be one associated with next tid and our alloc/free |
9aabf810 JK |
3793 | * request will be failed. In this case, we will retry. So, no problem. |
3794 | */ | |
3795 | barrier(); | |
8a5ec0ba | 3796 | |
8a5ec0ba CL |
3797 | /* |
3798 | * The transaction ids are globally unique per cpu and per operation on | |
3799 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
3800 | * occurs on the right processor and that there was no operation on the | |
3801 | * linked list in between. | |
3802 | */ | |
8a5ec0ba | 3803 | |
9dfc6e68 | 3804 | object = c->freelist; |
bb192ed9 | 3805 | slab = c->slab; |
1f04b07d TG |
3806 | |
3807 | if (!USE_LOCKLESS_FAST_PATH() || | |
bb192ed9 | 3808 | unlikely(!object || !slab || !node_match(slab, node))) { |
6edf2576 | 3809 | object = __slab_alloc(s, gfpflags, node, addr, c, orig_size); |
8eae1492 | 3810 | } else { |
0ad9500e ED |
3811 | void *next_object = get_freepointer_safe(s, object); |
3812 | ||
8a5ec0ba | 3813 | /* |
25985edc | 3814 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
3815 | * operation and if we are on the right processor. |
3816 | * | |
d0e0ac97 CG |
3817 | * The cmpxchg does the following atomically (without lock |
3818 | * semantics!) | |
8a5ec0ba CL |
3819 | * 1. Relocate first pointer to the current per cpu area. |
3820 | * 2. Verify that tid and freelist have not been changed | |
3821 | * 3. If they were not changed replace tid and freelist | |
3822 | * | |
d0e0ac97 CG |
3823 | * Since this is without lock semantics the protection is only |
3824 | * against code executing on this cpu *not* from access by | |
3825 | * other cpus. | |
8a5ec0ba | 3826 | */ |
6801be4f | 3827 | if (unlikely(!__update_cpu_freelist_fast(s, object, next_object, tid))) { |
8a5ec0ba CL |
3828 | note_cmpxchg_failure("slab_alloc", s, tid); |
3829 | goto redo; | |
3830 | } | |
0ad9500e | 3831 | prefetch_freepointer(s, next_object); |
84e554e6 | 3832 | stat(s, ALLOC_FASTPATH); |
894b8788 | 3833 | } |
0f181f9f | 3834 | |
56d5a2b9 VB |
3835 | return object; |
3836 | } | |
0af8489b VB |
3837 | #else /* CONFIG_SLUB_TINY */ |
3838 | static void *__slab_alloc_node(struct kmem_cache *s, | |
3839 | gfp_t gfpflags, int node, unsigned long addr, size_t orig_size) | |
3840 | { | |
3841 | struct partial_context pc; | |
3842 | struct slab *slab; | |
3843 | void *object; | |
3844 | ||
3845 | pc.flags = gfpflags; | |
0af8489b | 3846 | pc.orig_size = orig_size; |
43c4c349 | 3847 | slab = get_partial(s, node, &pc); |
0af8489b | 3848 | |
43c4c349 CZ |
3849 | if (slab) |
3850 | return pc.object; | |
0af8489b VB |
3851 | |
3852 | slab = new_slab(s, gfpflags, node); | |
3853 | if (unlikely(!slab)) { | |
3854 | slab_out_of_memory(s, gfpflags, node); | |
3855 | return NULL; | |
3856 | } | |
3857 | ||
3858 | object = alloc_single_from_new_slab(s, slab, orig_size); | |
3859 | ||
3860 | return object; | |
3861 | } | |
3862 | #endif /* CONFIG_SLUB_TINY */ | |
56d5a2b9 VB |
3863 | |
3864 | /* | |
3865 | * If the object has been wiped upon free, make sure it's fully initialized by | |
3866 | * zeroing out freelist pointer. | |
3867 | */ | |
3868 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
3869 | void *obj) | |
3870 | { | |
8f828aa4 NB |
3871 | if (unlikely(slab_want_init_on_free(s)) && obj && |
3872 | !freeptr_outside_object(s)) | |
56d5a2b9 VB |
3873 | memset((void *)((char *)kasan_reset_tag(obj) + s->offset), |
3874 | 0, sizeof(void *)); | |
3875 | } | |
3876 | ||
6011be59 VB |
3877 | noinline int should_failslab(struct kmem_cache *s, gfp_t gfpflags) |
3878 | { | |
3879 | if (__should_failslab(s, gfpflags)) | |
3880 | return -ENOMEM; | |
3881 | return 0; | |
3882 | } | |
3883 | ALLOW_ERROR_INJECTION(should_failslab, ERRNO); | |
3884 | ||
3450a0e5 | 3885 | static __fastpath_inline |
9f9796b4 | 3886 | struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) |
6011be59 VB |
3887 | { |
3888 | flags &= gfp_allowed_mask; | |
3889 | ||
3890 | might_alloc(flags); | |
3891 | ||
3450a0e5 | 3892 | if (unlikely(should_failslab(s, flags))) |
6011be59 VB |
3893 | return NULL; |
3894 | ||
6011be59 VB |
3895 | return s; |
3896 | } | |
3897 | ||
3450a0e5 | 3898 | static __fastpath_inline |
9f9796b4 | 3899 | bool slab_post_alloc_hook(struct kmem_cache *s, struct list_lru *lru, |
3450a0e5 VB |
3900 | gfp_t flags, size_t size, void **p, bool init, |
3901 | unsigned int orig_size) | |
6011be59 VB |
3902 | { |
3903 | unsigned int zero_size = s->object_size; | |
4b873696 | 3904 | struct slabobj_ext *obj_exts; |
6011be59 VB |
3905 | bool kasan_init = init; |
3906 | size_t i; | |
3450a0e5 | 3907 | gfp_t init_flags = flags & gfp_allowed_mask; |
6011be59 VB |
3908 | |
3909 | /* | |
3910 | * For kmalloc object, the allocated memory size(object_size) is likely | |
3911 | * larger than the requested size(orig_size). If redzone check is | |
3912 | * enabled for the extra space, don't zero it, as it will be redzoned | |
3913 | * soon. The redzone operation for this extra space could be seen as a | |
3914 | * replacement of current poisoning under certain debug option, and | |
3915 | * won't break other sanity checks. | |
3916 | */ | |
3917 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER | SLAB_RED_ZONE) && | |
3918 | (s->flags & SLAB_KMALLOC)) | |
3919 | zero_size = orig_size; | |
3920 | ||
3921 | /* | |
671776b3 | 3922 | * When slab_debug is enabled, avoid memory initialization integrated |
6011be59 VB |
3923 | * into KASAN and instead zero out the memory via the memset below with |
3924 | * the proper size. Otherwise, KASAN might overwrite SLUB redzones and | |
3925 | * cause false-positive reports. This does not lead to a performance | |
671776b3 | 3926 | * penalty on production builds, as slab_debug is not intended to be |
6011be59 VB |
3927 | * enabled there. |
3928 | */ | |
3929 | if (__slub_debug_enabled()) | |
3930 | kasan_init = false; | |
3931 | ||
3932 | /* | |
3933 | * As memory initialization might be integrated into KASAN, | |
3934 | * kasan_slab_alloc and initialization memset must be | |
3935 | * kept together to avoid discrepancies in behavior. | |
3936 | * | |
3937 | * As p[i] might get tagged, memset and kmemleak hook come after KASAN. | |
3938 | */ | |
3939 | for (i = 0; i < size; i++) { | |
3450a0e5 | 3940 | p[i] = kasan_slab_alloc(s, p[i], init_flags, kasan_init); |
6011be59 VB |
3941 | if (p[i] && init && (!kasan_init || |
3942 | !kasan_has_integrated_init())) | |
3943 | memset(p[i], 0, zero_size); | |
3944 | kmemleak_alloc_recursive(p[i], s->object_size, 1, | |
3450a0e5 VB |
3945 | s->flags, init_flags); |
3946 | kmsan_slab_alloc(s, p[i], init_flags); | |
4b873696 SB |
3947 | if (need_slab_obj_ext()) { |
3948 | obj_exts = prepare_slab_obj_exts_hook(s, flags, p[i]); | |
3949 | #ifdef CONFIG_MEM_ALLOC_PROFILING | |
3950 | /* | |
3951 | * Currently obj_exts is used only for allocation profiling. | |
3952 | * If other users appear then mem_alloc_profiling_enabled() | |
3953 | * check should be added before alloc_tag_add(). | |
3954 | */ | |
3955 | if (likely(obj_exts)) | |
3956 | alloc_tag_add(&obj_exts->ref, current->alloc_tag, s->size); | |
3957 | #endif | |
3958 | } | |
6011be59 VB |
3959 | } |
3960 | ||
9f9796b4 | 3961 | return memcg_slab_post_alloc_hook(s, lru, flags, size, p); |
6011be59 VB |
3962 | } |
3963 | ||
56d5a2b9 VB |
3964 | /* |
3965 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
3966 | * have the fastpath folded into their functions. So no function call | |
3967 | * overhead for requests that can be satisfied on the fastpath. | |
3968 | * | |
3969 | * The fastpath works by first checking if the lockless freelist can be used. | |
3970 | * If not then __slab_alloc is called for slow processing. | |
3971 | * | |
3972 | * Otherwise we can simply pick the next object from the lockless free list. | |
3973 | */ | |
be784ba8 | 3974 | static __fastpath_inline void *slab_alloc_node(struct kmem_cache *s, struct list_lru *lru, |
56d5a2b9 VB |
3975 | gfp_t gfpflags, int node, unsigned long addr, size_t orig_size) |
3976 | { | |
3977 | void *object; | |
56d5a2b9 VB |
3978 | bool init = false; |
3979 | ||
9f9796b4 | 3980 | s = slab_pre_alloc_hook(s, gfpflags); |
3450a0e5 | 3981 | if (unlikely(!s)) |
56d5a2b9 VB |
3982 | return NULL; |
3983 | ||
3984 | object = kfence_alloc(s, orig_size, gfpflags); | |
3985 | if (unlikely(object)) | |
3986 | goto out; | |
3987 | ||
3988 | object = __slab_alloc_node(s, gfpflags, node, addr, orig_size); | |
3989 | ||
ce5716c6 | 3990 | maybe_wipe_obj_freeptr(s, object); |
da844b78 | 3991 | init = slab_want_init_on_alloc(gfpflags, s); |
d07dbea4 | 3992 | |
b89fb5ef | 3993 | out: |
9ce67395 FT |
3994 | /* |
3995 | * When init equals 'true', like for kzalloc() family, only | |
3996 | * @orig_size bytes might be zeroed instead of s->object_size | |
9f9796b4 VB |
3997 | * In case this fails due to memcg_slab_post_alloc_hook(), |
3998 | * object is set to NULL | |
9ce67395 | 3999 | */ |
9f9796b4 | 4000 | slab_post_alloc_hook(s, lru, gfpflags, 1, &object, init, orig_size); |
5a896d9e | 4001 | |
894b8788 | 4002 | return object; |
81819f0f CL |
4003 | } |
4004 | ||
7bd230a2 | 4005 | void *kmem_cache_alloc_noprof(struct kmem_cache *s, gfp_t gfpflags) |
2b847c3c | 4006 | { |
49378a05 VB |
4007 | void *ret = slab_alloc_node(s, NULL, gfpflags, NUMA_NO_NODE, _RET_IP_, |
4008 | s->object_size); | |
5b882be4 | 4009 | |
2c1d697f | 4010 | trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, NUMA_NO_NODE); |
5b882be4 EGM |
4011 | |
4012 | return ret; | |
2b847c3c | 4013 | } |
7bd230a2 | 4014 | EXPORT_SYMBOL(kmem_cache_alloc_noprof); |
2b847c3c | 4015 | |
7bd230a2 | 4016 | void *kmem_cache_alloc_lru_noprof(struct kmem_cache *s, struct list_lru *lru, |
88f2ef73 | 4017 | gfp_t gfpflags) |
81819f0f | 4018 | { |
49378a05 VB |
4019 | void *ret = slab_alloc_node(s, lru, gfpflags, NUMA_NO_NODE, _RET_IP_, |
4020 | s->object_size); | |
5b882be4 | 4021 | |
2c1d697f | 4022 | trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, NUMA_NO_NODE); |
5b882be4 EGM |
4023 | |
4024 | return ret; | |
81819f0f | 4025 | } |
7bd230a2 | 4026 | EXPORT_SYMBOL(kmem_cache_alloc_lru_noprof); |
88f2ef73 | 4027 | |
0445ee00 VB |
4028 | /** |
4029 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
4030 | * @s: The cache to allocate from. | |
4031 | * @gfpflags: See kmalloc(). | |
4032 | * @node: node number of the target node. | |
4033 | * | |
4034 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
4035 | * node, which can improve the performance for cpu bound structures. | |
4036 | * | |
4037 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
4038 | * | |
4039 | * Return: pointer to the new object or %NULL in case of error | |
4040 | */ | |
7bd230a2 | 4041 | void *kmem_cache_alloc_node_noprof(struct kmem_cache *s, gfp_t gfpflags, int node) |
81819f0f | 4042 | { |
88f2ef73 | 4043 | void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, s->object_size); |
5b882be4 | 4044 | |
2c1d697f | 4045 | trace_kmem_cache_alloc(_RET_IP_, ret, s, gfpflags, node); |
5b882be4 EGM |
4046 | |
4047 | return ret; | |
81819f0f | 4048 | } |
7bd230a2 | 4049 | EXPORT_SYMBOL(kmem_cache_alloc_node_noprof); |
81819f0f | 4050 | |
4862caa5 VB |
4051 | /* |
4052 | * To avoid unnecessary overhead, we pass through large allocation requests | |
4053 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
4054 | * know the allocation order to free the pages properly in kfree. | |
4055 | */ | |
4056 | static void *__kmalloc_large_node(size_t size, gfp_t flags, int node) | |
88f2ef73 | 4057 | { |
fb46e22a | 4058 | struct folio *folio; |
4862caa5 VB |
4059 | void *ptr = NULL; |
4060 | unsigned int order = get_order(size); | |
4061 | ||
4062 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) | |
4063 | flags = kmalloc_fix_flags(flags); | |
4064 | ||
4065 | flags |= __GFP_COMP; | |
7bd230a2 | 4066 | folio = (struct folio *)alloc_pages_node_noprof(node, flags, order); |
fb46e22a LT |
4067 | if (folio) { |
4068 | ptr = folio_address(folio); | |
4069 | lruvec_stat_mod_folio(folio, NR_SLAB_UNRECLAIMABLE_B, | |
4862caa5 VB |
4070 | PAGE_SIZE << order); |
4071 | } | |
4072 | ||
4073 | ptr = kasan_kmalloc_large(ptr, size, flags); | |
4074 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ | |
4075 | kmemleak_alloc(ptr, size, 1, flags); | |
4076 | kmsan_kmalloc_large(ptr, size, flags); | |
4077 | ||
4078 | return ptr; | |
88f2ef73 | 4079 | } |
81819f0f | 4080 | |
7bd230a2 | 4081 | void *kmalloc_large_noprof(size_t size, gfp_t flags) |
88f2ef73 | 4082 | { |
4862caa5 VB |
4083 | void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE); |
4084 | ||
4085 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size), | |
4086 | flags, NUMA_NO_NODE); | |
4087 | return ret; | |
88f2ef73 | 4088 | } |
7bd230a2 | 4089 | EXPORT_SYMBOL(kmalloc_large_noprof); |
88f2ef73 | 4090 | |
7bd230a2 | 4091 | void *kmalloc_large_node_noprof(size_t size, gfp_t flags, int node) |
4a92379b | 4092 | { |
4862caa5 VB |
4093 | void *ret = __kmalloc_large_node(size, flags, node); |
4094 | ||
4095 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size), | |
4096 | flags, node); | |
4097 | return ret; | |
4a92379b | 4098 | } |
7bd230a2 | 4099 | EXPORT_SYMBOL(kmalloc_large_node_noprof); |
5b882be4 | 4100 | |
4862caa5 VB |
4101 | static __always_inline |
4102 | void *__do_kmalloc_node(size_t size, gfp_t flags, int node, | |
4103 | unsigned long caller) | |
81819f0f | 4104 | { |
4862caa5 VB |
4105 | struct kmem_cache *s; |
4106 | void *ret; | |
5b882be4 | 4107 | |
4862caa5 VB |
4108 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
4109 | ret = __kmalloc_large_node(size, flags, node); | |
4110 | trace_kmalloc(caller, ret, size, | |
4111 | PAGE_SIZE << get_order(size), flags, node); | |
4112 | return ret; | |
4113 | } | |
5b882be4 | 4114 | |
4862caa5 VB |
4115 | if (unlikely(!size)) |
4116 | return ZERO_SIZE_PTR; | |
4117 | ||
4118 | s = kmalloc_slab(size, flags, caller); | |
4119 | ||
4120 | ret = slab_alloc_node(s, NULL, flags, node, caller, size); | |
4121 | ret = kasan_kmalloc(s, ret, size, flags); | |
4122 | trace_kmalloc(caller, ret, size, s->size, flags, node); | |
5b882be4 | 4123 | return ret; |
81819f0f | 4124 | } |
4862caa5 | 4125 | |
7bd230a2 | 4126 | void *__kmalloc_node_noprof(size_t size, gfp_t flags, int node) |
4862caa5 VB |
4127 | { |
4128 | return __do_kmalloc_node(size, flags, node, _RET_IP_); | |
4129 | } | |
7bd230a2 | 4130 | EXPORT_SYMBOL(__kmalloc_node_noprof); |
4862caa5 | 4131 | |
7bd230a2 | 4132 | void *__kmalloc_noprof(size_t size, gfp_t flags) |
4862caa5 VB |
4133 | { |
4134 | return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_); | |
4135 | } | |
7bd230a2 | 4136 | EXPORT_SYMBOL(__kmalloc_noprof); |
4862caa5 | 4137 | |
7bd230a2 SB |
4138 | void *kmalloc_node_track_caller_noprof(size_t size, gfp_t flags, |
4139 | int node, unsigned long caller) | |
4862caa5 VB |
4140 | { |
4141 | return __do_kmalloc_node(size, flags, node, caller); | |
4142 | } | |
7bd230a2 | 4143 | EXPORT_SYMBOL(kmalloc_node_track_caller_noprof); |
4862caa5 | 4144 | |
7bd230a2 | 4145 | void *kmalloc_trace_noprof(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
4862caa5 VB |
4146 | { |
4147 | void *ret = slab_alloc_node(s, NULL, gfpflags, NUMA_NO_NODE, | |
4148 | _RET_IP_, size); | |
4149 | ||
4150 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE); | |
4151 | ||
4152 | ret = kasan_kmalloc(s, ret, size, gfpflags); | |
4153 | return ret; | |
4154 | } | |
7bd230a2 | 4155 | EXPORT_SYMBOL(kmalloc_trace_noprof); |
4862caa5 | 4156 | |
7bd230a2 | 4157 | void *kmalloc_node_trace_noprof(struct kmem_cache *s, gfp_t gfpflags, |
4862caa5 VB |
4158 | int node, size_t size) |
4159 | { | |
4160 | void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, size); | |
4161 | ||
4162 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node); | |
4163 | ||
4164 | ret = kasan_kmalloc(s, ret, size, gfpflags); | |
4165 | return ret; | |
4166 | } | |
7bd230a2 | 4167 | EXPORT_SYMBOL(kmalloc_node_trace_noprof); |
81819f0f | 4168 | |
fa9b88e4 VB |
4169 | static noinline void free_to_partial_list( |
4170 | struct kmem_cache *s, struct slab *slab, | |
4171 | void *head, void *tail, int bulk_cnt, | |
4172 | unsigned long addr) | |
4173 | { | |
4174 | struct kmem_cache_node *n = get_node(s, slab_nid(slab)); | |
4175 | struct slab *slab_free = NULL; | |
4176 | int cnt = bulk_cnt; | |
4177 | unsigned long flags; | |
4178 | depot_stack_handle_t handle = 0; | |
4179 | ||
4180 | if (s->flags & SLAB_STORE_USER) | |
4181 | handle = set_track_prepare(); | |
4182 | ||
4183 | spin_lock_irqsave(&n->list_lock, flags); | |
4184 | ||
4185 | if (free_debug_processing(s, slab, head, tail, &cnt, addr, handle)) { | |
4186 | void *prior = slab->freelist; | |
4187 | ||
4188 | /* Perform the actual freeing while we still hold the locks */ | |
4189 | slab->inuse -= cnt; | |
4190 | set_freepointer(s, tail, prior); | |
4191 | slab->freelist = head; | |
4192 | ||
4193 | /* | |
4194 | * If the slab is empty, and node's partial list is full, | |
4195 | * it should be discarded anyway no matter it's on full or | |
4196 | * partial list. | |
4197 | */ | |
4198 | if (slab->inuse == 0 && n->nr_partial >= s->min_partial) | |
4199 | slab_free = slab; | |
4200 | ||
4201 | if (!prior) { | |
4202 | /* was on full list */ | |
4203 | remove_full(s, n, slab); | |
4204 | if (!slab_free) { | |
4205 | add_partial(n, slab, DEACTIVATE_TO_TAIL); | |
4206 | stat(s, FREE_ADD_PARTIAL); | |
4207 | } | |
4208 | } else if (slab_free) { | |
4209 | remove_partial(n, slab); | |
4210 | stat(s, FREE_REMOVE_PARTIAL); | |
4211 | } | |
4212 | } | |
4213 | ||
4214 | if (slab_free) { | |
4215 | /* | |
4216 | * Update the counters while still holding n->list_lock to | |
4217 | * prevent spurious validation warnings | |
4218 | */ | |
4219 | dec_slabs_node(s, slab_nid(slab_free), slab_free->objects); | |
4220 | } | |
4221 | ||
4222 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4223 | ||
4224 | if (slab_free) { | |
4225 | stat(s, FREE_SLAB); | |
4226 | free_slab(s, slab_free); | |
4227 | } | |
4228 | } | |
4229 | ||
81819f0f | 4230 | /* |
94e4d712 | 4231 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 4232 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 4233 | * |
894b8788 | 4234 | * So we still attempt to reduce cache line usage. Just take the slab |
c2092c12 | 4235 | * lock and free the item. If there is no additional partial slab |
894b8788 | 4236 | * handling required then we can return immediately. |
81819f0f | 4237 | */ |
bb192ed9 | 4238 | static void __slab_free(struct kmem_cache *s, struct slab *slab, |
81084651 JDB |
4239 | void *head, void *tail, int cnt, |
4240 | unsigned long addr) | |
4241 | ||
81819f0f CL |
4242 | { |
4243 | void *prior; | |
2cfb7455 | 4244 | int was_frozen; |
bb192ed9 | 4245 | struct slab new; |
2cfb7455 CL |
4246 | unsigned long counters; |
4247 | struct kmem_cache_node *n = NULL; | |
3f649ab7 | 4248 | unsigned long flags; |
422e7d54 | 4249 | bool on_node_partial; |
81819f0f | 4250 | |
8a5ec0ba | 4251 | stat(s, FREE_SLOWPATH); |
81819f0f | 4252 | |
0af8489b | 4253 | if (IS_ENABLED(CONFIG_SLUB_TINY) || kmem_cache_debug(s)) { |
fa9b88e4 | 4254 | free_to_partial_list(s, slab, head, tail, cnt, addr); |
80f08c19 | 4255 | return; |
c7323a5a | 4256 | } |
6446faa2 | 4257 | |
2cfb7455 | 4258 | do { |
837d678d JK |
4259 | if (unlikely(n)) { |
4260 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4261 | n = NULL; | |
4262 | } | |
bb192ed9 VB |
4263 | prior = slab->freelist; |
4264 | counters = slab->counters; | |
81084651 | 4265 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
4266 | new.counters = counters; |
4267 | was_frozen = new.frozen; | |
81084651 | 4268 | new.inuse -= cnt; |
837d678d | 4269 | if ((!new.inuse || !prior) && !was_frozen) { |
8cd3fa42 CZ |
4270 | /* Needs to be taken off a list */ |
4271 | if (!kmem_cache_has_cpu_partial(s) || prior) { | |
49e22585 | 4272 | |
bb192ed9 | 4273 | n = get_node(s, slab_nid(slab)); |
49e22585 CL |
4274 | /* |
4275 | * Speculatively acquire the list_lock. | |
4276 | * If the cmpxchg does not succeed then we may | |
4277 | * drop the list_lock without any processing. | |
4278 | * | |
4279 | * Otherwise the list_lock will synchronize with | |
4280 | * other processors updating the list of slabs. | |
4281 | */ | |
4282 | spin_lock_irqsave(&n->list_lock, flags); | |
4283 | ||
422e7d54 | 4284 | on_node_partial = slab_test_node_partial(slab); |
49e22585 | 4285 | } |
2cfb7455 | 4286 | } |
81819f0f | 4287 | |
6801be4f | 4288 | } while (!slab_update_freelist(s, slab, |
2cfb7455 | 4289 | prior, counters, |
81084651 | 4290 | head, new.counters, |
2cfb7455 | 4291 | "__slab_free")); |
81819f0f | 4292 | |
2cfb7455 | 4293 | if (likely(!n)) { |
49e22585 | 4294 | |
c270cf30 AW |
4295 | if (likely(was_frozen)) { |
4296 | /* | |
4297 | * The list lock was not taken therefore no list | |
4298 | * activity can be necessary. | |
4299 | */ | |
4300 | stat(s, FREE_FROZEN); | |
8cd3fa42 | 4301 | } else if (kmem_cache_has_cpu_partial(s) && !prior) { |
c270cf30 | 4302 | /* |
8cd3fa42 | 4303 | * If we started with a full slab then put it onto the |
c270cf30 AW |
4304 | * per cpu partial list. |
4305 | */ | |
bb192ed9 | 4306 | put_cpu_partial(s, slab, 1); |
8028dcea AS |
4307 | stat(s, CPU_PARTIAL_FREE); |
4308 | } | |
c270cf30 | 4309 | |
b455def2 L |
4310 | return; |
4311 | } | |
81819f0f | 4312 | |
422e7d54 CZ |
4313 | /* |
4314 | * This slab was partially empty but not on the per-node partial list, | |
4315 | * in which case we shouldn't manipulate its list, just return. | |
4316 | */ | |
4317 | if (prior && !on_node_partial) { | |
4318 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4319 | return; | |
4320 | } | |
4321 | ||
8a5b20ae | 4322 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
4323 | goto slab_empty; |
4324 | ||
81819f0f | 4325 | /* |
837d678d JK |
4326 | * Objects left in the slab. If it was not on the partial list before |
4327 | * then add it. | |
81819f0f | 4328 | */ |
345c905d | 4329 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
bb192ed9 | 4330 | add_partial(n, slab, DEACTIVATE_TO_TAIL); |
837d678d | 4331 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 4332 | } |
80f08c19 | 4333 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
4334 | return; |
4335 | ||
4336 | slab_empty: | |
a973e9dd | 4337 | if (prior) { |
81819f0f | 4338 | /* |
6fbabb20 | 4339 | * Slab on the partial list. |
81819f0f | 4340 | */ |
bb192ed9 | 4341 | remove_partial(n, slab); |
84e554e6 | 4342 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 4343 | } |
2cfb7455 | 4344 | |
80f08c19 | 4345 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 4346 | stat(s, FREE_SLAB); |
bb192ed9 | 4347 | discard_slab(s, slab); |
81819f0f CL |
4348 | } |
4349 | ||
0af8489b | 4350 | #ifndef CONFIG_SLUB_TINY |
894b8788 CL |
4351 | /* |
4352 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
4353 | * can perform fastpath freeing without additional function calls. | |
4354 | * | |
4355 | * The fastpath is only possible if we are freeing to the current cpu slab | |
4356 | * of this processor. This typically the case if we have just allocated | |
4357 | * the item before. | |
4358 | * | |
4359 | * If fastpath is not possible then fall back to __slab_free where we deal | |
4360 | * with all sorts of special processing. | |
81084651 JDB |
4361 | * |
4362 | * Bulk free of a freelist with several objects (all pointing to the | |
c2092c12 | 4363 | * same slab) possible by specifying head and tail ptr, plus objects |
81084651 | 4364 | * count (cnt). Bulk free indicated by tail pointer being set. |
894b8788 | 4365 | */ |
80a9201a | 4366 | static __always_inline void do_slab_free(struct kmem_cache *s, |
bb192ed9 | 4367 | struct slab *slab, void *head, void *tail, |
80a9201a | 4368 | int cnt, unsigned long addr) |
894b8788 | 4369 | { |
dfb4f096 | 4370 | struct kmem_cache_cpu *c; |
8a5ec0ba | 4371 | unsigned long tid; |
1f04b07d | 4372 | void **freelist; |
964d4bd3 | 4373 | |
8a5ec0ba CL |
4374 | redo: |
4375 | /* | |
4376 | * Determine the currently cpus per cpu slab. | |
4377 | * The cpu may change afterward. However that does not matter since | |
4378 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 4379 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 4380 | */ |
9b4bc85a VB |
4381 | c = raw_cpu_ptr(s->cpu_slab); |
4382 | tid = READ_ONCE(c->tid); | |
c016b0bd | 4383 | |
b062539c | 4384 | /* Same with comment on barrier() in __slab_alloc_node() */ |
9aabf810 | 4385 | barrier(); |
c016b0bd | 4386 | |
1f04b07d | 4387 | if (unlikely(slab != c->slab)) { |
284f17ac | 4388 | __slab_free(s, slab, head, tail, cnt, addr); |
1f04b07d TG |
4389 | return; |
4390 | } | |
4391 | ||
4392 | if (USE_LOCKLESS_FAST_PATH()) { | |
4393 | freelist = READ_ONCE(c->freelist); | |
5076190d | 4394 | |
284f17ac | 4395 | set_freepointer(s, tail, freelist); |
8a5ec0ba | 4396 | |
6801be4f | 4397 | if (unlikely(!__update_cpu_freelist_fast(s, freelist, head, tid))) { |
8a5ec0ba CL |
4398 | note_cmpxchg_failure("slab_free", s, tid); |
4399 | goto redo; | |
4400 | } | |
1f04b07d TG |
4401 | } else { |
4402 | /* Update the free list under the local lock */ | |
bd0e7491 VB |
4403 | local_lock(&s->cpu_slab->lock); |
4404 | c = this_cpu_ptr(s->cpu_slab); | |
bb192ed9 | 4405 | if (unlikely(slab != c->slab)) { |
bd0e7491 VB |
4406 | local_unlock(&s->cpu_slab->lock); |
4407 | goto redo; | |
4408 | } | |
4409 | tid = c->tid; | |
4410 | freelist = c->freelist; | |
4411 | ||
284f17ac | 4412 | set_freepointer(s, tail, freelist); |
bd0e7491 VB |
4413 | c->freelist = head; |
4414 | c->tid = next_tid(tid); | |
4415 | ||
4416 | local_unlock(&s->cpu_slab->lock); | |
1f04b07d | 4417 | } |
6f3dd2c3 | 4418 | stat_add(s, FREE_FASTPATH, cnt); |
894b8788 | 4419 | } |
0af8489b VB |
4420 | #else /* CONFIG_SLUB_TINY */ |
4421 | static void do_slab_free(struct kmem_cache *s, | |
4422 | struct slab *slab, void *head, void *tail, | |
4423 | int cnt, unsigned long addr) | |
4424 | { | |
284f17ac | 4425 | __slab_free(s, slab, head, tail, cnt, addr); |
0af8489b VB |
4426 | } |
4427 | #endif /* CONFIG_SLUB_TINY */ | |
894b8788 | 4428 | |
284f17ac VB |
4429 | static __fastpath_inline |
4430 | void slab_free(struct kmem_cache *s, struct slab *slab, void *object, | |
4431 | unsigned long addr) | |
4432 | { | |
284f17ac | 4433 | memcg_slab_free_hook(s, slab, &object, 1); |
4b873696 | 4434 | alloc_tagging_slab_free_hook(s, slab, &object, 1); |
284f17ac | 4435 | |
782f8906 | 4436 | if (likely(slab_free_hook(s, object, slab_want_init_on_free(s)))) |
284f17ac VB |
4437 | do_slab_free(s, slab, object, object, 1, addr); |
4438 | } | |
4439 | ||
9f9796b4 VB |
4440 | #ifdef CONFIG_MEMCG_KMEM |
4441 | /* Do not inline the rare memcg charging failed path into the allocation path */ | |
4442 | static noinline | |
4443 | void memcg_alloc_abort_single(struct kmem_cache *s, void *object) | |
4444 | { | |
4445 | if (likely(slab_free_hook(s, object, slab_want_init_on_free(s)))) | |
4446 | do_slab_free(s, virt_to_slab(object), object, object, 1, _RET_IP_); | |
4447 | } | |
4448 | #endif | |
4449 | ||
284f17ac VB |
4450 | static __fastpath_inline |
4451 | void slab_free_bulk(struct kmem_cache *s, struct slab *slab, void *head, | |
4452 | void *tail, void **p, int cnt, unsigned long addr) | |
80a9201a | 4453 | { |
b77d5b1b | 4454 | memcg_slab_free_hook(s, slab, p, cnt); |
4b873696 | 4455 | alloc_tagging_slab_free_hook(s, slab, p, cnt); |
80a9201a | 4456 | /* |
c3895391 AK |
4457 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
4458 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 4459 | */ |
ecf9a253 | 4460 | if (likely(slab_free_freelist_hook(s, &head, &tail, &cnt))) |
bb192ed9 | 4461 | do_slab_free(s, slab, head, tail, cnt, addr); |
80a9201a AP |
4462 | } |
4463 | ||
2bd926b4 | 4464 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
4465 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
4466 | { | |
284f17ac | 4467 | do_slab_free(cache, virt_to_slab(x), x, x, 1, addr); |
80a9201a AP |
4468 | } |
4469 | #endif | |
4470 | ||
0bedcc66 | 4471 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
ed4cd17e | 4472 | { |
0bedcc66 VB |
4473 | struct slab *slab; |
4474 | ||
4475 | slab = virt_to_slab(obj); | |
4476 | if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n", __func__)) | |
4477 | return NULL; | |
4478 | return slab->slab_cache; | |
ed4cd17e HY |
4479 | } |
4480 | ||
0bedcc66 VB |
4481 | static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x) |
4482 | { | |
4483 | struct kmem_cache *cachep; | |
4484 | ||
4485 | if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) && | |
4486 | !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) | |
4487 | return s; | |
4488 | ||
4489 | cachep = virt_to_cache(x); | |
4490 | if (WARN(cachep && cachep != s, | |
4491 | "%s: Wrong slab cache. %s but object is from %s\n", | |
4492 | __func__, s->name, cachep->name)) | |
4493 | print_tracking(cachep, x); | |
4494 | return cachep; | |
4495 | } | |
4496 | ||
0445ee00 VB |
4497 | /** |
4498 | * kmem_cache_free - Deallocate an object | |
4499 | * @s: The cache the allocation was from. | |
4500 | * @x: The previously allocated object. | |
4501 | * | |
4502 | * Free an object which was previously allocated from this | |
4503 | * cache. | |
4504 | */ | |
81819f0f CL |
4505 | void kmem_cache_free(struct kmem_cache *s, void *x) |
4506 | { | |
b9ce5ef4 GC |
4507 | s = cache_from_obj(s, x); |
4508 | if (!s) | |
79576102 | 4509 | return; |
2c1d697f | 4510 | trace_kmem_cache_free(_RET_IP_, x, s); |
284f17ac | 4511 | slab_free(s, virt_to_slab(x), x, _RET_IP_); |
81819f0f CL |
4512 | } |
4513 | EXPORT_SYMBOL(kmem_cache_free); | |
4514 | ||
b774d3e3 VB |
4515 | static void free_large_kmalloc(struct folio *folio, void *object) |
4516 | { | |
4517 | unsigned int order = folio_order(folio); | |
4518 | ||
4519 | if (WARN_ON_ONCE(order == 0)) | |
4520 | pr_warn_once("object pointer: 0x%p\n", object); | |
4521 | ||
4522 | kmemleak_free(object); | |
4523 | kasan_kfree_large(object); | |
4524 | kmsan_kfree_large(object); | |
4525 | ||
fb46e22a | 4526 | lruvec_stat_mod_folio(folio, NR_SLAB_UNRECLAIMABLE_B, |
b774d3e3 | 4527 | -(PAGE_SIZE << order)); |
fb46e22a | 4528 | folio_put(folio); |
b774d3e3 VB |
4529 | } |
4530 | ||
4531 | /** | |
4532 | * kfree - free previously allocated memory | |
4533 | * @object: pointer returned by kmalloc() or kmem_cache_alloc() | |
4534 | * | |
4535 | * If @object is NULL, no operation is performed. | |
4536 | */ | |
4537 | void kfree(const void *object) | |
4538 | { | |
4539 | struct folio *folio; | |
4540 | struct slab *slab; | |
4541 | struct kmem_cache *s; | |
4542 | void *x = (void *)object; | |
4543 | ||
4544 | trace_kfree(_RET_IP_, object); | |
4545 | ||
4546 | if (unlikely(ZERO_OR_NULL_PTR(object))) | |
4547 | return; | |
4548 | ||
4549 | folio = virt_to_folio(object); | |
4550 | if (unlikely(!folio_test_slab(folio))) { | |
4551 | free_large_kmalloc(folio, (void *)object); | |
4552 | return; | |
4553 | } | |
4554 | ||
4555 | slab = folio_slab(folio); | |
4556 | s = slab->slab_cache; | |
284f17ac | 4557 | slab_free(s, slab, x, _RET_IP_); |
b774d3e3 VB |
4558 | } |
4559 | EXPORT_SYMBOL(kfree); | |
4560 | ||
d0ecd894 | 4561 | struct detached_freelist { |
cc465c3b | 4562 | struct slab *slab; |
d0ecd894 JDB |
4563 | void *tail; |
4564 | void *freelist; | |
4565 | int cnt; | |
376bf125 | 4566 | struct kmem_cache *s; |
d0ecd894 | 4567 | }; |
fbd02630 | 4568 | |
d0ecd894 JDB |
4569 | /* |
4570 | * This function progressively scans the array with free objects (with | |
4571 | * a limited look ahead) and extract objects belonging to the same | |
cc465c3b MWO |
4572 | * slab. It builds a detached freelist directly within the given |
4573 | * slab/objects. This can happen without any need for | |
d0ecd894 JDB |
4574 | * synchronization, because the objects are owned by running process. |
4575 | * The freelist is build up as a single linked list in the objects. | |
4576 | * The idea is, that this detached freelist can then be bulk | |
4577 | * transferred to the real freelist(s), but only requiring a single | |
4578 | * synchronization primitive. Look ahead in the array is limited due | |
4579 | * to performance reasons. | |
4580 | */ | |
376bf125 JDB |
4581 | static inline |
4582 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
4583 | void **p, struct detached_freelist *df) | |
d0ecd894 | 4584 | { |
d0ecd894 JDB |
4585 | int lookahead = 3; |
4586 | void *object; | |
cc465c3b | 4587 | struct folio *folio; |
b77d5b1b | 4588 | size_t same; |
fbd02630 | 4589 | |
b77d5b1b | 4590 | object = p[--size]; |
cc465c3b | 4591 | folio = virt_to_folio(object); |
ca257195 JDB |
4592 | if (!s) { |
4593 | /* Handle kalloc'ed objects */ | |
cc465c3b | 4594 | if (unlikely(!folio_test_slab(folio))) { |
d835eef4 | 4595 | free_large_kmalloc(folio, object); |
b77d5b1b | 4596 | df->slab = NULL; |
ca257195 JDB |
4597 | return size; |
4598 | } | |
4599 | /* Derive kmem_cache from object */ | |
b77d5b1b MS |
4600 | df->slab = folio_slab(folio); |
4601 | df->s = df->slab->slab_cache; | |
ca257195 | 4602 | } else { |
b77d5b1b | 4603 | df->slab = folio_slab(folio); |
ca257195 JDB |
4604 | df->s = cache_from_obj(s, object); /* Support for memcg */ |
4605 | } | |
376bf125 | 4606 | |
d0ecd894 | 4607 | /* Start new detached freelist */ |
d0ecd894 JDB |
4608 | df->tail = object; |
4609 | df->freelist = object; | |
d0ecd894 JDB |
4610 | df->cnt = 1; |
4611 | ||
b77d5b1b MS |
4612 | if (is_kfence_address(object)) |
4613 | return size; | |
4614 | ||
4615 | set_freepointer(df->s, object, NULL); | |
4616 | ||
4617 | same = size; | |
d0ecd894 JDB |
4618 | while (size) { |
4619 | object = p[--size]; | |
cc465c3b MWO |
4620 | /* df->slab is always set at this point */ |
4621 | if (df->slab == virt_to_slab(object)) { | |
d0ecd894 | 4622 | /* Opportunity build freelist */ |
376bf125 | 4623 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
4624 | df->freelist = object; |
4625 | df->cnt++; | |
b77d5b1b MS |
4626 | same--; |
4627 | if (size != same) | |
4628 | swap(p[size], p[same]); | |
d0ecd894 | 4629 | continue; |
fbd02630 | 4630 | } |
d0ecd894 JDB |
4631 | |
4632 | /* Limit look ahead search */ | |
4633 | if (!--lookahead) | |
4634 | break; | |
fbd02630 | 4635 | } |
d0ecd894 | 4636 | |
b77d5b1b | 4637 | return same; |
d0ecd894 JDB |
4638 | } |
4639 | ||
520a688a VB |
4640 | /* |
4641 | * Internal bulk free of objects that were not initialised by the post alloc | |
4642 | * hooks and thus should not be processed by the free hooks | |
4643 | */ | |
4644 | static void __kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) | |
4645 | { | |
4646 | if (!size) | |
4647 | return; | |
4648 | ||
4649 | do { | |
4650 | struct detached_freelist df; | |
4651 | ||
4652 | size = build_detached_freelist(s, size, p, &df); | |
4653 | if (!df.slab) | |
4654 | continue; | |
4655 | ||
4656 | do_slab_free(df.s, df.slab, df.freelist, df.tail, df.cnt, | |
4657 | _RET_IP_); | |
4658 | } while (likely(size)); | |
4659 | } | |
4660 | ||
d0ecd894 | 4661 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 4662 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 | 4663 | { |
2055e67b | 4664 | if (!size) |
d0ecd894 JDB |
4665 | return; |
4666 | ||
4667 | do { | |
4668 | struct detached_freelist df; | |
4669 | ||
4670 | size = build_detached_freelist(s, size, p, &df); | |
cc465c3b | 4671 | if (!df.slab) |
d0ecd894 JDB |
4672 | continue; |
4673 | ||
284f17ac VB |
4674 | slab_free_bulk(df.s, df.slab, df.freelist, df.tail, &p[size], |
4675 | df.cnt, _RET_IP_); | |
d0ecd894 | 4676 | } while (likely(size)); |
484748f0 CL |
4677 | } |
4678 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
4679 | ||
0af8489b | 4680 | #ifndef CONFIG_SLUB_TINY |
520a688a VB |
4681 | static inline |
4682 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, | |
4683 | void **p) | |
484748f0 | 4684 | { |
994eb764 | 4685 | struct kmem_cache_cpu *c; |
f5451547 | 4686 | unsigned long irqflags; |
994eb764 JDB |
4687 | int i; |
4688 | ||
994eb764 JDB |
4689 | /* |
4690 | * Drain objects in the per cpu slab, while disabling local | |
4691 | * IRQs, which protects against PREEMPT and interrupts | |
4692 | * handlers invoking normal fastpath. | |
4693 | */ | |
25c00c50 | 4694 | c = slub_get_cpu_ptr(s->cpu_slab); |
f5451547 | 4695 | local_lock_irqsave(&s->cpu_slab->lock, irqflags); |
994eb764 JDB |
4696 | |
4697 | for (i = 0; i < size; i++) { | |
b89fb5ef | 4698 | void *object = kfence_alloc(s, s->object_size, flags); |
994eb764 | 4699 | |
b89fb5ef AP |
4700 | if (unlikely(object)) { |
4701 | p[i] = object; | |
4702 | continue; | |
4703 | } | |
4704 | ||
4705 | object = c->freelist; | |
ebe909e0 | 4706 | if (unlikely(!object)) { |
fd4d9c7d JH |
4707 | /* |
4708 | * We may have removed an object from c->freelist using | |
4709 | * the fastpath in the previous iteration; in that case, | |
4710 | * c->tid has not been bumped yet. | |
4711 | * Since ___slab_alloc() may reenable interrupts while | |
4712 | * allocating memory, we should bump c->tid now. | |
4713 | */ | |
4714 | c->tid = next_tid(c->tid); | |
4715 | ||
f5451547 | 4716 | local_unlock_irqrestore(&s->cpu_slab->lock, irqflags); |
e500059b | 4717 | |
ebe909e0 JDB |
4718 | /* |
4719 | * Invoking slow path likely have side-effect | |
4720 | * of re-populating per CPU c->freelist | |
4721 | */ | |
87098373 | 4722 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
6edf2576 | 4723 | _RET_IP_, c, s->object_size); |
87098373 CL |
4724 | if (unlikely(!p[i])) |
4725 | goto error; | |
4726 | ||
ebe909e0 | 4727 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
4728 | maybe_wipe_obj_freeptr(s, p[i]); |
4729 | ||
f5451547 | 4730 | local_lock_irqsave(&s->cpu_slab->lock, irqflags); |
e500059b | 4731 | |
ebe909e0 JDB |
4732 | continue; /* goto for-loop */ |
4733 | } | |
994eb764 JDB |
4734 | c->freelist = get_freepointer(s, object); |
4735 | p[i] = object; | |
0f181f9f | 4736 | maybe_wipe_obj_freeptr(s, p[i]); |
6f3dd2c3 | 4737 | stat(s, ALLOC_FASTPATH); |
994eb764 JDB |
4738 | } |
4739 | c->tid = next_tid(c->tid); | |
f5451547 | 4740 | local_unlock_irqrestore(&s->cpu_slab->lock, irqflags); |
25c00c50 | 4741 | slub_put_cpu_ptr(s->cpu_slab); |
994eb764 | 4742 | |
865762a8 | 4743 | return i; |
56d5a2b9 | 4744 | |
87098373 | 4745 | error: |
25c00c50 | 4746 | slub_put_cpu_ptr(s->cpu_slab); |
520a688a | 4747 | __kmem_cache_free_bulk(s, i, p); |
865762a8 | 4748 | return 0; |
56d5a2b9 VB |
4749 | |
4750 | } | |
0af8489b VB |
4751 | #else /* CONFIG_SLUB_TINY */ |
4752 | static int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, | |
520a688a | 4753 | size_t size, void **p) |
0af8489b VB |
4754 | { |
4755 | int i; | |
4756 | ||
4757 | for (i = 0; i < size; i++) { | |
4758 | void *object = kfence_alloc(s, s->object_size, flags); | |
4759 | ||
4760 | if (unlikely(object)) { | |
4761 | p[i] = object; | |
4762 | continue; | |
4763 | } | |
4764 | ||
4765 | p[i] = __slab_alloc_node(s, flags, NUMA_NO_NODE, | |
4766 | _RET_IP_, s->object_size); | |
4767 | if (unlikely(!p[i])) | |
4768 | goto error; | |
4769 | ||
4770 | maybe_wipe_obj_freeptr(s, p[i]); | |
4771 | } | |
4772 | ||
4773 | return i; | |
4774 | ||
4775 | error: | |
520a688a | 4776 | __kmem_cache_free_bulk(s, i, p); |
0af8489b VB |
4777 | return 0; |
4778 | } | |
4779 | #endif /* CONFIG_SLUB_TINY */ | |
56d5a2b9 VB |
4780 | |
4781 | /* Note that interrupts must be enabled when calling this function. */ | |
7bd230a2 SB |
4782 | int kmem_cache_alloc_bulk_noprof(struct kmem_cache *s, gfp_t flags, size_t size, |
4783 | void **p) | |
56d5a2b9 VB |
4784 | { |
4785 | int i; | |
56d5a2b9 VB |
4786 | |
4787 | if (!size) | |
4788 | return 0; | |
4789 | ||
9f9796b4 | 4790 | s = slab_pre_alloc_hook(s, flags); |
56d5a2b9 VB |
4791 | if (unlikely(!s)) |
4792 | return 0; | |
4793 | ||
520a688a | 4794 | i = __kmem_cache_alloc_bulk(s, flags, size, p); |
9f9796b4 VB |
4795 | if (unlikely(i == 0)) |
4796 | return 0; | |
56d5a2b9 VB |
4797 | |
4798 | /* | |
4799 | * memcg and kmem_cache debug support and memory initialization. | |
4800 | * Done outside of the IRQ disabled fastpath loop. | |
4801 | */ | |
9f9796b4 VB |
4802 | if (unlikely(!slab_post_alloc_hook(s, NULL, flags, size, p, |
4803 | slab_want_init_on_alloc(flags, s), s->object_size))) { | |
4804 | return 0; | |
520a688a | 4805 | } |
56d5a2b9 | 4806 | return i; |
484748f0 | 4807 | } |
7bd230a2 | 4808 | EXPORT_SYMBOL(kmem_cache_alloc_bulk_noprof); |
484748f0 CL |
4809 | |
4810 | ||
81819f0f | 4811 | /* |
672bba3a CL |
4812 | * Object placement in a slab is made very easy because we always start at |
4813 | * offset 0. If we tune the size of the object to the alignment then we can | |
4814 | * get the required alignment by putting one properly sized object after | |
4815 | * another. | |
81819f0f CL |
4816 | * |
4817 | * Notice that the allocation order determines the sizes of the per cpu | |
4818 | * caches. Each processor has always one slab available for allocations. | |
4819 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 4820 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 4821 | * locking overhead. |
81819f0f CL |
4822 | */ |
4823 | ||
4824 | /* | |
f0953a1b | 4825 | * Minimum / Maximum order of slab pages. This influences locking overhead |
81819f0f CL |
4826 | * and slab fragmentation. A higher order reduces the number of partial slabs |
4827 | * and increases the number of allocations possible without having to | |
4828 | * take the list_lock. | |
4829 | */ | |
19af27af | 4830 | static unsigned int slub_min_order; |
90ce872c VB |
4831 | static unsigned int slub_max_order = |
4832 | IS_ENABLED(CONFIG_SLUB_TINY) ? 1 : PAGE_ALLOC_COSTLY_ORDER; | |
19af27af | 4833 | static unsigned int slub_min_objects; |
81819f0f | 4834 | |
81819f0f CL |
4835 | /* |
4836 | * Calculate the order of allocation given an slab object size. | |
4837 | * | |
672bba3a CL |
4838 | * The order of allocation has significant impact on performance and other |
4839 | * system components. Generally order 0 allocations should be preferred since | |
4840 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
4841 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 4842 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
4843 | * would be wasted. |
4844 | * | |
4845 | * In order to reach satisfactory performance we must ensure that a minimum | |
4846 | * number of objects is in one slab. Otherwise we may generate too much | |
4847 | * activity on the partial lists which requires taking the list_lock. This is | |
4848 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 4849 | * |
671776b3 XS |
4850 | * slab_max_order specifies the order where we begin to stop considering the |
4851 | * number of objects in a slab as critical. If we reach slab_max_order then | |
672bba3a CL |
4852 | * we try to keep the page order as low as possible. So we accept more waste |
4853 | * of space in favor of a small page order. | |
81819f0f | 4854 | * |
672bba3a CL |
4855 | * Higher order allocations also allow the placement of more objects in a |
4856 | * slab and thereby reduce object handling overhead. If the user has | |
dc84207d | 4857 | * requested a higher minimum order then we start with that one instead of |
672bba3a | 4858 | * the smallest order which will fit the object. |
81819f0f | 4859 | */ |
d122019b | 4860 | static inline unsigned int calc_slab_order(unsigned int size, |
90f055df | 4861 | unsigned int min_order, unsigned int max_order, |
9736d2a9 | 4862 | unsigned int fract_leftover) |
81819f0f | 4863 | { |
19af27af | 4864 | unsigned int order; |
81819f0f | 4865 | |
90f055df | 4866 | for (order = min_order; order <= max_order; order++) { |
81819f0f | 4867 | |
19af27af AD |
4868 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
4869 | unsigned int rem; | |
81819f0f | 4870 | |
9736d2a9 | 4871 | rem = slab_size % size; |
81819f0f | 4872 | |
5e6d444e | 4873 | if (rem <= slab_size / fract_leftover) |
81819f0f | 4874 | break; |
81819f0f | 4875 | } |
672bba3a | 4876 | |
81819f0f CL |
4877 | return order; |
4878 | } | |
4879 | ||
9736d2a9 | 4880 | static inline int calculate_order(unsigned int size) |
5e6d444e | 4881 | { |
19af27af AD |
4882 | unsigned int order; |
4883 | unsigned int min_objects; | |
4884 | unsigned int max_objects; | |
90f055df | 4885 | unsigned int min_order; |
5e6d444e | 4886 | |
5e6d444e | 4887 | min_objects = slub_min_objects; |
3286222f VB |
4888 | if (!min_objects) { |
4889 | /* | |
4890 | * Some architectures will only update present cpus when | |
4891 | * onlining them, so don't trust the number if it's just 1. But | |
4892 | * we also don't want to use nr_cpu_ids always, as on some other | |
4893 | * architectures, there can be many possible cpus, but never | |
4894 | * onlined. Here we compromise between trying to avoid too high | |
4895 | * order on systems that appear larger than they are, and too | |
4896 | * low order on systems that appear smaller than they are. | |
4897 | */ | |
90f055df | 4898 | unsigned int nr_cpus = num_present_cpus(); |
3286222f VB |
4899 | if (nr_cpus <= 1) |
4900 | nr_cpus = nr_cpu_ids; | |
4901 | min_objects = 4 * (fls(nr_cpus) + 1); | |
4902 | } | |
90f055df VB |
4903 | /* min_objects can't be 0 because get_order(0) is undefined */ |
4904 | max_objects = max(order_objects(slub_max_order, size), 1U); | |
e8120ff1 ZY |
4905 | min_objects = min(min_objects, max_objects); |
4906 | ||
90f055df VB |
4907 | min_order = max_t(unsigned int, slub_min_order, |
4908 | get_order(min_objects * size)); | |
4909 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) | |
4910 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | |
4911 | ||
0fe2735d VB |
4912 | /* |
4913 | * Attempt to find best configuration for a slab. This works by first | |
4914 | * attempting to generate a layout with the best possible configuration | |
4915 | * and backing off gradually. | |
4916 | * | |
4917 | * We start with accepting at most 1/16 waste and try to find the | |
671776b3 XS |
4918 | * smallest order from min_objects-derived/slab_min_order up to |
4919 | * slab_max_order that will satisfy the constraint. Note that increasing | |
0fe2735d VB |
4920 | * the order can only result in same or less fractional waste, not more. |
4921 | * | |
4922 | * If that fails, we increase the acceptable fraction of waste and try | |
5886fc82 VB |
4923 | * again. The last iteration with fraction of 1/2 would effectively |
4924 | * accept any waste and give us the order determined by min_objects, as | |
671776b3 | 4925 | * long as at least single object fits within slab_max_order. |
0fe2735d | 4926 | */ |
5886fc82 | 4927 | for (unsigned int fraction = 16; fraction > 1; fraction /= 2) { |
90f055df | 4928 | order = calc_slab_order(size, min_order, slub_max_order, |
0fe2735d VB |
4929 | fraction); |
4930 | if (order <= slub_max_order) | |
4931 | return order; | |
5e6d444e CL |
4932 | } |
4933 | ||
5e6d444e | 4934 | /* |
671776b3 | 4935 | * Doh this slab cannot be placed using slab_max_order. |
5e6d444e | 4936 | */ |
c7355d75 | 4937 | order = get_order(size); |
5e0a760b | 4938 | if (order <= MAX_PAGE_ORDER) |
5e6d444e CL |
4939 | return order; |
4940 | return -ENOSYS; | |
4941 | } | |
4942 | ||
5595cffc | 4943 | static void |
4053497d | 4944 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
4945 | { |
4946 | n->nr_partial = 0; | |
81819f0f CL |
4947 | spin_lock_init(&n->list_lock); |
4948 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 4949 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 4950 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 4951 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 4952 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 4953 | #endif |
81819f0f CL |
4954 | } |
4955 | ||
0af8489b | 4956 | #ifndef CONFIG_SLUB_TINY |
55136592 | 4957 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 4958 | { |
6c182dc0 | 4959 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
a0dc161a BH |
4960 | NR_KMALLOC_TYPES * KMALLOC_SHIFT_HIGH * |
4961 | sizeof(struct kmem_cache_cpu)); | |
4c93c355 | 4962 | |
8a5ec0ba | 4963 | /* |
d4d84fef CM |
4964 | * Must align to double word boundary for the double cmpxchg |
4965 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 4966 | */ |
d4d84fef CM |
4967 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
4968 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
4969 | |
4970 | if (!s->cpu_slab) | |
4971 | return 0; | |
4972 | ||
4973 | init_kmem_cache_cpus(s); | |
4c93c355 | 4974 | |
8a5ec0ba | 4975 | return 1; |
4c93c355 | 4976 | } |
0af8489b VB |
4977 | #else |
4978 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) | |
4979 | { | |
4980 | return 1; | |
4981 | } | |
4982 | #endif /* CONFIG_SLUB_TINY */ | |
4c93c355 | 4983 | |
51df1142 CL |
4984 | static struct kmem_cache *kmem_cache_node; |
4985 | ||
81819f0f CL |
4986 | /* |
4987 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
4988 | * slab on the node for this slabcache. There are no concurrent accesses | |
4989 | * possible. | |
4990 | * | |
721ae22a ZYW |
4991 | * Note that this function only works on the kmem_cache_node |
4992 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 4993 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 4994 | */ |
55136592 | 4995 | static void early_kmem_cache_node_alloc(int node) |
81819f0f | 4996 | { |
bb192ed9 | 4997 | struct slab *slab; |
81819f0f CL |
4998 | struct kmem_cache_node *n; |
4999 | ||
51df1142 | 5000 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 5001 | |
bb192ed9 | 5002 | slab = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f | 5003 | |
bb192ed9 VB |
5004 | BUG_ON(!slab); |
5005 | if (slab_nid(slab) != node) { | |
f9f58285 FF |
5006 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
5007 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
5008 | } |
5009 | ||
bb192ed9 | 5010 | n = slab->freelist; |
81819f0f | 5011 | BUG_ON(!n); |
8ab1372f | 5012 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 5013 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
8ab1372f | 5014 | #endif |
da844b78 | 5015 | n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false); |
bb192ed9 VB |
5016 | slab->freelist = get_freepointer(kmem_cache_node, n); |
5017 | slab->inuse = 1; | |
12b22386 | 5018 | kmem_cache_node->node[node] = n; |
4053497d | 5019 | init_kmem_cache_node(n); |
bb192ed9 | 5020 | inc_slabs_node(kmem_cache_node, node, slab->objects); |
6446faa2 | 5021 | |
67b6c900 | 5022 | /* |
1e4dd946 SR |
5023 | * No locks need to be taken here as it has just been |
5024 | * initialized and there is no concurrent access. | |
67b6c900 | 5025 | */ |
bb192ed9 | 5026 | __add_partial(n, slab, DEACTIVATE_TO_HEAD); |
81819f0f CL |
5027 | } |
5028 | ||
5029 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
5030 | { | |
5031 | int node; | |
fa45dc25 | 5032 | struct kmem_cache_node *n; |
81819f0f | 5033 | |
fa45dc25 | 5034 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 5035 | s->node[node] = NULL; |
ea37df54 | 5036 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
5037 | } |
5038 | } | |
5039 | ||
52b4b950 DS |
5040 | void __kmem_cache_release(struct kmem_cache *s) |
5041 | { | |
210e7a43 | 5042 | cache_random_seq_destroy(s); |
0af8489b | 5043 | #ifndef CONFIG_SLUB_TINY |
52b4b950 | 5044 | free_percpu(s->cpu_slab); |
0af8489b | 5045 | #endif |
52b4b950 DS |
5046 | free_kmem_cache_nodes(s); |
5047 | } | |
5048 | ||
55136592 | 5049 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
5050 | { |
5051 | int node; | |
81819f0f | 5052 | |
7e1fa93d | 5053 | for_each_node_mask(node, slab_nodes) { |
81819f0f CL |
5054 | struct kmem_cache_node *n; |
5055 | ||
73367bd8 | 5056 | if (slab_state == DOWN) { |
55136592 | 5057 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
5058 | continue; |
5059 | } | |
51df1142 | 5060 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 5061 | GFP_KERNEL, node); |
81819f0f | 5062 | |
73367bd8 AD |
5063 | if (!n) { |
5064 | free_kmem_cache_nodes(s); | |
5065 | return 0; | |
81819f0f | 5066 | } |
73367bd8 | 5067 | |
4053497d | 5068 | init_kmem_cache_node(n); |
ea37df54 | 5069 | s->node[node] = n; |
81819f0f CL |
5070 | } |
5071 | return 1; | |
5072 | } | |
81819f0f | 5073 | |
e6d0e1dc WY |
5074 | static void set_cpu_partial(struct kmem_cache *s) |
5075 | { | |
5076 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
b47291ef VB |
5077 | unsigned int nr_objects; |
5078 | ||
e6d0e1dc WY |
5079 | /* |
5080 | * cpu_partial determined the maximum number of objects kept in the | |
5081 | * per cpu partial lists of a processor. | |
5082 | * | |
5083 | * Per cpu partial lists mainly contain slabs that just have one | |
5084 | * object freed. If they are used for allocation then they can be | |
5085 | * filled up again with minimal effort. The slab will never hit the | |
5086 | * per node partial lists and therefore no locking will be required. | |
5087 | * | |
b47291ef VB |
5088 | * For backwards compatibility reasons, this is determined as number |
5089 | * of objects, even though we now limit maximum number of pages, see | |
5090 | * slub_set_cpu_partial() | |
e6d0e1dc WY |
5091 | */ |
5092 | if (!kmem_cache_has_cpu_partial(s)) | |
b47291ef | 5093 | nr_objects = 0; |
e6d0e1dc | 5094 | else if (s->size >= PAGE_SIZE) |
b47291ef | 5095 | nr_objects = 6; |
e6d0e1dc | 5096 | else if (s->size >= 1024) |
23e98ad1 | 5097 | nr_objects = 24; |
e6d0e1dc | 5098 | else if (s->size >= 256) |
23e98ad1 | 5099 | nr_objects = 52; |
e6d0e1dc | 5100 | else |
23e98ad1 | 5101 | nr_objects = 120; |
b47291ef VB |
5102 | |
5103 | slub_set_cpu_partial(s, nr_objects); | |
e6d0e1dc WY |
5104 | #endif |
5105 | } | |
5106 | ||
81819f0f CL |
5107 | /* |
5108 | * calculate_sizes() determines the order and the distribution of data within | |
5109 | * a slab object. | |
5110 | */ | |
ae44d81d | 5111 | static int calculate_sizes(struct kmem_cache *s) |
81819f0f | 5112 | { |
d50112ed | 5113 | slab_flags_t flags = s->flags; |
be4a7988 | 5114 | unsigned int size = s->object_size; |
19af27af | 5115 | unsigned int order; |
81819f0f | 5116 | |
d8b42bf5 CL |
5117 | /* |
5118 | * Round up object size to the next word boundary. We can only | |
5119 | * place the free pointer at word boundaries and this determines | |
5120 | * the possible location of the free pointer. | |
5121 | */ | |
5122 | size = ALIGN(size, sizeof(void *)); | |
5123 | ||
5124 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
5125 | /* |
5126 | * Determine if we can poison the object itself. If the user of | |
5127 | * the slab may touch the object after free or before allocation | |
5128 | * then we should never poison the object itself. | |
5129 | */ | |
5f0d5a3a | 5130 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 5131 | !s->ctor) |
81819f0f CL |
5132 | s->flags |= __OBJECT_POISON; |
5133 | else | |
5134 | s->flags &= ~__OBJECT_POISON; | |
5135 | ||
81819f0f CL |
5136 | |
5137 | /* | |
672bba3a | 5138 | * If we are Redzoning then check if there is some space between the |
81819f0f | 5139 | * end of the object and the free pointer. If not then add an |
672bba3a | 5140 | * additional word to have some bytes to store Redzone information. |
81819f0f | 5141 | */ |
3b0efdfa | 5142 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 5143 | size += sizeof(void *); |
41ecc55b | 5144 | #endif |
81819f0f CL |
5145 | |
5146 | /* | |
672bba3a | 5147 | * With that we have determined the number of bytes in actual use |
e41a49fa | 5148 | * by the object and redzoning. |
81819f0f CL |
5149 | */ |
5150 | s->inuse = size; | |
5151 | ||
946fa0db FT |
5152 | if (slub_debug_orig_size(s) || |
5153 | (flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || | |
74c1d3e0 KC |
5154 | ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) || |
5155 | s->ctor) { | |
81819f0f CL |
5156 | /* |
5157 | * Relocate free pointer after the object if it is not | |
5158 | * permitted to overwrite the first word of the object on | |
5159 | * kmem_cache_free. | |
5160 | * | |
5161 | * This is the case if we do RCU, have a constructor or | |
74c1d3e0 KC |
5162 | * destructor, are poisoning the objects, or are |
5163 | * redzoning an object smaller than sizeof(void *). | |
cbfc35a4 WL |
5164 | * |
5165 | * The assumption that s->offset >= s->inuse means free | |
5166 | * pointer is outside of the object is used in the | |
5167 | * freeptr_outside_object() function. If that is no | |
5168 | * longer true, the function needs to be modified. | |
81819f0f CL |
5169 | */ |
5170 | s->offset = size; | |
5171 | size += sizeof(void *); | |
e41a49fa | 5172 | } else { |
3202fa62 KC |
5173 | /* |
5174 | * Store freelist pointer near middle of object to keep | |
5175 | * it away from the edges of the object to avoid small | |
5176 | * sized over/underflows from neighboring allocations. | |
5177 | */ | |
e41a49fa | 5178 | s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *)); |
81819f0f CL |
5179 | } |
5180 | ||
c12b3c62 | 5181 | #ifdef CONFIG_SLUB_DEBUG |
6edf2576 | 5182 | if (flags & SLAB_STORE_USER) { |
81819f0f CL |
5183 | /* |
5184 | * Need to store information about allocs and frees after | |
5185 | * the object. | |
5186 | */ | |
5187 | size += 2 * sizeof(struct track); | |
6edf2576 FT |
5188 | |
5189 | /* Save the original kmalloc request size */ | |
5190 | if (flags & SLAB_KMALLOC) | |
5191 | size += sizeof(unsigned int); | |
5192 | } | |
80a9201a | 5193 | #endif |
81819f0f | 5194 | |
80a9201a AP |
5195 | kasan_cache_create(s, &size, &s->flags); |
5196 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 5197 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
5198 | /* |
5199 | * Add some empty padding so that we can catch | |
5200 | * overwrites from earlier objects rather than let | |
5201 | * tracking information or the free pointer be | |
0211a9c8 | 5202 | * corrupted if a user writes before the start |
81819f0f CL |
5203 | * of the object. |
5204 | */ | |
5205 | size += sizeof(void *); | |
d86bd1be JK |
5206 | |
5207 | s->red_left_pad = sizeof(void *); | |
5208 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
5209 | size += s->red_left_pad; | |
5210 | } | |
41ecc55b | 5211 | #endif |
672bba3a | 5212 | |
81819f0f CL |
5213 | /* |
5214 | * SLUB stores one object immediately after another beginning from | |
5215 | * offset 0. In order to align the objects we have to simply size | |
5216 | * each object to conform to the alignment. | |
5217 | */ | |
45906855 | 5218 | size = ALIGN(size, s->align); |
81819f0f | 5219 | s->size = size; |
4138fdfc | 5220 | s->reciprocal_size = reciprocal_value(size); |
ae44d81d | 5221 | order = calculate_order(size); |
81819f0f | 5222 | |
19af27af | 5223 | if ((int)order < 0) |
81819f0f CL |
5224 | return 0; |
5225 | ||
5b15f3fb | 5226 | s->allocflags = __GFP_COMP; |
b7a49f0d CL |
5227 | |
5228 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 5229 | s->allocflags |= GFP_DMA; |
b7a49f0d | 5230 | |
6d6ea1e9 NB |
5231 | if (s->flags & SLAB_CACHE_DMA32) |
5232 | s->allocflags |= GFP_DMA32; | |
5233 | ||
b7a49f0d CL |
5234 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5235 | s->allocflags |= __GFP_RECLAIMABLE; | |
5236 | ||
81819f0f CL |
5237 | /* |
5238 | * Determine the number of objects per slab | |
5239 | */ | |
9736d2a9 MW |
5240 | s->oo = oo_make(order, size); |
5241 | s->min = oo_make(get_order(size), size); | |
81819f0f | 5242 | |
834f3d11 | 5243 | return !!oo_objects(s->oo); |
81819f0f CL |
5244 | } |
5245 | ||
d50112ed | 5246 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 5247 | { |
303cd693 | 5248 | s->flags = kmem_cache_flags(flags, s->name); |
2482ddec KC |
5249 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
5250 | s->random = get_random_long(); | |
5251 | #endif | |
81819f0f | 5252 | |
ae44d81d | 5253 | if (!calculate_sizes(s)) |
81819f0f | 5254 | goto error; |
3de47213 DR |
5255 | if (disable_higher_order_debug) { |
5256 | /* | |
5257 | * Disable debugging flags that store metadata if the min slab | |
5258 | * order increased. | |
5259 | */ | |
3b0efdfa | 5260 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
5261 | s->flags &= ~DEBUG_METADATA_FLAGS; |
5262 | s->offset = 0; | |
ae44d81d | 5263 | if (!calculate_sizes(s)) |
3de47213 DR |
5264 | goto error; |
5265 | } | |
5266 | } | |
81819f0f | 5267 | |
6801be4f PZ |
5268 | #ifdef system_has_freelist_aba |
5269 | if (system_has_freelist_aba() && !(s->flags & SLAB_NO_CMPXCHG)) { | |
b789ef51 CL |
5270 | /* Enable fast mode */ |
5271 | s->flags |= __CMPXCHG_DOUBLE; | |
6801be4f | 5272 | } |
b789ef51 CL |
5273 | #endif |
5274 | ||
3b89d7d8 | 5275 | /* |
c2092c12 | 5276 | * The larger the object size is, the more slabs we want on the partial |
3b89d7d8 DR |
5277 | * list to avoid pounding the page allocator excessively. |
5278 | */ | |
5182f3c9 HY |
5279 | s->min_partial = min_t(unsigned long, MAX_PARTIAL, ilog2(s->size) / 2); |
5280 | s->min_partial = max_t(unsigned long, MIN_PARTIAL, s->min_partial); | |
49e22585 | 5281 | |
e6d0e1dc | 5282 | set_cpu_partial(s); |
49e22585 | 5283 | |
81819f0f | 5284 | #ifdef CONFIG_NUMA |
e2cb96b7 | 5285 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 5286 | #endif |
210e7a43 TG |
5287 | |
5288 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
5289 | if (slab_state >= UP) { | |
5290 | if (init_cache_random_seq(s)) | |
5291 | goto error; | |
5292 | } | |
5293 | ||
55136592 | 5294 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 5295 | goto error; |
81819f0f | 5296 | |
55136592 | 5297 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 5298 | return 0; |
ff12059e | 5299 | |
81819f0f | 5300 | error: |
9037c576 | 5301 | __kmem_cache_release(s); |
278b1bb1 | 5302 | return -EINVAL; |
81819f0f | 5303 | } |
81819f0f | 5304 | |
bb192ed9 | 5305 | static void list_slab_objects(struct kmem_cache *s, struct slab *slab, |
55860d96 | 5306 | const char *text) |
33b12c38 CL |
5307 | { |
5308 | #ifdef CONFIG_SLUB_DEBUG | |
bb192ed9 | 5309 | void *addr = slab_address(slab); |
33b12c38 | 5310 | void *p; |
aa456c7a | 5311 | |
bb192ed9 | 5312 | slab_err(s, slab, text, s->name); |
33b12c38 | 5313 | |
4ef3f5a3 VB |
5314 | spin_lock(&object_map_lock); |
5315 | __fill_map(object_map, s, slab); | |
5316 | ||
bb192ed9 | 5317 | for_each_object(p, s, addr, slab->objects) { |
33b12c38 | 5318 | |
4ef3f5a3 | 5319 | if (!test_bit(__obj_to_index(s, addr, p), object_map)) { |
96b94abc | 5320 | pr_err("Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
5321 | print_tracking(s, p); |
5322 | } | |
5323 | } | |
4ef3f5a3 | 5324 | spin_unlock(&object_map_lock); |
33b12c38 CL |
5325 | #endif |
5326 | } | |
5327 | ||
81819f0f | 5328 | /* |
599870b1 | 5329 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
5330 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
5331 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 5332 | */ |
599870b1 | 5333 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 5334 | { |
60398923 | 5335 | LIST_HEAD(discard); |
bb192ed9 | 5336 | struct slab *slab, *h; |
81819f0f | 5337 | |
52b4b950 DS |
5338 | BUG_ON(irqs_disabled()); |
5339 | spin_lock_irq(&n->list_lock); | |
bb192ed9 VB |
5340 | list_for_each_entry_safe(slab, h, &n->partial, slab_list) { |
5341 | if (!slab->inuse) { | |
5342 | remove_partial(n, slab); | |
5343 | list_add(&slab->slab_list, &discard); | |
33b12c38 | 5344 | } else { |
bb192ed9 | 5345 | list_slab_objects(s, slab, |
55860d96 | 5346 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 5347 | } |
33b12c38 | 5348 | } |
52b4b950 | 5349 | spin_unlock_irq(&n->list_lock); |
60398923 | 5350 | |
bb192ed9 VB |
5351 | list_for_each_entry_safe(slab, h, &discard, slab_list) |
5352 | discard_slab(s, slab); | |
81819f0f CL |
5353 | } |
5354 | ||
f9e13c0a SB |
5355 | bool __kmem_cache_empty(struct kmem_cache *s) |
5356 | { | |
5357 | int node; | |
5358 | struct kmem_cache_node *n; | |
5359 | ||
5360 | for_each_kmem_cache_node(s, node, n) | |
4f174a8b | 5361 | if (n->nr_partial || node_nr_slabs(n)) |
f9e13c0a SB |
5362 | return false; |
5363 | return true; | |
5364 | } | |
5365 | ||
81819f0f | 5366 | /* |
672bba3a | 5367 | * Release all resources used by a slab cache. |
81819f0f | 5368 | */ |
52b4b950 | 5369 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
5370 | { |
5371 | int node; | |
fa45dc25 | 5372 | struct kmem_cache_node *n; |
81819f0f | 5373 | |
5a836bf6 | 5374 | flush_all_cpus_locked(s); |
81819f0f | 5375 | /* Attempt to free all objects */ |
fa45dc25 | 5376 | for_each_kmem_cache_node(s, node, n) { |
599870b1 | 5377 | free_partial(s, n); |
4f174a8b | 5378 | if (n->nr_partial || node_nr_slabs(n)) |
81819f0f CL |
5379 | return 1; |
5380 | } | |
81819f0f CL |
5381 | return 0; |
5382 | } | |
5383 | ||
5bb1bb35 | 5384 | #ifdef CONFIG_PRINTK |
2dfe63e6 | 5385 | void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab) |
8e7f37f2 PM |
5386 | { |
5387 | void *base; | |
5388 | int __maybe_unused i; | |
5389 | unsigned int objnr; | |
5390 | void *objp; | |
5391 | void *objp0; | |
7213230a | 5392 | struct kmem_cache *s = slab->slab_cache; |
8e7f37f2 PM |
5393 | struct track __maybe_unused *trackp; |
5394 | ||
5395 | kpp->kp_ptr = object; | |
7213230a | 5396 | kpp->kp_slab = slab; |
8e7f37f2 | 5397 | kpp->kp_slab_cache = s; |
7213230a | 5398 | base = slab_address(slab); |
8e7f37f2 PM |
5399 | objp0 = kasan_reset_tag(object); |
5400 | #ifdef CONFIG_SLUB_DEBUG | |
5401 | objp = restore_red_left(s, objp0); | |
5402 | #else | |
5403 | objp = objp0; | |
5404 | #endif | |
40f3bf0c | 5405 | objnr = obj_to_index(s, slab, objp); |
8e7f37f2 PM |
5406 | kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp); |
5407 | objp = base + s->size * objnr; | |
5408 | kpp->kp_objp = objp; | |
7213230a MWO |
5409 | if (WARN_ON_ONCE(objp < base || objp >= base + slab->objects * s->size |
5410 | || (objp - base) % s->size) || | |
8e7f37f2 PM |
5411 | !(s->flags & SLAB_STORE_USER)) |
5412 | return; | |
5413 | #ifdef CONFIG_SLUB_DEBUG | |
0cbc124b | 5414 | objp = fixup_red_left(s, objp); |
8e7f37f2 PM |
5415 | trackp = get_track(s, objp, TRACK_ALLOC); |
5416 | kpp->kp_ret = (void *)trackp->addr; | |
5cf909c5 OG |
5417 | #ifdef CONFIG_STACKDEPOT |
5418 | { | |
5419 | depot_stack_handle_t handle; | |
5420 | unsigned long *entries; | |
5421 | unsigned int nr_entries; | |
78869146 | 5422 | |
5cf909c5 OG |
5423 | handle = READ_ONCE(trackp->handle); |
5424 | if (handle) { | |
5425 | nr_entries = stack_depot_fetch(handle, &entries); | |
5426 | for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++) | |
5427 | kpp->kp_stack[i] = (void *)entries[i]; | |
5428 | } | |
78869146 | 5429 | |
5cf909c5 OG |
5430 | trackp = get_track(s, objp, TRACK_FREE); |
5431 | handle = READ_ONCE(trackp->handle); | |
5432 | if (handle) { | |
5433 | nr_entries = stack_depot_fetch(handle, &entries); | |
5434 | for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++) | |
5435 | kpp->kp_free_stack[i] = (void *)entries[i]; | |
5436 | } | |
e548eaa1 | 5437 | } |
8e7f37f2 PM |
5438 | #endif |
5439 | #endif | |
5440 | } | |
5bb1bb35 | 5441 | #endif |
8e7f37f2 | 5442 | |
81819f0f CL |
5443 | /******************************************************************** |
5444 | * Kmalloc subsystem | |
5445 | *******************************************************************/ | |
5446 | ||
81819f0f CL |
5447 | static int __init setup_slub_min_order(char *str) |
5448 | { | |
19af27af | 5449 | get_option(&str, (int *)&slub_min_order); |
81819f0f | 5450 | |
e519ce7a FT |
5451 | if (slub_min_order > slub_max_order) |
5452 | slub_max_order = slub_min_order; | |
5453 | ||
81819f0f CL |
5454 | return 1; |
5455 | } | |
5456 | ||
671776b3 XS |
5457 | __setup("slab_min_order=", setup_slub_min_order); |
5458 | __setup_param("slub_min_order=", slub_min_order, setup_slub_min_order, 0); | |
5459 | ||
81819f0f CL |
5460 | |
5461 | static int __init setup_slub_max_order(char *str) | |
5462 | { | |
19af27af | 5463 | get_option(&str, (int *)&slub_max_order); |
5e0a760b | 5464 | slub_max_order = min_t(unsigned int, slub_max_order, MAX_PAGE_ORDER); |
81819f0f | 5465 | |
e519ce7a FT |
5466 | if (slub_min_order > slub_max_order) |
5467 | slub_min_order = slub_max_order; | |
5468 | ||
81819f0f CL |
5469 | return 1; |
5470 | } | |
5471 | ||
671776b3 XS |
5472 | __setup("slab_max_order=", setup_slub_max_order); |
5473 | __setup_param("slub_max_order=", slub_max_order, setup_slub_max_order, 0); | |
81819f0f CL |
5474 | |
5475 | static int __init setup_slub_min_objects(char *str) | |
5476 | { | |
19af27af | 5477 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
5478 | |
5479 | return 1; | |
5480 | } | |
5481 | ||
671776b3 XS |
5482 | __setup("slab_min_objects=", setup_slub_min_objects); |
5483 | __setup_param("slub_min_objects=", slub_min_objects, setup_slub_min_objects, 0); | |
81819f0f | 5484 | |
ed18adc1 KC |
5485 | #ifdef CONFIG_HARDENED_USERCOPY |
5486 | /* | |
afcc90f8 KC |
5487 | * Rejects incorrectly sized objects and objects that are to be copied |
5488 | * to/from userspace but do not fall entirely within the containing slab | |
5489 | * cache's usercopy region. | |
ed18adc1 KC |
5490 | * |
5491 | * Returns NULL if check passes, otherwise const char * to name of cache | |
5492 | * to indicate an error. | |
5493 | */ | |
0b3eb091 MWO |
5494 | void __check_heap_object(const void *ptr, unsigned long n, |
5495 | const struct slab *slab, bool to_user) | |
ed18adc1 KC |
5496 | { |
5497 | struct kmem_cache *s; | |
44065b2e | 5498 | unsigned int offset; |
b89fb5ef | 5499 | bool is_kfence = is_kfence_address(ptr); |
ed18adc1 | 5500 | |
96fedce2 AK |
5501 | ptr = kasan_reset_tag(ptr); |
5502 | ||
ed18adc1 | 5503 | /* Find object and usable object size. */ |
0b3eb091 | 5504 | s = slab->slab_cache; |
ed18adc1 KC |
5505 | |
5506 | /* Reject impossible pointers. */ | |
0b3eb091 | 5507 | if (ptr < slab_address(slab)) |
f4e6e289 KC |
5508 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
5509 | to_user, 0, n); | |
ed18adc1 KC |
5510 | |
5511 | /* Find offset within object. */ | |
b89fb5ef AP |
5512 | if (is_kfence) |
5513 | offset = ptr - kfence_object_start(ptr); | |
5514 | else | |
0b3eb091 | 5515 | offset = (ptr - slab_address(slab)) % s->size; |
ed18adc1 KC |
5516 | |
5517 | /* Adjust for redzone and reject if within the redzone. */ | |
b89fb5ef | 5518 | if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) { |
ed18adc1 | 5519 | if (offset < s->red_left_pad) |
f4e6e289 KC |
5520 | usercopy_abort("SLUB object in left red zone", |
5521 | s->name, to_user, offset, n); | |
ed18adc1 KC |
5522 | offset -= s->red_left_pad; |
5523 | } | |
5524 | ||
afcc90f8 KC |
5525 | /* Allow address range falling entirely within usercopy region. */ |
5526 | if (offset >= s->useroffset && | |
5527 | offset - s->useroffset <= s->usersize && | |
5528 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 5529 | return; |
ed18adc1 | 5530 | |
f4e6e289 | 5531 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
5532 | } |
5533 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
5534 | ||
832f37f5 VD |
5535 | #define SHRINK_PROMOTE_MAX 32 |
5536 | ||
2086d26a | 5537 | /* |
832f37f5 VD |
5538 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
5539 | * up most to the head of the partial lists. New allocations will then | |
5540 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
5541 | * |
5542 | * The slabs with the least items are placed last. This results in them | |
5543 | * being allocated from last increasing the chance that the last objects | |
5544 | * are freed in them. | |
2086d26a | 5545 | */ |
5a836bf6 | 5546 | static int __kmem_cache_do_shrink(struct kmem_cache *s) |
2086d26a CL |
5547 | { |
5548 | int node; | |
5549 | int i; | |
5550 | struct kmem_cache_node *n; | |
bb192ed9 VB |
5551 | struct slab *slab; |
5552 | struct slab *t; | |
832f37f5 VD |
5553 | struct list_head discard; |
5554 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 5555 | unsigned long flags; |
ce3712d7 | 5556 | int ret = 0; |
2086d26a | 5557 | |
fa45dc25 | 5558 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
5559 | INIT_LIST_HEAD(&discard); |
5560 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
5561 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
5562 | |
5563 | spin_lock_irqsave(&n->list_lock, flags); | |
5564 | ||
5565 | /* | |
832f37f5 | 5566 | * Build lists of slabs to discard or promote. |
2086d26a | 5567 | * |
672bba3a | 5568 | * Note that concurrent frees may occur while we hold the |
c2092c12 | 5569 | * list_lock. slab->inuse here is the upper limit. |
2086d26a | 5570 | */ |
bb192ed9 VB |
5571 | list_for_each_entry_safe(slab, t, &n->partial, slab_list) { |
5572 | int free = slab->objects - slab->inuse; | |
832f37f5 | 5573 | |
c2092c12 | 5574 | /* Do not reread slab->inuse */ |
832f37f5 VD |
5575 | barrier(); |
5576 | ||
5577 | /* We do not keep full slabs on the list */ | |
5578 | BUG_ON(free <= 0); | |
5579 | ||
bb192ed9 VB |
5580 | if (free == slab->objects) { |
5581 | list_move(&slab->slab_list, &discard); | |
8a399e2f | 5582 | slab_clear_node_partial(slab); |
69cb8e6b | 5583 | n->nr_partial--; |
c7323a5a | 5584 | dec_slabs_node(s, node, slab->objects); |
832f37f5 | 5585 | } else if (free <= SHRINK_PROMOTE_MAX) |
bb192ed9 | 5586 | list_move(&slab->slab_list, promote + free - 1); |
2086d26a CL |
5587 | } |
5588 | ||
2086d26a | 5589 | /* |
832f37f5 VD |
5590 | * Promote the slabs filled up most to the head of the |
5591 | * partial list. | |
2086d26a | 5592 | */ |
832f37f5 VD |
5593 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
5594 | list_splice(promote + i, &n->partial); | |
2086d26a | 5595 | |
2086d26a | 5596 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
5597 | |
5598 | /* Release empty slabs */ | |
bb192ed9 | 5599 | list_for_each_entry_safe(slab, t, &discard, slab_list) |
c7323a5a | 5600 | free_slab(s, slab); |
ce3712d7 | 5601 | |
4f174a8b | 5602 | if (node_nr_slabs(n)) |
ce3712d7 | 5603 | ret = 1; |
2086d26a CL |
5604 | } |
5605 | ||
ce3712d7 | 5606 | return ret; |
2086d26a | 5607 | } |
2086d26a | 5608 | |
5a836bf6 SAS |
5609 | int __kmem_cache_shrink(struct kmem_cache *s) |
5610 | { | |
5611 | flush_all(s); | |
5612 | return __kmem_cache_do_shrink(s); | |
5613 | } | |
5614 | ||
b9049e23 YG |
5615 | static int slab_mem_going_offline_callback(void *arg) |
5616 | { | |
5617 | struct kmem_cache *s; | |
5618 | ||
18004c5d | 5619 | mutex_lock(&slab_mutex); |
5a836bf6 SAS |
5620 | list_for_each_entry(s, &slab_caches, list) { |
5621 | flush_all_cpus_locked(s); | |
5622 | __kmem_cache_do_shrink(s); | |
5623 | } | |
18004c5d | 5624 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
5625 | |
5626 | return 0; | |
5627 | } | |
5628 | ||
5629 | static void slab_mem_offline_callback(void *arg) | |
5630 | { | |
b9049e23 YG |
5631 | struct memory_notify *marg = arg; |
5632 | int offline_node; | |
5633 | ||
b9d5ab25 | 5634 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
5635 | |
5636 | /* | |
5637 | * If the node still has available memory. we need kmem_cache_node | |
5638 | * for it yet. | |
5639 | */ | |
5640 | if (offline_node < 0) | |
5641 | return; | |
5642 | ||
18004c5d | 5643 | mutex_lock(&slab_mutex); |
7e1fa93d | 5644 | node_clear(offline_node, slab_nodes); |
666716fd VB |
5645 | /* |
5646 | * We no longer free kmem_cache_node structures here, as it would be | |
5647 | * racy with all get_node() users, and infeasible to protect them with | |
5648 | * slab_mutex. | |
5649 | */ | |
18004c5d | 5650 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
5651 | } |
5652 | ||
5653 | static int slab_mem_going_online_callback(void *arg) | |
5654 | { | |
5655 | struct kmem_cache_node *n; | |
5656 | struct kmem_cache *s; | |
5657 | struct memory_notify *marg = arg; | |
b9d5ab25 | 5658 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
5659 | int ret = 0; |
5660 | ||
5661 | /* | |
5662 | * If the node's memory is already available, then kmem_cache_node is | |
5663 | * already created. Nothing to do. | |
5664 | */ | |
5665 | if (nid < 0) | |
5666 | return 0; | |
5667 | ||
5668 | /* | |
0121c619 | 5669 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
5670 | * allocate a kmem_cache_node structure in order to bring the node |
5671 | * online. | |
5672 | */ | |
18004c5d | 5673 | mutex_lock(&slab_mutex); |
b9049e23 | 5674 | list_for_each_entry(s, &slab_caches, list) { |
666716fd VB |
5675 | /* |
5676 | * The structure may already exist if the node was previously | |
5677 | * onlined and offlined. | |
5678 | */ | |
5679 | if (get_node(s, nid)) | |
5680 | continue; | |
b9049e23 YG |
5681 | /* |
5682 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
5683 | * since memory is not yet available from the node that | |
5684 | * is brought up. | |
5685 | */ | |
8de66a0c | 5686 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
5687 | if (!n) { |
5688 | ret = -ENOMEM; | |
5689 | goto out; | |
5690 | } | |
4053497d | 5691 | init_kmem_cache_node(n); |
b9049e23 YG |
5692 | s->node[nid] = n; |
5693 | } | |
7e1fa93d VB |
5694 | /* |
5695 | * Any cache created after this point will also have kmem_cache_node | |
5696 | * initialized for the new node. | |
5697 | */ | |
5698 | node_set(nid, slab_nodes); | |
b9049e23 | 5699 | out: |
18004c5d | 5700 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
5701 | return ret; |
5702 | } | |
5703 | ||
5704 | static int slab_memory_callback(struct notifier_block *self, | |
5705 | unsigned long action, void *arg) | |
5706 | { | |
5707 | int ret = 0; | |
5708 | ||
5709 | switch (action) { | |
5710 | case MEM_GOING_ONLINE: | |
5711 | ret = slab_mem_going_online_callback(arg); | |
5712 | break; | |
5713 | case MEM_GOING_OFFLINE: | |
5714 | ret = slab_mem_going_offline_callback(arg); | |
5715 | break; | |
5716 | case MEM_OFFLINE: | |
5717 | case MEM_CANCEL_ONLINE: | |
5718 | slab_mem_offline_callback(arg); | |
5719 | break; | |
5720 | case MEM_ONLINE: | |
5721 | case MEM_CANCEL_OFFLINE: | |
5722 | break; | |
5723 | } | |
dc19f9db KH |
5724 | if (ret) |
5725 | ret = notifier_from_errno(ret); | |
5726 | else | |
5727 | ret = NOTIFY_OK; | |
b9049e23 YG |
5728 | return ret; |
5729 | } | |
5730 | ||
81819f0f CL |
5731 | /******************************************************************** |
5732 | * Basic setup of slabs | |
5733 | *******************************************************************/ | |
5734 | ||
51df1142 CL |
5735 | /* |
5736 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
5737 | * the page allocator. Allocate them properly then fix up the pointers |
5738 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
5739 | */ |
5740 | ||
dffb4d60 | 5741 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
5742 | { |
5743 | int node; | |
dffb4d60 | 5744 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 5745 | struct kmem_cache_node *n; |
51df1142 | 5746 | |
dffb4d60 | 5747 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 5748 | |
7d557b3c GC |
5749 | /* |
5750 | * This runs very early, and only the boot processor is supposed to be | |
5751 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
5752 | * IPIs around. | |
5753 | */ | |
5754 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 5755 | for_each_kmem_cache_node(s, node, n) { |
bb192ed9 | 5756 | struct slab *p; |
51df1142 | 5757 | |
916ac052 | 5758 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 5759 | p->slab_cache = s; |
51df1142 | 5760 | |
607bf324 | 5761 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 5762 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 5763 | p->slab_cache = s; |
51df1142 | 5764 | #endif |
51df1142 | 5765 | } |
dffb4d60 CL |
5766 | list_add(&s->list, &slab_caches); |
5767 | return s; | |
51df1142 CL |
5768 | } |
5769 | ||
81819f0f CL |
5770 | void __init kmem_cache_init(void) |
5771 | { | |
dffb4d60 CL |
5772 | static __initdata struct kmem_cache boot_kmem_cache, |
5773 | boot_kmem_cache_node; | |
7e1fa93d | 5774 | int node; |
51df1142 | 5775 | |
fc8d8620 SG |
5776 | if (debug_guardpage_minorder()) |
5777 | slub_max_order = 0; | |
5778 | ||
79270291 SB |
5779 | /* Print slub debugging pointers without hashing */ |
5780 | if (__slub_debug_enabled()) | |
5781 | no_hash_pointers_enable(NULL); | |
5782 | ||
dffb4d60 CL |
5783 | kmem_cache_node = &boot_kmem_cache_node; |
5784 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 5785 | |
7e1fa93d VB |
5786 | /* |
5787 | * Initialize the nodemask for which we will allocate per node | |
5788 | * structures. Here we don't need taking slab_mutex yet. | |
5789 | */ | |
5790 | for_each_node_state(node, N_NORMAL_MEMORY) | |
5791 | node_set(node, slab_nodes); | |
5792 | ||
dffb4d60 | 5793 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
45012241 SB |
5794 | sizeof(struct kmem_cache_node), |
5795 | SLAB_HWCACHE_ALIGN | SLAB_NO_OBJ_EXT, 0, 0); | |
b9049e23 | 5796 | |
946d5f9c | 5797 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); |
81819f0f CL |
5798 | |
5799 | /* Able to allocate the per node structures */ | |
5800 | slab_state = PARTIAL; | |
5801 | ||
dffb4d60 CL |
5802 | create_boot_cache(kmem_cache, "kmem_cache", |
5803 | offsetof(struct kmem_cache, node) + | |
5804 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
45012241 | 5805 | SLAB_HWCACHE_ALIGN | SLAB_NO_OBJ_EXT, 0, 0); |
8a13a4cc | 5806 | |
dffb4d60 | 5807 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 5808 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
5809 | |
5810 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 5811 | setup_kmalloc_cache_index_table(); |
66b3dc1f | 5812 | create_kmalloc_caches(); |
81819f0f | 5813 | |
210e7a43 TG |
5814 | /* Setup random freelists for each cache */ |
5815 | init_freelist_randomization(); | |
5816 | ||
a96a87bf SAS |
5817 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
5818 | slub_cpu_dead); | |
81819f0f | 5819 | |
b9726c26 | 5820 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 5821 | cache_line_size(), |
81819f0f CL |
5822 | slub_min_order, slub_max_order, slub_min_objects, |
5823 | nr_cpu_ids, nr_node_ids); | |
5824 | } | |
5825 | ||
7e85ee0c PE |
5826 | void __init kmem_cache_init_late(void) |
5827 | { | |
0af8489b | 5828 | #ifndef CONFIG_SLUB_TINY |
e45cc288 ML |
5829 | flushwq = alloc_workqueue("slub_flushwq", WQ_MEM_RECLAIM, 0); |
5830 | WARN_ON(!flushwq); | |
0af8489b | 5831 | #endif |
7e85ee0c PE |
5832 | } |
5833 | ||
2633d7a0 | 5834 | struct kmem_cache * |
f4957d5b | 5835 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 5836 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 5837 | { |
10befea9 | 5838 | struct kmem_cache *s; |
81819f0f | 5839 | |
a44cb944 | 5840 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 5841 | if (s) { |
efb93527 XS |
5842 | if (sysfs_slab_alias(s, name)) |
5843 | return NULL; | |
5844 | ||
81819f0f | 5845 | s->refcount++; |
84d0ddd6 | 5846 | |
81819f0f CL |
5847 | /* |
5848 | * Adjust the object sizes so that we clear | |
5849 | * the complete object on kzalloc. | |
5850 | */ | |
1b473f29 | 5851 | s->object_size = max(s->object_size, size); |
52ee6d74 | 5852 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 5853 | } |
6446faa2 | 5854 | |
cbb79694 CL |
5855 | return s; |
5856 | } | |
84c1cf62 | 5857 | |
d50112ed | 5858 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 5859 | { |
aac3a166 PE |
5860 | int err; |
5861 | ||
5862 | err = kmem_cache_open(s, flags); | |
5863 | if (err) | |
5864 | return err; | |
20cea968 | 5865 | |
45530c44 CL |
5866 | /* Mutex is not taken during early boot */ |
5867 | if (slab_state <= UP) | |
5868 | return 0; | |
5869 | ||
aac3a166 | 5870 | err = sysfs_slab_add(s); |
67823a54 | 5871 | if (err) { |
52b4b950 | 5872 | __kmem_cache_release(s); |
67823a54 ML |
5873 | return err; |
5874 | } | |
20cea968 | 5875 | |
64dd6849 FM |
5876 | if (s->flags & SLAB_STORE_USER) |
5877 | debugfs_slab_add(s); | |
5878 | ||
67823a54 | 5879 | return 0; |
81819f0f | 5880 | } |
81819f0f | 5881 | |
b1a413a3 | 5882 | #ifdef SLAB_SUPPORTS_SYSFS |
bb192ed9 | 5883 | static int count_inuse(struct slab *slab) |
205ab99d | 5884 | { |
bb192ed9 | 5885 | return slab->inuse; |
205ab99d CL |
5886 | } |
5887 | ||
bb192ed9 | 5888 | static int count_total(struct slab *slab) |
205ab99d | 5889 | { |
bb192ed9 | 5890 | return slab->objects; |
205ab99d | 5891 | } |
ab4d5ed5 | 5892 | #endif |
205ab99d | 5893 | |
ab4d5ed5 | 5894 | #ifdef CONFIG_SLUB_DEBUG |
bb192ed9 | 5895 | static void validate_slab(struct kmem_cache *s, struct slab *slab, |
0a19e7dd | 5896 | unsigned long *obj_map) |
53e15af0 CL |
5897 | { |
5898 | void *p; | |
bb192ed9 | 5899 | void *addr = slab_address(slab); |
53e15af0 | 5900 | |
bb192ed9 | 5901 | if (!check_slab(s, slab) || !on_freelist(s, slab, NULL)) |
41bec7c3 | 5902 | return; |
53e15af0 CL |
5903 | |
5904 | /* Now we know that a valid freelist exists */ | |
bb192ed9 VB |
5905 | __fill_map(obj_map, s, slab); |
5906 | for_each_object(p, s, addr, slab->objects) { | |
0a19e7dd | 5907 | u8 val = test_bit(__obj_to_index(s, addr, p), obj_map) ? |
dd98afd4 | 5908 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; |
53e15af0 | 5909 | |
bb192ed9 | 5910 | if (!check_object(s, slab, p, val)) |
dd98afd4 YZ |
5911 | break; |
5912 | } | |
53e15af0 CL |
5913 | } |
5914 | ||
434e245d | 5915 | static int validate_slab_node(struct kmem_cache *s, |
0a19e7dd | 5916 | struct kmem_cache_node *n, unsigned long *obj_map) |
53e15af0 CL |
5917 | { |
5918 | unsigned long count = 0; | |
bb192ed9 | 5919 | struct slab *slab; |
53e15af0 CL |
5920 | unsigned long flags; |
5921 | ||
5922 | spin_lock_irqsave(&n->list_lock, flags); | |
5923 | ||
bb192ed9 VB |
5924 | list_for_each_entry(slab, &n->partial, slab_list) { |
5925 | validate_slab(s, slab, obj_map); | |
53e15af0 CL |
5926 | count++; |
5927 | } | |
1f9f78b1 | 5928 | if (count != n->nr_partial) { |
f9f58285 FF |
5929 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
5930 | s->name, count, n->nr_partial); | |
1f9f78b1 OG |
5931 | slab_add_kunit_errors(); |
5932 | } | |
53e15af0 CL |
5933 | |
5934 | if (!(s->flags & SLAB_STORE_USER)) | |
5935 | goto out; | |
5936 | ||
bb192ed9 VB |
5937 | list_for_each_entry(slab, &n->full, slab_list) { |
5938 | validate_slab(s, slab, obj_map); | |
53e15af0 CL |
5939 | count++; |
5940 | } | |
8040cbf5 | 5941 | if (count != node_nr_slabs(n)) { |
f9f58285 | 5942 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
8040cbf5 | 5943 | s->name, count, node_nr_slabs(n)); |
1f9f78b1 OG |
5944 | slab_add_kunit_errors(); |
5945 | } | |
53e15af0 CL |
5946 | |
5947 | out: | |
5948 | spin_unlock_irqrestore(&n->list_lock, flags); | |
5949 | return count; | |
5950 | } | |
5951 | ||
1f9f78b1 | 5952 | long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
5953 | { |
5954 | int node; | |
5955 | unsigned long count = 0; | |
fa45dc25 | 5956 | struct kmem_cache_node *n; |
0a19e7dd VB |
5957 | unsigned long *obj_map; |
5958 | ||
5959 | obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL); | |
5960 | if (!obj_map) | |
5961 | return -ENOMEM; | |
53e15af0 CL |
5962 | |
5963 | flush_all(s); | |
fa45dc25 | 5964 | for_each_kmem_cache_node(s, node, n) |
0a19e7dd VB |
5965 | count += validate_slab_node(s, n, obj_map); |
5966 | ||
5967 | bitmap_free(obj_map); | |
90e9f6a6 | 5968 | |
53e15af0 CL |
5969 | return count; |
5970 | } | |
1f9f78b1 OG |
5971 | EXPORT_SYMBOL(validate_slab_cache); |
5972 | ||
64dd6849 | 5973 | #ifdef CONFIG_DEBUG_FS |
88a420e4 | 5974 | /* |
672bba3a | 5975 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
5976 | * and freed. |
5977 | */ | |
5978 | ||
5979 | struct location { | |
8ea9fb92 | 5980 | depot_stack_handle_t handle; |
88a420e4 | 5981 | unsigned long count; |
ce71e27c | 5982 | unsigned long addr; |
6edf2576 | 5983 | unsigned long waste; |
45edfa58 CL |
5984 | long long sum_time; |
5985 | long min_time; | |
5986 | long max_time; | |
5987 | long min_pid; | |
5988 | long max_pid; | |
174596a0 | 5989 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 5990 | nodemask_t nodes; |
88a420e4 CL |
5991 | }; |
5992 | ||
5993 | struct loc_track { | |
5994 | unsigned long max; | |
5995 | unsigned long count; | |
5996 | struct location *loc; | |
005a79e5 | 5997 | loff_t idx; |
88a420e4 CL |
5998 | }; |
5999 | ||
64dd6849 FM |
6000 | static struct dentry *slab_debugfs_root; |
6001 | ||
88a420e4 CL |
6002 | static void free_loc_track(struct loc_track *t) |
6003 | { | |
6004 | if (t->max) | |
6005 | free_pages((unsigned long)t->loc, | |
6006 | get_order(sizeof(struct location) * t->max)); | |
6007 | } | |
6008 | ||
68dff6a9 | 6009 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
6010 | { |
6011 | struct location *l; | |
6012 | int order; | |
6013 | ||
88a420e4 CL |
6014 | order = get_order(sizeof(struct location) * max); |
6015 | ||
68dff6a9 | 6016 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
6017 | if (!l) |
6018 | return 0; | |
6019 | ||
6020 | if (t->count) { | |
6021 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
6022 | free_loc_track(t); | |
6023 | } | |
6024 | t->max = max; | |
6025 | t->loc = l; | |
6026 | return 1; | |
6027 | } | |
6028 | ||
6029 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
6edf2576 FT |
6030 | const struct track *track, |
6031 | unsigned int orig_size) | |
88a420e4 CL |
6032 | { |
6033 | long start, end, pos; | |
6034 | struct location *l; | |
6edf2576 | 6035 | unsigned long caddr, chandle, cwaste; |
45edfa58 | 6036 | unsigned long age = jiffies - track->when; |
8ea9fb92 | 6037 | depot_stack_handle_t handle = 0; |
6edf2576 | 6038 | unsigned int waste = s->object_size - orig_size; |
88a420e4 | 6039 | |
8ea9fb92 OG |
6040 | #ifdef CONFIG_STACKDEPOT |
6041 | handle = READ_ONCE(track->handle); | |
6042 | #endif | |
88a420e4 CL |
6043 | start = -1; |
6044 | end = t->count; | |
6045 | ||
6046 | for ( ; ; ) { | |
6047 | pos = start + (end - start + 1) / 2; | |
6048 | ||
6049 | /* | |
6050 | * There is nothing at "end". If we end up there | |
6051 | * we need to add something to before end. | |
6052 | */ | |
6053 | if (pos == end) | |
6054 | break; | |
6055 | ||
6edf2576 FT |
6056 | l = &t->loc[pos]; |
6057 | caddr = l->addr; | |
6058 | chandle = l->handle; | |
6059 | cwaste = l->waste; | |
6060 | if ((track->addr == caddr) && (handle == chandle) && | |
6061 | (waste == cwaste)) { | |
45edfa58 | 6062 | |
45edfa58 CL |
6063 | l->count++; |
6064 | if (track->when) { | |
6065 | l->sum_time += age; | |
6066 | if (age < l->min_time) | |
6067 | l->min_time = age; | |
6068 | if (age > l->max_time) | |
6069 | l->max_time = age; | |
6070 | ||
6071 | if (track->pid < l->min_pid) | |
6072 | l->min_pid = track->pid; | |
6073 | if (track->pid > l->max_pid) | |
6074 | l->max_pid = track->pid; | |
6075 | ||
174596a0 RR |
6076 | cpumask_set_cpu(track->cpu, |
6077 | to_cpumask(l->cpus)); | |
45edfa58 CL |
6078 | } |
6079 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
6080 | return 1; |
6081 | } | |
6082 | ||
45edfa58 | 6083 | if (track->addr < caddr) |
88a420e4 | 6084 | end = pos; |
8ea9fb92 OG |
6085 | else if (track->addr == caddr && handle < chandle) |
6086 | end = pos; | |
6edf2576 FT |
6087 | else if (track->addr == caddr && handle == chandle && |
6088 | waste < cwaste) | |
6089 | end = pos; | |
88a420e4 CL |
6090 | else |
6091 | start = pos; | |
6092 | } | |
6093 | ||
6094 | /* | |
672bba3a | 6095 | * Not found. Insert new tracking element. |
88a420e4 | 6096 | */ |
68dff6a9 | 6097 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
6098 | return 0; |
6099 | ||
6100 | l = t->loc + pos; | |
6101 | if (pos < t->count) | |
6102 | memmove(l + 1, l, | |
6103 | (t->count - pos) * sizeof(struct location)); | |
6104 | t->count++; | |
6105 | l->count = 1; | |
45edfa58 CL |
6106 | l->addr = track->addr; |
6107 | l->sum_time = age; | |
6108 | l->min_time = age; | |
6109 | l->max_time = age; | |
6110 | l->min_pid = track->pid; | |
6111 | l->max_pid = track->pid; | |
8ea9fb92 | 6112 | l->handle = handle; |
6edf2576 | 6113 | l->waste = waste; |
174596a0 RR |
6114 | cpumask_clear(to_cpumask(l->cpus)); |
6115 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
6116 | nodes_clear(l->nodes); |
6117 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
6118 | return 1; |
6119 | } | |
6120 | ||
6121 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bb192ed9 | 6122 | struct slab *slab, enum track_item alloc, |
b3fd64e1 | 6123 | unsigned long *obj_map) |
88a420e4 | 6124 | { |
bb192ed9 | 6125 | void *addr = slab_address(slab); |
6edf2576 | 6126 | bool is_alloc = (alloc == TRACK_ALLOC); |
88a420e4 CL |
6127 | void *p; |
6128 | ||
bb192ed9 | 6129 | __fill_map(obj_map, s, slab); |
b3fd64e1 | 6130 | |
bb192ed9 | 6131 | for_each_object(p, s, addr, slab->objects) |
b3fd64e1 | 6132 | if (!test_bit(__obj_to_index(s, addr, p), obj_map)) |
6edf2576 FT |
6133 | add_location(t, s, get_track(s, p, alloc), |
6134 | is_alloc ? get_orig_size(s, p) : | |
6135 | s->object_size); | |
88a420e4 | 6136 | } |
64dd6849 | 6137 | #endif /* CONFIG_DEBUG_FS */ |
6dfd1b65 | 6138 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 6139 | |
b1a413a3 | 6140 | #ifdef SLAB_SUPPORTS_SYSFS |
81819f0f | 6141 | enum slab_stat_type { |
205ab99d CL |
6142 | SL_ALL, /* All slabs */ |
6143 | SL_PARTIAL, /* Only partially allocated slabs */ | |
6144 | SL_CPU, /* Only slabs used for cpu caches */ | |
6145 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
6146 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
6147 | }; |
6148 | ||
205ab99d | 6149 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
6150 | #define SO_PARTIAL (1 << SL_PARTIAL) |
6151 | #define SO_CPU (1 << SL_CPU) | |
6152 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 6153 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 6154 | |
62e5c4b4 | 6155 | static ssize_t show_slab_objects(struct kmem_cache *s, |
bf16d19a | 6156 | char *buf, unsigned long flags) |
81819f0f CL |
6157 | { |
6158 | unsigned long total = 0; | |
81819f0f CL |
6159 | int node; |
6160 | int x; | |
6161 | unsigned long *nodes; | |
bf16d19a | 6162 | int len = 0; |
81819f0f | 6163 | |
6396bb22 | 6164 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
6165 | if (!nodes) |
6166 | return -ENOMEM; | |
81819f0f | 6167 | |
205ab99d CL |
6168 | if (flags & SO_CPU) { |
6169 | int cpu; | |
81819f0f | 6170 | |
205ab99d | 6171 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
6172 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
6173 | cpu); | |
ec3ab083 | 6174 | int node; |
bb192ed9 | 6175 | struct slab *slab; |
dfb4f096 | 6176 | |
bb192ed9 VB |
6177 | slab = READ_ONCE(c->slab); |
6178 | if (!slab) | |
ec3ab083 | 6179 | continue; |
205ab99d | 6180 | |
bb192ed9 | 6181 | node = slab_nid(slab); |
ec3ab083 | 6182 | if (flags & SO_TOTAL) |
bb192ed9 | 6183 | x = slab->objects; |
ec3ab083 | 6184 | else if (flags & SO_OBJECTS) |
bb192ed9 | 6185 | x = slab->inuse; |
ec3ab083 CL |
6186 | else |
6187 | x = 1; | |
49e22585 | 6188 | |
ec3ab083 CL |
6189 | total += x; |
6190 | nodes[node] += x; | |
6191 | ||
9c01e9af | 6192 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
bb192ed9 VB |
6193 | slab = slub_percpu_partial_read_once(c); |
6194 | if (slab) { | |
6195 | node = slab_nid(slab); | |
8afb1474 LZ |
6196 | if (flags & SO_TOTAL) |
6197 | WARN_ON_ONCE(1); | |
6198 | else if (flags & SO_OBJECTS) | |
6199 | WARN_ON_ONCE(1); | |
6200 | else | |
87654cf7 | 6201 | x = data_race(slab->slabs); |
bc6697d8 ED |
6202 | total += x; |
6203 | nodes[node] += x; | |
49e22585 | 6204 | } |
9c01e9af | 6205 | #endif |
81819f0f CL |
6206 | } |
6207 | } | |
6208 | ||
e4f8e513 QC |
6209 | /* |
6210 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
6211 | * already held which will conflict with an existing lock order: | |
6212 | * | |
6213 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
6214 | * | |
6215 | * We don't really need mem_hotplug_lock (to hold off | |
6216 | * slab_mem_going_offline_callback) here because slab's memory hot | |
6217 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
6218 | */ | |
6219 | ||
ab4d5ed5 | 6220 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 6221 | if (flags & SO_ALL) { |
fa45dc25 CL |
6222 | struct kmem_cache_node *n; |
6223 | ||
6224 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 6225 | |
d0e0ac97 | 6226 | if (flags & SO_TOTAL) |
8040cbf5 | 6227 | x = node_nr_objs(n); |
d0e0ac97 | 6228 | else if (flags & SO_OBJECTS) |
8040cbf5 | 6229 | x = node_nr_objs(n) - count_partial(n, count_free); |
81819f0f | 6230 | else |
8040cbf5 | 6231 | x = node_nr_slabs(n); |
81819f0f CL |
6232 | total += x; |
6233 | nodes[node] += x; | |
6234 | } | |
6235 | ||
ab4d5ed5 CL |
6236 | } else |
6237 | #endif | |
6238 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 6239 | struct kmem_cache_node *n; |
81819f0f | 6240 | |
fa45dc25 | 6241 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
6242 | if (flags & SO_TOTAL) |
6243 | x = count_partial(n, count_total); | |
6244 | else if (flags & SO_OBJECTS) | |
6245 | x = count_partial(n, count_inuse); | |
81819f0f | 6246 | else |
205ab99d | 6247 | x = n->nr_partial; |
81819f0f CL |
6248 | total += x; |
6249 | nodes[node] += x; | |
6250 | } | |
6251 | } | |
bf16d19a JP |
6252 | |
6253 | len += sysfs_emit_at(buf, len, "%lu", total); | |
81819f0f | 6254 | #ifdef CONFIG_NUMA |
bf16d19a | 6255 | for (node = 0; node < nr_node_ids; node++) { |
81819f0f | 6256 | if (nodes[node]) |
bf16d19a JP |
6257 | len += sysfs_emit_at(buf, len, " N%d=%lu", |
6258 | node, nodes[node]); | |
6259 | } | |
81819f0f | 6260 | #endif |
bf16d19a | 6261 | len += sysfs_emit_at(buf, len, "\n"); |
81819f0f | 6262 | kfree(nodes); |
bf16d19a JP |
6263 | |
6264 | return len; | |
81819f0f CL |
6265 | } |
6266 | ||
81819f0f | 6267 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) |
497888cf | 6268 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
6269 | |
6270 | struct slab_attribute { | |
6271 | struct attribute attr; | |
6272 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
6273 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
6274 | }; | |
6275 | ||
6276 | #define SLAB_ATTR_RO(_name) \ | |
d1d28bd9 | 6277 | static struct slab_attribute _name##_attr = __ATTR_RO_MODE(_name, 0400) |
81819f0f CL |
6278 | |
6279 | #define SLAB_ATTR(_name) \ | |
d1d28bd9 | 6280 | static struct slab_attribute _name##_attr = __ATTR_RW_MODE(_name, 0600) |
81819f0f | 6281 | |
81819f0f CL |
6282 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
6283 | { | |
bf16d19a | 6284 | return sysfs_emit(buf, "%u\n", s->size); |
81819f0f CL |
6285 | } |
6286 | SLAB_ATTR_RO(slab_size); | |
6287 | ||
6288 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
6289 | { | |
bf16d19a | 6290 | return sysfs_emit(buf, "%u\n", s->align); |
81819f0f CL |
6291 | } |
6292 | SLAB_ATTR_RO(align); | |
6293 | ||
6294 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
6295 | { | |
bf16d19a | 6296 | return sysfs_emit(buf, "%u\n", s->object_size); |
81819f0f CL |
6297 | } |
6298 | SLAB_ATTR_RO(object_size); | |
6299 | ||
6300 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
6301 | { | |
bf16d19a | 6302 | return sysfs_emit(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
6303 | } |
6304 | SLAB_ATTR_RO(objs_per_slab); | |
6305 | ||
6306 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
6307 | { | |
bf16d19a | 6308 | return sysfs_emit(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 6309 | } |
32a6f409 | 6310 | SLAB_ATTR_RO(order); |
81819f0f | 6311 | |
73d342b1 DR |
6312 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
6313 | { | |
bf16d19a | 6314 | return sysfs_emit(buf, "%lu\n", s->min_partial); |
73d342b1 DR |
6315 | } |
6316 | ||
6317 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
6318 | size_t length) | |
6319 | { | |
6320 | unsigned long min; | |
6321 | int err; | |
6322 | ||
3dbb95f7 | 6323 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
6324 | if (err) |
6325 | return err; | |
6326 | ||
5182f3c9 | 6327 | s->min_partial = min; |
73d342b1 DR |
6328 | return length; |
6329 | } | |
6330 | SLAB_ATTR(min_partial); | |
6331 | ||
49e22585 CL |
6332 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
6333 | { | |
b47291ef VB |
6334 | unsigned int nr_partial = 0; |
6335 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
6336 | nr_partial = s->cpu_partial; | |
6337 | #endif | |
6338 | ||
6339 | return sysfs_emit(buf, "%u\n", nr_partial); | |
49e22585 CL |
6340 | } |
6341 | ||
6342 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
6343 | size_t length) | |
6344 | { | |
e5d9998f | 6345 | unsigned int objects; |
49e22585 CL |
6346 | int err; |
6347 | ||
e5d9998f | 6348 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
6349 | if (err) |
6350 | return err; | |
345c905d | 6351 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 6352 | return -EINVAL; |
49e22585 | 6353 | |
e6d0e1dc | 6354 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
6355 | flush_all(s); |
6356 | return length; | |
6357 | } | |
6358 | SLAB_ATTR(cpu_partial); | |
6359 | ||
81819f0f CL |
6360 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
6361 | { | |
62c70bce JP |
6362 | if (!s->ctor) |
6363 | return 0; | |
bf16d19a | 6364 | return sysfs_emit(buf, "%pS\n", s->ctor); |
81819f0f CL |
6365 | } |
6366 | SLAB_ATTR_RO(ctor); | |
6367 | ||
81819f0f CL |
6368 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
6369 | { | |
bf16d19a | 6370 | return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
6371 | } |
6372 | SLAB_ATTR_RO(aliases); | |
6373 | ||
81819f0f CL |
6374 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
6375 | { | |
d9acf4b7 | 6376 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
6377 | } |
6378 | SLAB_ATTR_RO(partial); | |
6379 | ||
6380 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
6381 | { | |
d9acf4b7 | 6382 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
6383 | } |
6384 | SLAB_ATTR_RO(cpu_slabs); | |
6385 | ||
205ab99d CL |
6386 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
6387 | { | |
6388 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
6389 | } | |
6390 | SLAB_ATTR_RO(objects_partial); | |
6391 | ||
49e22585 CL |
6392 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
6393 | { | |
6394 | int objects = 0; | |
bb192ed9 | 6395 | int slabs = 0; |
9c01e9af | 6396 | int cpu __maybe_unused; |
bf16d19a | 6397 | int len = 0; |
49e22585 | 6398 | |
9c01e9af | 6399 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 | 6400 | for_each_online_cpu(cpu) { |
bb192ed9 | 6401 | struct slab *slab; |
a93cf07b | 6402 | |
bb192ed9 | 6403 | slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); |
49e22585 | 6404 | |
bb192ed9 | 6405 | if (slab) |
87654cf7 | 6406 | slabs += data_race(slab->slabs); |
49e22585 | 6407 | } |
9c01e9af | 6408 | #endif |
49e22585 | 6409 | |
c2092c12 | 6410 | /* Approximate half-full slabs, see slub_set_cpu_partial() */ |
bb192ed9 VB |
6411 | objects = (slabs * oo_objects(s->oo)) / 2; |
6412 | len += sysfs_emit_at(buf, len, "%d(%d)", objects, slabs); | |
49e22585 | 6413 | |
c6c17c4d | 6414 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 | 6415 | for_each_online_cpu(cpu) { |
bb192ed9 | 6416 | struct slab *slab; |
a93cf07b | 6417 | |
bb192ed9 VB |
6418 | slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); |
6419 | if (slab) { | |
87654cf7 | 6420 | slabs = data_race(slab->slabs); |
bb192ed9 | 6421 | objects = (slabs * oo_objects(s->oo)) / 2; |
bf16d19a | 6422 | len += sysfs_emit_at(buf, len, " C%d=%d(%d)", |
bb192ed9 | 6423 | cpu, objects, slabs); |
b47291ef | 6424 | } |
49e22585 CL |
6425 | } |
6426 | #endif | |
bf16d19a JP |
6427 | len += sysfs_emit_at(buf, len, "\n"); |
6428 | ||
6429 | return len; | |
49e22585 CL |
6430 | } |
6431 | SLAB_ATTR_RO(slabs_cpu_partial); | |
6432 | ||
a5a84755 CL |
6433 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
6434 | { | |
bf16d19a | 6435 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); |
a5a84755 | 6436 | } |
8f58119a | 6437 | SLAB_ATTR_RO(reclaim_account); |
a5a84755 CL |
6438 | |
6439 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
6440 | { | |
bf16d19a | 6441 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
a5a84755 CL |
6442 | } |
6443 | SLAB_ATTR_RO(hwcache_align); | |
6444 | ||
6445 | #ifdef CONFIG_ZONE_DMA | |
6446 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
6447 | { | |
bf16d19a | 6448 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); |
a5a84755 CL |
6449 | } |
6450 | SLAB_ATTR_RO(cache_dma); | |
6451 | #endif | |
6452 | ||
346907ce | 6453 | #ifdef CONFIG_HARDENED_USERCOPY |
8eb8284b DW |
6454 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
6455 | { | |
bf16d19a | 6456 | return sysfs_emit(buf, "%u\n", s->usersize); |
8eb8284b DW |
6457 | } |
6458 | SLAB_ATTR_RO(usersize); | |
346907ce | 6459 | #endif |
8eb8284b | 6460 | |
a5a84755 CL |
6461 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
6462 | { | |
bf16d19a | 6463 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
6464 | } |
6465 | SLAB_ATTR_RO(destroy_by_rcu); | |
6466 | ||
ab4d5ed5 | 6467 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
6468 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
6469 | { | |
6470 | return show_slab_objects(s, buf, SO_ALL); | |
6471 | } | |
6472 | SLAB_ATTR_RO(slabs); | |
6473 | ||
205ab99d CL |
6474 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
6475 | { | |
6476 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
6477 | } | |
6478 | SLAB_ATTR_RO(total_objects); | |
6479 | ||
81bd3179 XS |
6480 | static ssize_t objects_show(struct kmem_cache *s, char *buf) |
6481 | { | |
6482 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); | |
6483 | } | |
6484 | SLAB_ATTR_RO(objects); | |
6485 | ||
81819f0f CL |
6486 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
6487 | { | |
bf16d19a | 6488 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f | 6489 | } |
060807f8 | 6490 | SLAB_ATTR_RO(sanity_checks); |
81819f0f CL |
6491 | |
6492 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
6493 | { | |
bf16d19a | 6494 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE)); |
81819f0f | 6495 | } |
060807f8 | 6496 | SLAB_ATTR_RO(trace); |
81819f0f | 6497 | |
81819f0f CL |
6498 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
6499 | { | |
bf16d19a | 6500 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); |
81819f0f CL |
6501 | } |
6502 | ||
ad38b5b1 | 6503 | SLAB_ATTR_RO(red_zone); |
81819f0f CL |
6504 | |
6505 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
6506 | { | |
bf16d19a | 6507 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON)); |
81819f0f CL |
6508 | } |
6509 | ||
ad38b5b1 | 6510 | SLAB_ATTR_RO(poison); |
81819f0f CL |
6511 | |
6512 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
6513 | { | |
bf16d19a | 6514 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); |
81819f0f CL |
6515 | } |
6516 | ||
ad38b5b1 | 6517 | SLAB_ATTR_RO(store_user); |
81819f0f | 6518 | |
53e15af0 CL |
6519 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
6520 | { | |
6521 | return 0; | |
6522 | } | |
6523 | ||
6524 | static ssize_t validate_store(struct kmem_cache *s, | |
6525 | const char *buf, size_t length) | |
6526 | { | |
434e245d CL |
6527 | int ret = -EINVAL; |
6528 | ||
c7323a5a | 6529 | if (buf[0] == '1' && kmem_cache_debug(s)) { |
434e245d CL |
6530 | ret = validate_slab_cache(s); |
6531 | if (ret >= 0) | |
6532 | ret = length; | |
6533 | } | |
6534 | return ret; | |
53e15af0 CL |
6535 | } |
6536 | SLAB_ATTR(validate); | |
a5a84755 | 6537 | |
a5a84755 CL |
6538 | #endif /* CONFIG_SLUB_DEBUG */ |
6539 | ||
6540 | #ifdef CONFIG_FAILSLAB | |
6541 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
6542 | { | |
bf16d19a | 6543 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); |
a5a84755 | 6544 | } |
7c82b3b3 AA |
6545 | |
6546 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
6547 | size_t length) | |
6548 | { | |
6549 | if (s->refcount > 1) | |
6550 | return -EINVAL; | |
6551 | ||
6552 | if (buf[0] == '1') | |
6553 | WRITE_ONCE(s->flags, s->flags | SLAB_FAILSLAB); | |
6554 | else | |
6555 | WRITE_ONCE(s->flags, s->flags & ~SLAB_FAILSLAB); | |
6556 | ||
6557 | return length; | |
6558 | } | |
6559 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 6560 | #endif |
53e15af0 | 6561 | |
2086d26a CL |
6562 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
6563 | { | |
6564 | return 0; | |
6565 | } | |
6566 | ||
6567 | static ssize_t shrink_store(struct kmem_cache *s, | |
6568 | const char *buf, size_t length) | |
6569 | { | |
832f37f5 | 6570 | if (buf[0] == '1') |
10befea9 | 6571 | kmem_cache_shrink(s); |
832f37f5 | 6572 | else |
2086d26a CL |
6573 | return -EINVAL; |
6574 | return length; | |
6575 | } | |
6576 | SLAB_ATTR(shrink); | |
6577 | ||
81819f0f | 6578 | #ifdef CONFIG_NUMA |
9824601e | 6579 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 6580 | { |
bf16d19a | 6581 | return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
6582 | } |
6583 | ||
9824601e | 6584 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
6585 | const char *buf, size_t length) |
6586 | { | |
eb7235eb | 6587 | unsigned int ratio; |
0121c619 CL |
6588 | int err; |
6589 | ||
eb7235eb | 6590 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
6591 | if (err) |
6592 | return err; | |
eb7235eb AD |
6593 | if (ratio > 100) |
6594 | return -ERANGE; | |
0121c619 | 6595 | |
eb7235eb | 6596 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 6597 | |
81819f0f CL |
6598 | return length; |
6599 | } | |
9824601e | 6600 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
6601 | #endif |
6602 | ||
8ff12cfc | 6603 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
6604 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
6605 | { | |
6606 | unsigned long sum = 0; | |
6607 | int cpu; | |
bf16d19a | 6608 | int len = 0; |
6da2ec56 | 6609 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
6610 | |
6611 | if (!data) | |
6612 | return -ENOMEM; | |
6613 | ||
6614 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 6615 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
6616 | |
6617 | data[cpu] = x; | |
6618 | sum += x; | |
6619 | } | |
6620 | ||
bf16d19a | 6621 | len += sysfs_emit_at(buf, len, "%lu", sum); |
8ff12cfc | 6622 | |
50ef37b9 | 6623 | #ifdef CONFIG_SMP |
8ff12cfc | 6624 | for_each_online_cpu(cpu) { |
bf16d19a JP |
6625 | if (data[cpu]) |
6626 | len += sysfs_emit_at(buf, len, " C%d=%u", | |
6627 | cpu, data[cpu]); | |
8ff12cfc | 6628 | } |
50ef37b9 | 6629 | #endif |
8ff12cfc | 6630 | kfree(data); |
bf16d19a JP |
6631 | len += sysfs_emit_at(buf, len, "\n"); |
6632 | ||
6633 | return len; | |
8ff12cfc CL |
6634 | } |
6635 | ||
78eb00cc DR |
6636 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
6637 | { | |
6638 | int cpu; | |
6639 | ||
6640 | for_each_online_cpu(cpu) | |
9dfc6e68 | 6641 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
6642 | } |
6643 | ||
8ff12cfc CL |
6644 | #define STAT_ATTR(si, text) \ |
6645 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
6646 | { \ | |
6647 | return show_stat(s, buf, si); \ | |
6648 | } \ | |
78eb00cc DR |
6649 | static ssize_t text##_store(struct kmem_cache *s, \ |
6650 | const char *buf, size_t length) \ | |
6651 | { \ | |
6652 | if (buf[0] != '0') \ | |
6653 | return -EINVAL; \ | |
6654 | clear_stat(s, si); \ | |
6655 | return length; \ | |
6656 | } \ | |
6657 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
6658 | |
6659 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
6660 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
6661 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
6662 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
6663 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
6664 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
6665 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
6666 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
6667 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
6668 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 6669 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
6670 | STAT_ATTR(FREE_SLAB, free_slab); |
6671 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
6672 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
6673 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
6674 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
6675 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
6676 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 6677 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 6678 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
6679 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
6680 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
6681 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
6682 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
6683 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
6684 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 6685 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 6686 | |
b84e04f1 IK |
6687 | #ifdef CONFIG_KFENCE |
6688 | static ssize_t skip_kfence_show(struct kmem_cache *s, char *buf) | |
6689 | { | |
6690 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_SKIP_KFENCE)); | |
6691 | } | |
6692 | ||
6693 | static ssize_t skip_kfence_store(struct kmem_cache *s, | |
6694 | const char *buf, size_t length) | |
6695 | { | |
6696 | int ret = length; | |
6697 | ||
6698 | if (buf[0] == '0') | |
6699 | s->flags &= ~SLAB_SKIP_KFENCE; | |
6700 | else if (buf[0] == '1') | |
6701 | s->flags |= SLAB_SKIP_KFENCE; | |
6702 | else | |
6703 | ret = -EINVAL; | |
6704 | ||
6705 | return ret; | |
6706 | } | |
6707 | SLAB_ATTR(skip_kfence); | |
6708 | #endif | |
6709 | ||
06428780 | 6710 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
6711 | &slab_size_attr.attr, |
6712 | &object_size_attr.attr, | |
6713 | &objs_per_slab_attr.attr, | |
6714 | &order_attr.attr, | |
73d342b1 | 6715 | &min_partial_attr.attr, |
49e22585 | 6716 | &cpu_partial_attr.attr, |
205ab99d | 6717 | &objects_partial_attr.attr, |
81819f0f CL |
6718 | &partial_attr.attr, |
6719 | &cpu_slabs_attr.attr, | |
6720 | &ctor_attr.attr, | |
81819f0f CL |
6721 | &aliases_attr.attr, |
6722 | &align_attr.attr, | |
81819f0f CL |
6723 | &hwcache_align_attr.attr, |
6724 | &reclaim_account_attr.attr, | |
6725 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 6726 | &shrink_attr.attr, |
49e22585 | 6727 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 6728 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 | 6729 | &total_objects_attr.attr, |
81bd3179 | 6730 | &objects_attr.attr, |
a5a84755 CL |
6731 | &slabs_attr.attr, |
6732 | &sanity_checks_attr.attr, | |
6733 | &trace_attr.attr, | |
81819f0f CL |
6734 | &red_zone_attr.attr, |
6735 | &poison_attr.attr, | |
6736 | &store_user_attr.attr, | |
53e15af0 | 6737 | &validate_attr.attr, |
ab4d5ed5 | 6738 | #endif |
81819f0f CL |
6739 | #ifdef CONFIG_ZONE_DMA |
6740 | &cache_dma_attr.attr, | |
6741 | #endif | |
6742 | #ifdef CONFIG_NUMA | |
9824601e | 6743 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
6744 | #endif |
6745 | #ifdef CONFIG_SLUB_STATS | |
6746 | &alloc_fastpath_attr.attr, | |
6747 | &alloc_slowpath_attr.attr, | |
6748 | &free_fastpath_attr.attr, | |
6749 | &free_slowpath_attr.attr, | |
6750 | &free_frozen_attr.attr, | |
6751 | &free_add_partial_attr.attr, | |
6752 | &free_remove_partial_attr.attr, | |
6753 | &alloc_from_partial_attr.attr, | |
6754 | &alloc_slab_attr.attr, | |
6755 | &alloc_refill_attr.attr, | |
e36a2652 | 6756 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
6757 | &free_slab_attr.attr, |
6758 | &cpuslab_flush_attr.attr, | |
6759 | &deactivate_full_attr.attr, | |
6760 | &deactivate_empty_attr.attr, | |
6761 | &deactivate_to_head_attr.attr, | |
6762 | &deactivate_to_tail_attr.attr, | |
6763 | &deactivate_remote_frees_attr.attr, | |
03e404af | 6764 | &deactivate_bypass_attr.attr, |
65c3376a | 6765 | &order_fallback_attr.attr, |
b789ef51 CL |
6766 | &cmpxchg_double_fail_attr.attr, |
6767 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
6768 | &cpu_partial_alloc_attr.attr, |
6769 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
6770 | &cpu_partial_node_attr.attr, |
6771 | &cpu_partial_drain_attr.attr, | |
81819f0f | 6772 | #endif |
4c13dd3b DM |
6773 | #ifdef CONFIG_FAILSLAB |
6774 | &failslab_attr.attr, | |
6775 | #endif | |
346907ce | 6776 | #ifdef CONFIG_HARDENED_USERCOPY |
8eb8284b | 6777 | &usersize_attr.attr, |
346907ce | 6778 | #endif |
b84e04f1 IK |
6779 | #ifdef CONFIG_KFENCE |
6780 | &skip_kfence_attr.attr, | |
6781 | #endif | |
4c13dd3b | 6782 | |
81819f0f CL |
6783 | NULL |
6784 | }; | |
6785 | ||
1fdaaa23 | 6786 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
6787 | .attrs = slab_attrs, |
6788 | }; | |
6789 | ||
6790 | static ssize_t slab_attr_show(struct kobject *kobj, | |
6791 | struct attribute *attr, | |
6792 | char *buf) | |
6793 | { | |
6794 | struct slab_attribute *attribute; | |
6795 | struct kmem_cache *s; | |
81819f0f CL |
6796 | |
6797 | attribute = to_slab_attr(attr); | |
6798 | s = to_slab(kobj); | |
6799 | ||
6800 | if (!attribute->show) | |
6801 | return -EIO; | |
6802 | ||
2bfbb027 | 6803 | return attribute->show(s, buf); |
81819f0f CL |
6804 | } |
6805 | ||
6806 | static ssize_t slab_attr_store(struct kobject *kobj, | |
6807 | struct attribute *attr, | |
6808 | const char *buf, size_t len) | |
6809 | { | |
6810 | struct slab_attribute *attribute; | |
6811 | struct kmem_cache *s; | |
81819f0f CL |
6812 | |
6813 | attribute = to_slab_attr(attr); | |
6814 | s = to_slab(kobj); | |
6815 | ||
6816 | if (!attribute->store) | |
6817 | return -EIO; | |
6818 | ||
2bfbb027 | 6819 | return attribute->store(s, buf, len); |
81819f0f CL |
6820 | } |
6821 | ||
41a21285 CL |
6822 | static void kmem_cache_release(struct kobject *k) |
6823 | { | |
6824 | slab_kmem_cache_release(to_slab(k)); | |
6825 | } | |
6826 | ||
52cf25d0 | 6827 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
6828 | .show = slab_attr_show, |
6829 | .store = slab_attr_store, | |
6830 | }; | |
6831 | ||
9ebe720e | 6832 | static const struct kobj_type slab_ktype = { |
81819f0f | 6833 | .sysfs_ops = &slab_sysfs_ops, |
41a21285 | 6834 | .release = kmem_cache_release, |
81819f0f CL |
6835 | }; |
6836 | ||
27c3a314 | 6837 | static struct kset *slab_kset; |
81819f0f | 6838 | |
9a41707b VD |
6839 | static inline struct kset *cache_kset(struct kmem_cache *s) |
6840 | { | |
9a41707b VD |
6841 | return slab_kset; |
6842 | } | |
6843 | ||
d65360f2 | 6844 | #define ID_STR_LENGTH 32 |
81819f0f CL |
6845 | |
6846 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
6847 | * |
6848 | * Format :[flags-]size | |
81819f0f CL |
6849 | */ |
6850 | static char *create_unique_id(struct kmem_cache *s) | |
6851 | { | |
6852 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
6853 | char *p = name; | |
6854 | ||
7e9c323c CY |
6855 | if (!name) |
6856 | return ERR_PTR(-ENOMEM); | |
81819f0f CL |
6857 | |
6858 | *p++ = ':'; | |
6859 | /* | |
6860 | * First flags affecting slabcache operations. We will only | |
6861 | * get here for aliasable slabs so we do not need to support | |
6862 | * too many flags. The flags here must cover all flags that | |
6863 | * are matched during merging to guarantee that the id is | |
6864 | * unique. | |
6865 | */ | |
6866 | if (s->flags & SLAB_CACHE_DMA) | |
6867 | *p++ = 'd'; | |
6d6ea1e9 NB |
6868 | if (s->flags & SLAB_CACHE_DMA32) |
6869 | *p++ = 'D'; | |
81819f0f CL |
6870 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
6871 | *p++ = 'a'; | |
becfda68 | 6872 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 6873 | *p++ = 'F'; |
230e9fc2 VD |
6874 | if (s->flags & SLAB_ACCOUNT) |
6875 | *p++ = 'A'; | |
81819f0f CL |
6876 | if (p != name + 1) |
6877 | *p++ = '-'; | |
d65360f2 | 6878 | p += snprintf(p, ID_STR_LENGTH - (p - name), "%07u", s->size); |
2633d7a0 | 6879 | |
d65360f2 CY |
6880 | if (WARN_ON(p > name + ID_STR_LENGTH - 1)) { |
6881 | kfree(name); | |
6882 | return ERR_PTR(-EINVAL); | |
6883 | } | |
68ef169a | 6884 | kmsan_unpoison_memory(name, p - name); |
81819f0f CL |
6885 | return name; |
6886 | } | |
6887 | ||
6888 | static int sysfs_slab_add(struct kmem_cache *s) | |
6889 | { | |
6890 | int err; | |
6891 | const char *name; | |
1663f26d | 6892 | struct kset *kset = cache_kset(s); |
45530c44 | 6893 | int unmergeable = slab_unmergeable(s); |
81819f0f | 6894 | |
11066386 MC |
6895 | if (!unmergeable && disable_higher_order_debug && |
6896 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
6897 | unmergeable = 1; | |
6898 | ||
81819f0f CL |
6899 | if (unmergeable) { |
6900 | /* | |
6901 | * Slabcache can never be merged so we can use the name proper. | |
6902 | * This is typically the case for debug situations. In that | |
6903 | * case we can catch duplicate names easily. | |
6904 | */ | |
27c3a314 | 6905 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
6906 | name = s->name; |
6907 | } else { | |
6908 | /* | |
6909 | * Create a unique name for the slab as a target | |
6910 | * for the symlinks. | |
6911 | */ | |
6912 | name = create_unique_id(s); | |
7e9c323c CY |
6913 | if (IS_ERR(name)) |
6914 | return PTR_ERR(name); | |
81819f0f CL |
6915 | } |
6916 | ||
1663f26d | 6917 | s->kobj.kset = kset; |
26e4f205 | 6918 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
757fed1d | 6919 | if (err) |
80da026a | 6920 | goto out; |
81819f0f CL |
6921 | |
6922 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
6923 | if (err) |
6924 | goto out_del_kobj; | |
9a41707b | 6925 | |
81819f0f CL |
6926 | if (!unmergeable) { |
6927 | /* Setup first alias */ | |
6928 | sysfs_slab_alias(s, s->name); | |
81819f0f | 6929 | } |
54b6a731 DJ |
6930 | out: |
6931 | if (!unmergeable) | |
6932 | kfree(name); | |
6933 | return err; | |
6934 | out_del_kobj: | |
6935 | kobject_del(&s->kobj); | |
54b6a731 | 6936 | goto out; |
81819f0f CL |
6937 | } |
6938 | ||
d50d82fa MP |
6939 | void sysfs_slab_unlink(struct kmem_cache *s) |
6940 | { | |
011568eb | 6941 | kobject_del(&s->kobj); |
d50d82fa MP |
6942 | } |
6943 | ||
bf5eb3de TH |
6944 | void sysfs_slab_release(struct kmem_cache *s) |
6945 | { | |
011568eb | 6946 | kobject_put(&s->kobj); |
81819f0f CL |
6947 | } |
6948 | ||
6949 | /* | |
6950 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 6951 | * available lest we lose that information. |
81819f0f CL |
6952 | */ |
6953 | struct saved_alias { | |
6954 | struct kmem_cache *s; | |
6955 | const char *name; | |
6956 | struct saved_alias *next; | |
6957 | }; | |
6958 | ||
5af328a5 | 6959 | static struct saved_alias *alias_list; |
81819f0f CL |
6960 | |
6961 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
6962 | { | |
6963 | struct saved_alias *al; | |
6964 | ||
97d06609 | 6965 | if (slab_state == FULL) { |
81819f0f CL |
6966 | /* |
6967 | * If we have a leftover link then remove it. | |
6968 | */ | |
27c3a314 GKH |
6969 | sysfs_remove_link(&slab_kset->kobj, name); |
6970 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
6971 | } |
6972 | ||
6973 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
6974 | if (!al) | |
6975 | return -ENOMEM; | |
6976 | ||
6977 | al->s = s; | |
6978 | al->name = name; | |
6979 | al->next = alias_list; | |
6980 | alias_list = al; | |
68ef169a | 6981 | kmsan_unpoison_memory(al, sizeof(*al)); |
81819f0f CL |
6982 | return 0; |
6983 | } | |
6984 | ||
6985 | static int __init slab_sysfs_init(void) | |
6986 | { | |
5b95a4ac | 6987 | struct kmem_cache *s; |
81819f0f CL |
6988 | int err; |
6989 | ||
18004c5d | 6990 | mutex_lock(&slab_mutex); |
2bce6485 | 6991 | |
d7660ce5 | 6992 | slab_kset = kset_create_and_add("slab", NULL, kernel_kobj); |
27c3a314 | 6993 | if (!slab_kset) { |
18004c5d | 6994 | mutex_unlock(&slab_mutex); |
f9f58285 | 6995 | pr_err("Cannot register slab subsystem.\n"); |
35973232 | 6996 | return -ENOMEM; |
81819f0f CL |
6997 | } |
6998 | ||
97d06609 | 6999 | slab_state = FULL; |
26a7bd03 | 7000 | |
5b95a4ac | 7001 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 7002 | err = sysfs_slab_add(s); |
5d540fb7 | 7003 | if (err) |
f9f58285 FF |
7004 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
7005 | s->name); | |
26a7bd03 | 7006 | } |
81819f0f CL |
7007 | |
7008 | while (alias_list) { | |
7009 | struct saved_alias *al = alias_list; | |
7010 | ||
7011 | alias_list = alias_list->next; | |
7012 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 7013 | if (err) |
f9f58285 FF |
7014 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
7015 | al->name); | |
81819f0f CL |
7016 | kfree(al); |
7017 | } | |
7018 | ||
18004c5d | 7019 | mutex_unlock(&slab_mutex); |
81819f0f CL |
7020 | return 0; |
7021 | } | |
1a5ad30b | 7022 | late_initcall(slab_sysfs_init); |
b1a413a3 | 7023 | #endif /* SLAB_SUPPORTS_SYSFS */ |
57ed3eda | 7024 | |
64dd6849 FM |
7025 | #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS) |
7026 | static int slab_debugfs_show(struct seq_file *seq, void *v) | |
7027 | { | |
64dd6849 | 7028 | struct loc_track *t = seq->private; |
005a79e5 GS |
7029 | struct location *l; |
7030 | unsigned long idx; | |
64dd6849 | 7031 | |
005a79e5 | 7032 | idx = (unsigned long) t->idx; |
64dd6849 FM |
7033 | if (idx < t->count) { |
7034 | l = &t->loc[idx]; | |
7035 | ||
7036 | seq_printf(seq, "%7ld ", l->count); | |
7037 | ||
7038 | if (l->addr) | |
7039 | seq_printf(seq, "%pS", (void *)l->addr); | |
7040 | else | |
7041 | seq_puts(seq, "<not-available>"); | |
7042 | ||
6edf2576 FT |
7043 | if (l->waste) |
7044 | seq_printf(seq, " waste=%lu/%lu", | |
7045 | l->count * l->waste, l->waste); | |
7046 | ||
64dd6849 FM |
7047 | if (l->sum_time != l->min_time) { |
7048 | seq_printf(seq, " age=%ld/%llu/%ld", | |
7049 | l->min_time, div_u64(l->sum_time, l->count), | |
7050 | l->max_time); | |
7051 | } else | |
7052 | seq_printf(seq, " age=%ld", l->min_time); | |
7053 | ||
7054 | if (l->min_pid != l->max_pid) | |
7055 | seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid); | |
7056 | else | |
7057 | seq_printf(seq, " pid=%ld", | |
7058 | l->min_pid); | |
7059 | ||
7060 | if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus))) | |
7061 | seq_printf(seq, " cpus=%*pbl", | |
7062 | cpumask_pr_args(to_cpumask(l->cpus))); | |
7063 | ||
7064 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes)) | |
7065 | seq_printf(seq, " nodes=%*pbl", | |
7066 | nodemask_pr_args(&l->nodes)); | |
7067 | ||
8ea9fb92 OG |
7068 | #ifdef CONFIG_STACKDEPOT |
7069 | { | |
7070 | depot_stack_handle_t handle; | |
7071 | unsigned long *entries; | |
7072 | unsigned int nr_entries, j; | |
7073 | ||
7074 | handle = READ_ONCE(l->handle); | |
7075 | if (handle) { | |
7076 | nr_entries = stack_depot_fetch(handle, &entries); | |
7077 | seq_puts(seq, "\n"); | |
7078 | for (j = 0; j < nr_entries; j++) | |
7079 | seq_printf(seq, " %pS\n", (void *)entries[j]); | |
7080 | } | |
7081 | } | |
7082 | #endif | |
64dd6849 FM |
7083 | seq_puts(seq, "\n"); |
7084 | } | |
7085 | ||
7086 | if (!idx && !t->count) | |
7087 | seq_puts(seq, "No data\n"); | |
7088 | ||
7089 | return 0; | |
7090 | } | |
7091 | ||
7092 | static void slab_debugfs_stop(struct seq_file *seq, void *v) | |
7093 | { | |
7094 | } | |
7095 | ||
7096 | static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos) | |
7097 | { | |
7098 | struct loc_track *t = seq->private; | |
7099 | ||
005a79e5 | 7100 | t->idx = ++(*ppos); |
64dd6849 | 7101 | if (*ppos <= t->count) |
005a79e5 | 7102 | return ppos; |
64dd6849 FM |
7103 | |
7104 | return NULL; | |
7105 | } | |
7106 | ||
553c0369 OG |
7107 | static int cmp_loc_by_count(const void *a, const void *b, const void *data) |
7108 | { | |
7109 | struct location *loc1 = (struct location *)a; | |
7110 | struct location *loc2 = (struct location *)b; | |
7111 | ||
7112 | if (loc1->count > loc2->count) | |
7113 | return -1; | |
7114 | else | |
7115 | return 1; | |
7116 | } | |
7117 | ||
64dd6849 FM |
7118 | static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos) |
7119 | { | |
005a79e5 GS |
7120 | struct loc_track *t = seq->private; |
7121 | ||
7122 | t->idx = *ppos; | |
64dd6849 FM |
7123 | return ppos; |
7124 | } | |
7125 | ||
7126 | static const struct seq_operations slab_debugfs_sops = { | |
7127 | .start = slab_debugfs_start, | |
7128 | .next = slab_debugfs_next, | |
7129 | .stop = slab_debugfs_stop, | |
7130 | .show = slab_debugfs_show, | |
7131 | }; | |
7132 | ||
7133 | static int slab_debug_trace_open(struct inode *inode, struct file *filep) | |
7134 | { | |
7135 | ||
7136 | struct kmem_cache_node *n; | |
7137 | enum track_item alloc; | |
7138 | int node; | |
7139 | struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops, | |
7140 | sizeof(struct loc_track)); | |
7141 | struct kmem_cache *s = file_inode(filep)->i_private; | |
b3fd64e1 VB |
7142 | unsigned long *obj_map; |
7143 | ||
2127d225 ML |
7144 | if (!t) |
7145 | return -ENOMEM; | |
7146 | ||
b3fd64e1 | 7147 | obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL); |
2127d225 ML |
7148 | if (!obj_map) { |
7149 | seq_release_private(inode, filep); | |
b3fd64e1 | 7150 | return -ENOMEM; |
2127d225 | 7151 | } |
64dd6849 FM |
7152 | |
7153 | if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0) | |
7154 | alloc = TRACK_ALLOC; | |
7155 | else | |
7156 | alloc = TRACK_FREE; | |
7157 | ||
b3fd64e1 VB |
7158 | if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) { |
7159 | bitmap_free(obj_map); | |
2127d225 | 7160 | seq_release_private(inode, filep); |
64dd6849 | 7161 | return -ENOMEM; |
b3fd64e1 | 7162 | } |
64dd6849 | 7163 | |
64dd6849 FM |
7164 | for_each_kmem_cache_node(s, node, n) { |
7165 | unsigned long flags; | |
bb192ed9 | 7166 | struct slab *slab; |
64dd6849 | 7167 | |
8040cbf5 | 7168 | if (!node_nr_slabs(n)) |
64dd6849 FM |
7169 | continue; |
7170 | ||
7171 | spin_lock_irqsave(&n->list_lock, flags); | |
bb192ed9 VB |
7172 | list_for_each_entry(slab, &n->partial, slab_list) |
7173 | process_slab(t, s, slab, alloc, obj_map); | |
7174 | list_for_each_entry(slab, &n->full, slab_list) | |
7175 | process_slab(t, s, slab, alloc, obj_map); | |
64dd6849 FM |
7176 | spin_unlock_irqrestore(&n->list_lock, flags); |
7177 | } | |
7178 | ||
553c0369 OG |
7179 | /* Sort locations by count */ |
7180 | sort_r(t->loc, t->count, sizeof(struct location), | |
7181 | cmp_loc_by_count, NULL, NULL); | |
7182 | ||
b3fd64e1 | 7183 | bitmap_free(obj_map); |
64dd6849 FM |
7184 | return 0; |
7185 | } | |
7186 | ||
7187 | static int slab_debug_trace_release(struct inode *inode, struct file *file) | |
7188 | { | |
7189 | struct seq_file *seq = file->private_data; | |
7190 | struct loc_track *t = seq->private; | |
7191 | ||
7192 | free_loc_track(t); | |
7193 | return seq_release_private(inode, file); | |
7194 | } | |
7195 | ||
7196 | static const struct file_operations slab_debugfs_fops = { | |
7197 | .open = slab_debug_trace_open, | |
7198 | .read = seq_read, | |
7199 | .llseek = seq_lseek, | |
7200 | .release = slab_debug_trace_release, | |
7201 | }; | |
7202 | ||
7203 | static void debugfs_slab_add(struct kmem_cache *s) | |
7204 | { | |
7205 | struct dentry *slab_cache_dir; | |
7206 | ||
7207 | if (unlikely(!slab_debugfs_root)) | |
7208 | return; | |
7209 | ||
7210 | slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root); | |
7211 | ||
7212 | debugfs_create_file("alloc_traces", 0400, | |
7213 | slab_cache_dir, s, &slab_debugfs_fops); | |
7214 | ||
7215 | debugfs_create_file("free_traces", 0400, | |
7216 | slab_cache_dir, s, &slab_debugfs_fops); | |
7217 | } | |
7218 | ||
7219 | void debugfs_slab_release(struct kmem_cache *s) | |
7220 | { | |
aa4a8605 | 7221 | debugfs_lookup_and_remove(s->name, slab_debugfs_root); |
64dd6849 FM |
7222 | } |
7223 | ||
7224 | static int __init slab_debugfs_init(void) | |
7225 | { | |
7226 | struct kmem_cache *s; | |
7227 | ||
7228 | slab_debugfs_root = debugfs_create_dir("slab", NULL); | |
7229 | ||
7230 | list_for_each_entry(s, &slab_caches, list) | |
7231 | if (s->flags & SLAB_STORE_USER) | |
7232 | debugfs_slab_add(s); | |
7233 | ||
7234 | return 0; | |
7235 | ||
7236 | } | |
7237 | __initcall(slab_debugfs_init); | |
7238 | #endif | |
57ed3eda PE |
7239 | /* |
7240 | * The /proc/slabinfo ABI | |
7241 | */ | |
5b365771 | 7242 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 7243 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 7244 | { |
57ed3eda | 7245 | unsigned long nr_slabs = 0; |
205ab99d CL |
7246 | unsigned long nr_objs = 0; |
7247 | unsigned long nr_free = 0; | |
57ed3eda | 7248 | int node; |
fa45dc25 | 7249 | struct kmem_cache_node *n; |
57ed3eda | 7250 | |
fa45dc25 | 7251 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
7252 | nr_slabs += node_nr_slabs(n); |
7253 | nr_objs += node_nr_objs(n); | |
046f4c69 | 7254 | nr_free += count_partial_free_approx(n); |
57ed3eda PE |
7255 | } |
7256 | ||
0d7561c6 GC |
7257 | sinfo->active_objs = nr_objs - nr_free; |
7258 | sinfo->num_objs = nr_objs; | |
7259 | sinfo->active_slabs = nr_slabs; | |
7260 | sinfo->num_slabs = nr_slabs; | |
7261 | sinfo->objects_per_slab = oo_objects(s->oo); | |
7262 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda | 7263 | } |
5b365771 | 7264 | #endif /* CONFIG_SLUB_DEBUG */ |