]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
5 | * The allocator synchronizes using per slab locks and only | |
6 | * uses a centralized lock to manage a pool of partial slabs. | |
7 | * | |
8 | * (C) 2007 SGI, Christoph Lameter <clameter@sgi.com> | |
9 | */ | |
10 | ||
11 | #include <linux/mm.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/bit_spinlock.h> | |
14 | #include <linux/interrupt.h> | |
15 | #include <linux/bitops.h> | |
16 | #include <linux/slab.h> | |
17 | #include <linux/seq_file.h> | |
18 | #include <linux/cpu.h> | |
19 | #include <linux/cpuset.h> | |
20 | #include <linux/mempolicy.h> | |
21 | #include <linux/ctype.h> | |
22 | #include <linux/kallsyms.h> | |
23 | ||
24 | /* | |
25 | * Lock order: | |
26 | * 1. slab_lock(page) | |
27 | * 2. slab->list_lock | |
28 | * | |
29 | * The slab_lock protects operations on the object of a particular | |
30 | * slab and its metadata in the page struct. If the slab lock | |
31 | * has been taken then no allocations nor frees can be performed | |
32 | * on the objects in the slab nor can the slab be added or removed | |
33 | * from the partial or full lists since this would mean modifying | |
34 | * the page_struct of the slab. | |
35 | * | |
36 | * The list_lock protects the partial and full list on each node and | |
37 | * the partial slab counter. If taken then no new slabs may be added or | |
38 | * removed from the lists nor make the number of partial slabs be modified. | |
39 | * (Note that the total number of slabs is an atomic value that may be | |
40 | * modified without taking the list lock). | |
41 | * | |
42 | * The list_lock is a centralized lock and thus we avoid taking it as | |
43 | * much as possible. As long as SLUB does not have to handle partial | |
44 | * slabs, operations can continue without any centralized lock. F.e. | |
45 | * allocating a long series of objects that fill up slabs does not require | |
46 | * the list lock. | |
47 | * | |
48 | * The lock order is sometimes inverted when we are trying to get a slab | |
49 | * off a list. We take the list_lock and then look for a page on the list | |
50 | * to use. While we do that objects in the slabs may be freed. We can | |
51 | * only operate on the slab if we have also taken the slab_lock. So we use | |
52 | * a slab_trylock() on the slab. If trylock was successful then no frees | |
53 | * can occur anymore and we can use the slab for allocations etc. If the | |
54 | * slab_trylock() does not succeed then frees are in progress in the slab and | |
55 | * we must stay away from it for a while since we may cause a bouncing | |
56 | * cacheline if we try to acquire the lock. So go onto the next slab. | |
57 | * If all pages are busy then we may allocate a new slab instead of reusing | |
58 | * a partial slab. A new slab has noone operating on it and thus there is | |
59 | * no danger of cacheline contention. | |
60 | * | |
61 | * Interrupts are disabled during allocation and deallocation in order to | |
62 | * make the slab allocator safe to use in the context of an irq. In addition | |
63 | * interrupts are disabled to ensure that the processor does not change | |
64 | * while handling per_cpu slabs, due to kernel preemption. | |
65 | * | |
66 | * SLUB assigns one slab for allocation to each processor. | |
67 | * Allocations only occur from these slabs called cpu slabs. | |
68 | * | |
69 | * Slabs with free elements are kept on a partial list. | |
70 | * There is no list for full slabs. If an object in a full slab is | |
71 | * freed then the slab will show up again on the partial lists. | |
72 | * Otherwise there is no need to track full slabs unless we have to | |
73 | * track full slabs for debugging purposes. | |
74 | * | |
75 | * Slabs are freed when they become empty. Teardown and setup is | |
76 | * minimal so we rely on the page allocators per cpu caches for | |
77 | * fast frees and allocs. | |
78 | * | |
79 | * Overloading of page flags that are otherwise used for LRU management. | |
80 | * | |
81 | * PageActive The slab is used as a cpu cache. Allocations | |
82 | * may be performed from the slab. The slab is not | |
83 | * on any slab list and cannot be moved onto one. | |
84 | * | |
85 | * PageError Slab requires special handling due to debug | |
86 | * options set. This moves slab handling out of | |
87 | * the fast path. | |
88 | */ | |
89 | ||
90 | /* | |
91 | * Issues still to be resolved: | |
92 | * | |
93 | * - The per cpu array is updated for each new slab and and is a remote | |
94 | * cacheline for most nodes. This could become a bouncing cacheline given | |
95 | * enough frequent updates. There are 16 pointers in a cacheline.so at | |
96 | * max 16 cpus could compete. Likely okay. | |
97 | * | |
98 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. | |
99 | * | |
100 | * - Support DEBUG_SLAB_LEAK. Trouble is we do not know where the full | |
101 | * slabs are in SLUB. | |
102 | * | |
103 | * - SLAB_DEBUG_INITIAL is not supported but I have never seen a use of | |
104 | * it. | |
105 | * | |
106 | * - Variable sizing of the per node arrays | |
107 | */ | |
108 | ||
109 | /* Enable to test recovery from slab corruption on boot */ | |
110 | #undef SLUB_RESILIENCY_TEST | |
111 | ||
112 | #if PAGE_SHIFT <= 12 | |
113 | ||
114 | /* | |
115 | * Small page size. Make sure that we do not fragment memory | |
116 | */ | |
117 | #define DEFAULT_MAX_ORDER 1 | |
118 | #define DEFAULT_MIN_OBJECTS 4 | |
119 | ||
120 | #else | |
121 | ||
122 | /* | |
123 | * Large page machines are customarily able to handle larger | |
124 | * page orders. | |
125 | */ | |
126 | #define DEFAULT_MAX_ORDER 2 | |
127 | #define DEFAULT_MIN_OBJECTS 8 | |
128 | ||
129 | #endif | |
130 | ||
131 | /* | |
132 | * Flags from the regular SLAB that SLUB does not support: | |
133 | */ | |
134 | #define SLUB_UNIMPLEMENTED (SLAB_DEBUG_INITIAL) | |
135 | ||
136 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ | |
137 | SLAB_POISON | SLAB_STORE_USER) | |
138 | /* | |
139 | * Set of flags that will prevent slab merging | |
140 | */ | |
141 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
142 | SLAB_TRACE | SLAB_DESTROY_BY_RCU) | |
143 | ||
144 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
145 | SLAB_CACHE_DMA) | |
146 | ||
147 | #ifndef ARCH_KMALLOC_MINALIGN | |
47bfdc0d | 148 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
149 | #endif |
150 | ||
151 | #ifndef ARCH_SLAB_MINALIGN | |
47bfdc0d | 152 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) |
81819f0f CL |
153 | #endif |
154 | ||
155 | /* Internal SLUB flags */ | |
156 | #define __OBJECT_POISON 0x80000000 /* Poison object */ | |
157 | ||
158 | static int kmem_size = sizeof(struct kmem_cache); | |
159 | ||
160 | #ifdef CONFIG_SMP | |
161 | static struct notifier_block slab_notifier; | |
162 | #endif | |
163 | ||
164 | static enum { | |
165 | DOWN, /* No slab functionality available */ | |
166 | PARTIAL, /* kmem_cache_open() works but kmalloc does not */ | |
167 | UP, /* Everything works */ | |
168 | SYSFS /* Sysfs up */ | |
169 | } slab_state = DOWN; | |
170 | ||
171 | /* A list of all slab caches on the system */ | |
172 | static DECLARE_RWSEM(slub_lock); | |
173 | LIST_HEAD(slab_caches); | |
174 | ||
175 | #ifdef CONFIG_SYSFS | |
176 | static int sysfs_slab_add(struct kmem_cache *); | |
177 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
178 | static void sysfs_slab_remove(struct kmem_cache *); | |
179 | #else | |
180 | static int sysfs_slab_add(struct kmem_cache *s) { return 0; } | |
181 | static int sysfs_slab_alias(struct kmem_cache *s, const char *p) { return 0; } | |
182 | static void sysfs_slab_remove(struct kmem_cache *s) {} | |
183 | #endif | |
184 | ||
185 | /******************************************************************** | |
186 | * Core slab cache functions | |
187 | *******************************************************************/ | |
188 | ||
189 | int slab_is_available(void) | |
190 | { | |
191 | return slab_state >= UP; | |
192 | } | |
193 | ||
194 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
195 | { | |
196 | #ifdef CONFIG_NUMA | |
197 | return s->node[node]; | |
198 | #else | |
199 | return &s->local_node; | |
200 | #endif | |
201 | } | |
202 | ||
203 | /* | |
204 | * Object debugging | |
205 | */ | |
206 | static void print_section(char *text, u8 *addr, unsigned int length) | |
207 | { | |
208 | int i, offset; | |
209 | int newline = 1; | |
210 | char ascii[17]; | |
211 | ||
212 | ascii[16] = 0; | |
213 | ||
214 | for (i = 0; i < length; i++) { | |
215 | if (newline) { | |
216 | printk(KERN_ERR "%10s 0x%p: ", text, addr + i); | |
217 | newline = 0; | |
218 | } | |
219 | printk(" %02x", addr[i]); | |
220 | offset = i % 16; | |
221 | ascii[offset] = isgraph(addr[i]) ? addr[i] : '.'; | |
222 | if (offset == 15) { | |
223 | printk(" %s\n",ascii); | |
224 | newline = 1; | |
225 | } | |
226 | } | |
227 | if (!newline) { | |
228 | i %= 16; | |
229 | while (i < 16) { | |
230 | printk(" "); | |
231 | ascii[i] = ' '; | |
232 | i++; | |
233 | } | |
234 | printk(" %s\n", ascii); | |
235 | } | |
236 | } | |
237 | ||
238 | /* | |
239 | * Slow version of get and set free pointer. | |
240 | * | |
241 | * This requires touching the cache lines of kmem_cache. | |
242 | * The offset can also be obtained from the page. In that | |
243 | * case it is in the cacheline that we already need to touch. | |
244 | */ | |
245 | static void *get_freepointer(struct kmem_cache *s, void *object) | |
246 | { | |
247 | return *(void **)(object + s->offset); | |
248 | } | |
249 | ||
250 | static void set_freepointer(struct kmem_cache *s, void *object, void *fp) | |
251 | { | |
252 | *(void **)(object + s->offset) = fp; | |
253 | } | |
254 | ||
255 | /* | |
256 | * Tracking user of a slab. | |
257 | */ | |
258 | struct track { | |
259 | void *addr; /* Called from address */ | |
260 | int cpu; /* Was running on cpu */ | |
261 | int pid; /* Pid context */ | |
262 | unsigned long when; /* When did the operation occur */ | |
263 | }; | |
264 | ||
265 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
266 | ||
267 | static struct track *get_track(struct kmem_cache *s, void *object, | |
268 | enum track_item alloc) | |
269 | { | |
270 | struct track *p; | |
271 | ||
272 | if (s->offset) | |
273 | p = object + s->offset + sizeof(void *); | |
274 | else | |
275 | p = object + s->inuse; | |
276 | ||
277 | return p + alloc; | |
278 | } | |
279 | ||
280 | static void set_track(struct kmem_cache *s, void *object, | |
281 | enum track_item alloc, void *addr) | |
282 | { | |
283 | struct track *p; | |
284 | ||
285 | if (s->offset) | |
286 | p = object + s->offset + sizeof(void *); | |
287 | else | |
288 | p = object + s->inuse; | |
289 | ||
290 | p += alloc; | |
291 | if (addr) { | |
292 | p->addr = addr; | |
293 | p->cpu = smp_processor_id(); | |
294 | p->pid = current ? current->pid : -1; | |
295 | p->when = jiffies; | |
296 | } else | |
297 | memset(p, 0, sizeof(struct track)); | |
298 | } | |
299 | ||
81819f0f CL |
300 | static void init_tracking(struct kmem_cache *s, void *object) |
301 | { | |
302 | if (s->flags & SLAB_STORE_USER) { | |
303 | set_track(s, object, TRACK_FREE, NULL); | |
304 | set_track(s, object, TRACK_ALLOC, NULL); | |
305 | } | |
306 | } | |
307 | ||
308 | static void print_track(const char *s, struct track *t) | |
309 | { | |
310 | if (!t->addr) | |
311 | return; | |
312 | ||
313 | printk(KERN_ERR "%s: ", s); | |
314 | __print_symbol("%s", (unsigned long)t->addr); | |
315 | printk(" jiffies_ago=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid); | |
316 | } | |
317 | ||
318 | static void print_trailer(struct kmem_cache *s, u8 *p) | |
319 | { | |
320 | unsigned int off; /* Offset of last byte */ | |
321 | ||
322 | if (s->flags & SLAB_RED_ZONE) | |
323 | print_section("Redzone", p + s->objsize, | |
324 | s->inuse - s->objsize); | |
325 | ||
326 | printk(KERN_ERR "FreePointer 0x%p -> 0x%p\n", | |
327 | p + s->offset, | |
328 | get_freepointer(s, p)); | |
329 | ||
330 | if (s->offset) | |
331 | off = s->offset + sizeof(void *); | |
332 | else | |
333 | off = s->inuse; | |
334 | ||
335 | if (s->flags & SLAB_STORE_USER) { | |
336 | print_track("Last alloc", get_track(s, p, TRACK_ALLOC)); | |
337 | print_track("Last free ", get_track(s, p, TRACK_FREE)); | |
338 | off += 2 * sizeof(struct track); | |
339 | } | |
340 | ||
341 | if (off != s->size) | |
342 | /* Beginning of the filler is the free pointer */ | |
343 | print_section("Filler", p + off, s->size - off); | |
344 | } | |
345 | ||
346 | static void object_err(struct kmem_cache *s, struct page *page, | |
347 | u8 *object, char *reason) | |
348 | { | |
349 | u8 *addr = page_address(page); | |
350 | ||
351 | printk(KERN_ERR "*** SLUB %s: %s@0x%p slab 0x%p\n", | |
352 | s->name, reason, object, page); | |
353 | printk(KERN_ERR " offset=%tu flags=0x%04lx inuse=%u freelist=0x%p\n", | |
354 | object - addr, page->flags, page->inuse, page->freelist); | |
355 | if (object > addr + 16) | |
356 | print_section("Bytes b4", object - 16, 16); | |
357 | print_section("Object", object, min(s->objsize, 128)); | |
358 | print_trailer(s, object); | |
359 | dump_stack(); | |
360 | } | |
361 | ||
362 | static void slab_err(struct kmem_cache *s, struct page *page, char *reason, ...) | |
363 | { | |
364 | va_list args; | |
365 | char buf[100]; | |
366 | ||
367 | va_start(args, reason); | |
368 | vsnprintf(buf, sizeof(buf), reason, args); | |
369 | va_end(args); | |
370 | printk(KERN_ERR "*** SLUB %s: %s in slab @0x%p\n", s->name, buf, | |
371 | page); | |
372 | dump_stack(); | |
373 | } | |
374 | ||
375 | static void init_object(struct kmem_cache *s, void *object, int active) | |
376 | { | |
377 | u8 *p = object; | |
378 | ||
379 | if (s->flags & __OBJECT_POISON) { | |
380 | memset(p, POISON_FREE, s->objsize - 1); | |
381 | p[s->objsize -1] = POISON_END; | |
382 | } | |
383 | ||
384 | if (s->flags & SLAB_RED_ZONE) | |
385 | memset(p + s->objsize, | |
386 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE, | |
387 | s->inuse - s->objsize); | |
388 | } | |
389 | ||
390 | static int check_bytes(u8 *start, unsigned int value, unsigned int bytes) | |
391 | { | |
392 | while (bytes) { | |
393 | if (*start != (u8)value) | |
394 | return 0; | |
395 | start++; | |
396 | bytes--; | |
397 | } | |
398 | return 1; | |
399 | } | |
400 | ||
401 | ||
402 | static int check_valid_pointer(struct kmem_cache *s, struct page *page, | |
403 | void *object) | |
404 | { | |
405 | void *base; | |
406 | ||
407 | if (!object) | |
408 | return 1; | |
409 | ||
410 | base = page_address(page); | |
411 | if (object < base || object >= base + s->objects * s->size || | |
412 | (object - base) % s->size) { | |
413 | return 0; | |
414 | } | |
415 | ||
416 | return 1; | |
417 | } | |
418 | ||
419 | /* | |
420 | * Object layout: | |
421 | * | |
422 | * object address | |
423 | * Bytes of the object to be managed. | |
424 | * If the freepointer may overlay the object then the free | |
425 | * pointer is the first word of the object. | |
426 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is | |
427 | * 0xa5 (POISON_END) | |
428 | * | |
429 | * object + s->objsize | |
430 | * Padding to reach word boundary. This is also used for Redzoning. | |
431 | * Padding is extended to word size if Redzoning is enabled | |
432 | * and objsize == inuse. | |
433 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with | |
434 | * 0xcc (RED_ACTIVE) for objects in use. | |
435 | * | |
436 | * object + s->inuse | |
437 | * A. Free pointer (if we cannot overwrite object on free) | |
438 | * B. Tracking data for SLAB_STORE_USER | |
439 | * C. Padding to reach required alignment boundary | |
440 | * Padding is done using 0x5a (POISON_INUSE) | |
441 | * | |
442 | * object + s->size | |
443 | * | |
444 | * If slabcaches are merged then the objsize and inuse boundaries are to | |
445 | * be ignored. And therefore no slab options that rely on these boundaries | |
446 | * may be used with merged slabcaches. | |
447 | */ | |
448 | ||
449 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, | |
450 | void *from, void *to) | |
451 | { | |
452 | printk(KERN_ERR "@@@ SLUB: %s Restoring %s (0x%x) from 0x%p-0x%p\n", | |
453 | s->name, message, data, from, to - 1); | |
454 | memset(from, data, to - from); | |
455 | } | |
456 | ||
457 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) | |
458 | { | |
459 | unsigned long off = s->inuse; /* The end of info */ | |
460 | ||
461 | if (s->offset) | |
462 | /* Freepointer is placed after the object. */ | |
463 | off += sizeof(void *); | |
464 | ||
465 | if (s->flags & SLAB_STORE_USER) | |
466 | /* We also have user information there */ | |
467 | off += 2 * sizeof(struct track); | |
468 | ||
469 | if (s->size == off) | |
470 | return 1; | |
471 | ||
472 | if (check_bytes(p + off, POISON_INUSE, s->size - off)) | |
473 | return 1; | |
474 | ||
475 | object_err(s, page, p, "Object padding check fails"); | |
476 | ||
477 | /* | |
478 | * Restore padding | |
479 | */ | |
480 | restore_bytes(s, "object padding", POISON_INUSE, p + off, p + s->size); | |
481 | return 0; | |
482 | } | |
483 | ||
484 | static int slab_pad_check(struct kmem_cache *s, struct page *page) | |
485 | { | |
486 | u8 *p; | |
487 | int length, remainder; | |
488 | ||
489 | if (!(s->flags & SLAB_POISON)) | |
490 | return 1; | |
491 | ||
492 | p = page_address(page); | |
493 | length = s->objects * s->size; | |
494 | remainder = (PAGE_SIZE << s->order) - length; | |
495 | if (!remainder) | |
496 | return 1; | |
497 | ||
498 | if (!check_bytes(p + length, POISON_INUSE, remainder)) { | |
499 | printk(KERN_ERR "SLUB: %s slab 0x%p: Padding fails check\n", | |
500 | s->name, p); | |
501 | dump_stack(); | |
502 | restore_bytes(s, "slab padding", POISON_INUSE, p + length, | |
503 | p + length + remainder); | |
504 | return 0; | |
505 | } | |
506 | return 1; | |
507 | } | |
508 | ||
509 | static int check_object(struct kmem_cache *s, struct page *page, | |
510 | void *object, int active) | |
511 | { | |
512 | u8 *p = object; | |
513 | u8 *endobject = object + s->objsize; | |
514 | ||
515 | if (s->flags & SLAB_RED_ZONE) { | |
516 | unsigned int red = | |
517 | active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE; | |
518 | ||
519 | if (!check_bytes(endobject, red, s->inuse - s->objsize)) { | |
520 | object_err(s, page, object, | |
521 | active ? "Redzone Active" : "Redzone Inactive"); | |
522 | restore_bytes(s, "redzone", red, | |
523 | endobject, object + s->inuse); | |
524 | return 0; | |
525 | } | |
526 | } else { | |
527 | if ((s->flags & SLAB_POISON) && s->objsize < s->inuse && | |
528 | !check_bytes(endobject, POISON_INUSE, | |
529 | s->inuse - s->objsize)) { | |
530 | object_err(s, page, p, "Alignment padding check fails"); | |
531 | /* | |
532 | * Fix it so that there will not be another report. | |
533 | * | |
534 | * Hmmm... We may be corrupting an object that now expects | |
535 | * to be longer than allowed. | |
536 | */ | |
537 | restore_bytes(s, "alignment padding", POISON_INUSE, | |
538 | endobject, object + s->inuse); | |
539 | } | |
540 | } | |
541 | ||
542 | if (s->flags & SLAB_POISON) { | |
543 | if (!active && (s->flags & __OBJECT_POISON) && | |
544 | (!check_bytes(p, POISON_FREE, s->objsize - 1) || | |
545 | p[s->objsize - 1] != POISON_END)) { | |
546 | ||
547 | object_err(s, page, p, "Poison check failed"); | |
548 | restore_bytes(s, "Poison", POISON_FREE, | |
549 | p, p + s->objsize -1); | |
550 | restore_bytes(s, "Poison", POISON_END, | |
551 | p + s->objsize - 1, p + s->objsize); | |
552 | return 0; | |
553 | } | |
554 | /* | |
555 | * check_pad_bytes cleans up on its own. | |
556 | */ | |
557 | check_pad_bytes(s, page, p); | |
558 | } | |
559 | ||
560 | if (!s->offset && active) | |
561 | /* | |
562 | * Object and freepointer overlap. Cannot check | |
563 | * freepointer while object is allocated. | |
564 | */ | |
565 | return 1; | |
566 | ||
567 | /* Check free pointer validity */ | |
568 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
569 | object_err(s, page, p, "Freepointer corrupt"); | |
570 | /* | |
571 | * No choice but to zap it and thus loose the remainder | |
572 | * of the free objects in this slab. May cause | |
573 | * another error because the object count maybe | |
574 | * wrong now. | |
575 | */ | |
576 | set_freepointer(s, p, NULL); | |
577 | return 0; | |
578 | } | |
579 | return 1; | |
580 | } | |
581 | ||
582 | static int check_slab(struct kmem_cache *s, struct page *page) | |
583 | { | |
584 | VM_BUG_ON(!irqs_disabled()); | |
585 | ||
586 | if (!PageSlab(page)) { | |
587 | printk(KERN_ERR "SLUB: %s Not a valid slab page @0x%p " | |
588 | "flags=%lx mapping=0x%p count=%d \n", | |
589 | s->name, page, page->flags, page->mapping, | |
590 | page_count(page)); | |
591 | return 0; | |
592 | } | |
593 | if (page->offset * sizeof(void *) != s->offset) { | |
594 | printk(KERN_ERR "SLUB: %s Corrupted offset %lu in slab @0x%p" | |
595 | " flags=0x%lx mapping=0x%p count=%d\n", | |
596 | s->name, | |
597 | (unsigned long)(page->offset * sizeof(void *)), | |
598 | page, | |
599 | page->flags, | |
600 | page->mapping, | |
601 | page_count(page)); | |
602 | dump_stack(); | |
603 | return 0; | |
604 | } | |
605 | if (page->inuse > s->objects) { | |
606 | printk(KERN_ERR "SLUB: %s Inuse %u > max %u in slab " | |
607 | "page @0x%p flags=%lx mapping=0x%p count=%d\n", | |
608 | s->name, page->inuse, s->objects, page, page->flags, | |
609 | page->mapping, page_count(page)); | |
610 | dump_stack(); | |
611 | return 0; | |
612 | } | |
613 | /* Slab_pad_check fixes things up after itself */ | |
614 | slab_pad_check(s, page); | |
615 | return 1; | |
616 | } | |
617 | ||
618 | /* | |
619 | * Determine if a certain object on a page is on the freelist and | |
620 | * therefore free. Must hold the slab lock for cpu slabs to | |
621 | * guarantee that the chains are consistent. | |
622 | */ | |
623 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
624 | { | |
625 | int nr = 0; | |
626 | void *fp = page->freelist; | |
627 | void *object = NULL; | |
628 | ||
629 | while (fp && nr <= s->objects) { | |
630 | if (fp == search) | |
631 | return 1; | |
632 | if (!check_valid_pointer(s, page, fp)) { | |
633 | if (object) { | |
634 | object_err(s, page, object, | |
635 | "Freechain corrupt"); | |
636 | set_freepointer(s, object, NULL); | |
637 | break; | |
638 | } else { | |
639 | printk(KERN_ERR "SLUB: %s slab 0x%p " | |
640 | "freepointer 0x%p corrupted.\n", | |
641 | s->name, page, fp); | |
642 | dump_stack(); | |
643 | page->freelist = NULL; | |
644 | page->inuse = s->objects; | |
645 | return 0; | |
646 | } | |
647 | break; | |
648 | } | |
649 | object = fp; | |
650 | fp = get_freepointer(s, object); | |
651 | nr++; | |
652 | } | |
653 | ||
654 | if (page->inuse != s->objects - nr) { | |
655 | printk(KERN_ERR "slab %s: page 0x%p wrong object count." | |
656 | " counter is %d but counted were %d\n", | |
657 | s->name, page, page->inuse, | |
658 | s->objects - nr); | |
659 | page->inuse = s->objects - nr; | |
660 | } | |
661 | return search == NULL; | |
662 | } | |
663 | ||
664 | static int alloc_object_checks(struct kmem_cache *s, struct page *page, | |
665 | void *object) | |
666 | { | |
667 | if (!check_slab(s, page)) | |
668 | goto bad; | |
669 | ||
670 | if (object && !on_freelist(s, page, object)) { | |
671 | printk(KERN_ERR "SLUB: %s Object 0x%p@0x%p " | |
672 | "already allocated.\n", | |
673 | s->name, object, page); | |
674 | goto dump; | |
675 | } | |
676 | ||
677 | if (!check_valid_pointer(s, page, object)) { | |
678 | object_err(s, page, object, "Freelist Pointer check fails"); | |
679 | goto dump; | |
680 | } | |
681 | ||
682 | if (!object) | |
683 | return 1; | |
684 | ||
685 | if (!check_object(s, page, object, 0)) | |
686 | goto bad; | |
687 | init_object(s, object, 1); | |
688 | ||
689 | if (s->flags & SLAB_TRACE) { | |
690 | printk(KERN_INFO "TRACE %s alloc 0x%p inuse=%d fp=0x%p\n", | |
691 | s->name, object, page->inuse, | |
692 | page->freelist); | |
693 | dump_stack(); | |
694 | } | |
695 | return 1; | |
696 | dump: | |
697 | dump_stack(); | |
698 | bad: | |
699 | if (PageSlab(page)) { | |
700 | /* | |
701 | * If this is a slab page then lets do the best we can | |
702 | * to avoid issues in the future. Marking all objects | |
703 | * as used avoids touching the remainder. | |
704 | */ | |
705 | printk(KERN_ERR "@@@ SLUB: %s slab 0x%p. Marking all objects used.\n", | |
706 | s->name, page); | |
707 | page->inuse = s->objects; | |
708 | page->freelist = NULL; | |
709 | /* Fix up fields that may be corrupted */ | |
710 | page->offset = s->offset / sizeof(void *); | |
711 | } | |
712 | return 0; | |
713 | } | |
714 | ||
715 | static int free_object_checks(struct kmem_cache *s, struct page *page, | |
716 | void *object) | |
717 | { | |
718 | if (!check_slab(s, page)) | |
719 | goto fail; | |
720 | ||
721 | if (!check_valid_pointer(s, page, object)) { | |
722 | printk(KERN_ERR "SLUB: %s slab 0x%p invalid " | |
723 | "object pointer 0x%p\n", | |
724 | s->name, page, object); | |
725 | goto fail; | |
726 | } | |
727 | ||
728 | if (on_freelist(s, page, object)) { | |
729 | printk(KERN_ERR "SLUB: %s slab 0x%p object " | |
730 | "0x%p already free.\n", s->name, page, object); | |
731 | goto fail; | |
732 | } | |
733 | ||
734 | if (!check_object(s, page, object, 1)) | |
735 | return 0; | |
736 | ||
737 | if (unlikely(s != page->slab)) { | |
738 | if (!PageSlab(page)) | |
739 | printk(KERN_ERR "slab_free %s size %d: attempt to" | |
740 | "free object(0x%p) outside of slab.\n", | |
741 | s->name, s->size, object); | |
742 | else | |
743 | if (!page->slab) | |
744 | printk(KERN_ERR | |
745 | "slab_free : no slab(NULL) for object 0x%p.\n", | |
746 | object); | |
747 | else | |
748 | printk(KERN_ERR "slab_free %s(%d): object at 0x%p" | |
749 | " belongs to slab %s(%d)\n", | |
750 | s->name, s->size, object, | |
751 | page->slab->name, page->slab->size); | |
752 | goto fail; | |
753 | } | |
754 | if (s->flags & SLAB_TRACE) { | |
755 | printk(KERN_INFO "TRACE %s free 0x%p inuse=%d fp=0x%p\n", | |
756 | s->name, object, page->inuse, | |
757 | page->freelist); | |
758 | print_section("Object", object, s->objsize); | |
759 | dump_stack(); | |
760 | } | |
761 | init_object(s, object, 0); | |
762 | return 1; | |
763 | fail: | |
764 | dump_stack(); | |
765 | printk(KERN_ERR "@@@ SLUB: %s slab 0x%p object at 0x%p not freed.\n", | |
766 | s->name, page, object); | |
767 | return 0; | |
768 | } | |
769 | ||
770 | /* | |
771 | * Slab allocation and freeing | |
772 | */ | |
773 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) | |
774 | { | |
775 | struct page * page; | |
776 | int pages = 1 << s->order; | |
777 | ||
778 | if (s->order) | |
779 | flags |= __GFP_COMP; | |
780 | ||
781 | if (s->flags & SLAB_CACHE_DMA) | |
782 | flags |= SLUB_DMA; | |
783 | ||
784 | if (node == -1) | |
785 | page = alloc_pages(flags, s->order); | |
786 | else | |
787 | page = alloc_pages_node(node, flags, s->order); | |
788 | ||
789 | if (!page) | |
790 | return NULL; | |
791 | ||
792 | mod_zone_page_state(page_zone(page), | |
793 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
794 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
795 | pages); | |
796 | ||
797 | return page; | |
798 | } | |
799 | ||
800 | static void setup_object(struct kmem_cache *s, struct page *page, | |
801 | void *object) | |
802 | { | |
803 | if (PageError(page)) { | |
804 | init_object(s, object, 0); | |
805 | init_tracking(s, object); | |
806 | } | |
807 | ||
808 | if (unlikely(s->ctor)) { | |
809 | int mode = SLAB_CTOR_CONSTRUCTOR; | |
810 | ||
811 | if (!(s->flags & __GFP_WAIT)) | |
812 | mode |= SLAB_CTOR_ATOMIC; | |
813 | ||
814 | s->ctor(object, s, mode); | |
815 | } | |
816 | } | |
817 | ||
818 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
819 | { | |
820 | struct page *page; | |
821 | struct kmem_cache_node *n; | |
822 | void *start; | |
823 | void *end; | |
824 | void *last; | |
825 | void *p; | |
826 | ||
827 | if (flags & __GFP_NO_GROW) | |
828 | return NULL; | |
829 | ||
830 | BUG_ON(flags & ~(GFP_DMA | GFP_LEVEL_MASK)); | |
831 | ||
832 | if (flags & __GFP_WAIT) | |
833 | local_irq_enable(); | |
834 | ||
835 | page = allocate_slab(s, flags & GFP_LEVEL_MASK, node); | |
836 | if (!page) | |
837 | goto out; | |
838 | ||
839 | n = get_node(s, page_to_nid(page)); | |
840 | if (n) | |
841 | atomic_long_inc(&n->nr_slabs); | |
842 | page->offset = s->offset / sizeof(void *); | |
843 | page->slab = s; | |
844 | page->flags |= 1 << PG_slab; | |
845 | if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON | | |
846 | SLAB_STORE_USER | SLAB_TRACE)) | |
847 | page->flags |= 1 << PG_error; | |
848 | ||
849 | start = page_address(page); | |
850 | end = start + s->objects * s->size; | |
851 | ||
852 | if (unlikely(s->flags & SLAB_POISON)) | |
853 | memset(start, POISON_INUSE, PAGE_SIZE << s->order); | |
854 | ||
855 | last = start; | |
856 | for (p = start + s->size; p < end; p += s->size) { | |
857 | setup_object(s, page, last); | |
858 | set_freepointer(s, last, p); | |
859 | last = p; | |
860 | } | |
861 | setup_object(s, page, last); | |
862 | set_freepointer(s, last, NULL); | |
863 | ||
864 | page->freelist = start; | |
865 | page->inuse = 0; | |
866 | out: | |
867 | if (flags & __GFP_WAIT) | |
868 | local_irq_disable(); | |
869 | return page; | |
870 | } | |
871 | ||
872 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
873 | { | |
874 | int pages = 1 << s->order; | |
875 | ||
876 | if (unlikely(PageError(page) || s->dtor)) { | |
877 | void *start = page_address(page); | |
878 | void *end = start + (pages << PAGE_SHIFT); | |
879 | void *p; | |
880 | ||
881 | slab_pad_check(s, page); | |
882 | for (p = start; p <= end - s->size; p += s->size) { | |
883 | if (s->dtor) | |
884 | s->dtor(p, s, 0); | |
885 | check_object(s, page, p, 0); | |
886 | } | |
887 | } | |
888 | ||
889 | mod_zone_page_state(page_zone(page), | |
890 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
891 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
892 | - pages); | |
893 | ||
894 | page->mapping = NULL; | |
895 | __free_pages(page, s->order); | |
896 | } | |
897 | ||
898 | static void rcu_free_slab(struct rcu_head *h) | |
899 | { | |
900 | struct page *page; | |
901 | ||
902 | page = container_of((struct list_head *)h, struct page, lru); | |
903 | __free_slab(page->slab, page); | |
904 | } | |
905 | ||
906 | static void free_slab(struct kmem_cache *s, struct page *page) | |
907 | { | |
908 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
909 | /* | |
910 | * RCU free overloads the RCU head over the LRU | |
911 | */ | |
912 | struct rcu_head *head = (void *)&page->lru; | |
913 | ||
914 | call_rcu(head, rcu_free_slab); | |
915 | } else | |
916 | __free_slab(s, page); | |
917 | } | |
918 | ||
919 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
920 | { | |
921 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
922 | ||
923 | atomic_long_dec(&n->nr_slabs); | |
924 | reset_page_mapcount(page); | |
925 | page->flags &= ~(1 << PG_slab | 1 << PG_error); | |
926 | free_slab(s, page); | |
927 | } | |
928 | ||
929 | /* | |
930 | * Per slab locking using the pagelock | |
931 | */ | |
932 | static __always_inline void slab_lock(struct page *page) | |
933 | { | |
934 | bit_spin_lock(PG_locked, &page->flags); | |
935 | } | |
936 | ||
937 | static __always_inline void slab_unlock(struct page *page) | |
938 | { | |
939 | bit_spin_unlock(PG_locked, &page->flags); | |
940 | } | |
941 | ||
942 | static __always_inline int slab_trylock(struct page *page) | |
943 | { | |
944 | int rc = 1; | |
945 | ||
946 | rc = bit_spin_trylock(PG_locked, &page->flags); | |
947 | return rc; | |
948 | } | |
949 | ||
950 | /* | |
951 | * Management of partially allocated slabs | |
952 | */ | |
953 | static void add_partial(struct kmem_cache *s, struct page *page) | |
954 | { | |
955 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
956 | ||
957 | spin_lock(&n->list_lock); | |
958 | n->nr_partial++; | |
959 | list_add(&page->lru, &n->partial); | |
960 | spin_unlock(&n->list_lock); | |
961 | } | |
962 | ||
963 | static void remove_partial(struct kmem_cache *s, | |
964 | struct page *page) | |
965 | { | |
966 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
967 | ||
968 | spin_lock(&n->list_lock); | |
969 | list_del(&page->lru); | |
970 | n->nr_partial--; | |
971 | spin_unlock(&n->list_lock); | |
972 | } | |
973 | ||
974 | /* | |
975 | * Lock page and remove it from the partial list | |
976 | * | |
977 | * Must hold list_lock | |
978 | */ | |
979 | static int lock_and_del_slab(struct kmem_cache_node *n, struct page *page) | |
980 | { | |
981 | if (slab_trylock(page)) { | |
982 | list_del(&page->lru); | |
983 | n->nr_partial--; | |
984 | return 1; | |
985 | } | |
986 | return 0; | |
987 | } | |
988 | ||
989 | /* | |
990 | * Try to get a partial slab from a specific node | |
991 | */ | |
992 | static struct page *get_partial_node(struct kmem_cache_node *n) | |
993 | { | |
994 | struct page *page; | |
995 | ||
996 | /* | |
997 | * Racy check. If we mistakenly see no partial slabs then we | |
998 | * just allocate an empty slab. If we mistakenly try to get a | |
999 | * partial slab then get_partials() will return NULL. | |
1000 | */ | |
1001 | if (!n || !n->nr_partial) | |
1002 | return NULL; | |
1003 | ||
1004 | spin_lock(&n->list_lock); | |
1005 | list_for_each_entry(page, &n->partial, lru) | |
1006 | if (lock_and_del_slab(n, page)) | |
1007 | goto out; | |
1008 | page = NULL; | |
1009 | out: | |
1010 | spin_unlock(&n->list_lock); | |
1011 | return page; | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * Get a page from somewhere. Search in increasing NUMA | |
1016 | * distances. | |
1017 | */ | |
1018 | static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags) | |
1019 | { | |
1020 | #ifdef CONFIG_NUMA | |
1021 | struct zonelist *zonelist; | |
1022 | struct zone **z; | |
1023 | struct page *page; | |
1024 | ||
1025 | /* | |
1026 | * The defrag ratio allows to configure the tradeoffs between | |
1027 | * inter node defragmentation and node local allocations. | |
1028 | * A lower defrag_ratio increases the tendency to do local | |
1029 | * allocations instead of scanning throught the partial | |
1030 | * lists on other nodes. | |
1031 | * | |
1032 | * If defrag_ratio is set to 0 then kmalloc() always | |
1033 | * returns node local objects. If its higher then kmalloc() | |
1034 | * may return off node objects in order to avoid fragmentation. | |
1035 | * | |
1036 | * A higher ratio means slabs may be taken from other nodes | |
1037 | * thus reducing the number of partial slabs on those nodes. | |
1038 | * | |
1039 | * If /sys/slab/xx/defrag_ratio is set to 100 (which makes | |
1040 | * defrag_ratio = 1000) then every (well almost) allocation | |
1041 | * will first attempt to defrag slab caches on other nodes. This | |
1042 | * means scanning over all nodes to look for partial slabs which | |
1043 | * may be a bit expensive to do on every slab allocation. | |
1044 | */ | |
1045 | if (!s->defrag_ratio || get_cycles() % 1024 > s->defrag_ratio) | |
1046 | return NULL; | |
1047 | ||
1048 | zonelist = &NODE_DATA(slab_node(current->mempolicy)) | |
1049 | ->node_zonelists[gfp_zone(flags)]; | |
1050 | for (z = zonelist->zones; *z; z++) { | |
1051 | struct kmem_cache_node *n; | |
1052 | ||
1053 | n = get_node(s, zone_to_nid(*z)); | |
1054 | ||
1055 | if (n && cpuset_zone_allowed_hardwall(*z, flags) && | |
1056 | n->nr_partial > 2) { | |
1057 | page = get_partial_node(n); | |
1058 | if (page) | |
1059 | return page; | |
1060 | } | |
1061 | } | |
1062 | #endif | |
1063 | return NULL; | |
1064 | } | |
1065 | ||
1066 | /* | |
1067 | * Get a partial page, lock it and return it. | |
1068 | */ | |
1069 | static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node) | |
1070 | { | |
1071 | struct page *page; | |
1072 | int searchnode = (node == -1) ? numa_node_id() : node; | |
1073 | ||
1074 | page = get_partial_node(get_node(s, searchnode)); | |
1075 | if (page || (flags & __GFP_THISNODE)) | |
1076 | return page; | |
1077 | ||
1078 | return get_any_partial(s, flags); | |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * Move a page back to the lists. | |
1083 | * | |
1084 | * Must be called with the slab lock held. | |
1085 | * | |
1086 | * On exit the slab lock will have been dropped. | |
1087 | */ | |
1088 | static void putback_slab(struct kmem_cache *s, struct page *page) | |
1089 | { | |
1090 | if (page->inuse) { | |
1091 | if (page->freelist) | |
1092 | add_partial(s, page); | |
1093 | slab_unlock(page); | |
1094 | } else { | |
1095 | slab_unlock(page); | |
1096 | discard_slab(s, page); | |
1097 | } | |
1098 | } | |
1099 | ||
1100 | /* | |
1101 | * Remove the cpu slab | |
1102 | */ | |
1103 | static void deactivate_slab(struct kmem_cache *s, struct page *page, int cpu) | |
1104 | { | |
1105 | s->cpu_slab[cpu] = NULL; | |
1106 | ClearPageActive(page); | |
1107 | ||
1108 | putback_slab(s, page); | |
1109 | } | |
1110 | ||
1111 | static void flush_slab(struct kmem_cache *s, struct page *page, int cpu) | |
1112 | { | |
1113 | slab_lock(page); | |
1114 | deactivate_slab(s, page, cpu); | |
1115 | } | |
1116 | ||
1117 | /* | |
1118 | * Flush cpu slab. | |
1119 | * Called from IPI handler with interrupts disabled. | |
1120 | */ | |
1121 | static void __flush_cpu_slab(struct kmem_cache *s, int cpu) | |
1122 | { | |
1123 | struct page *page = s->cpu_slab[cpu]; | |
1124 | ||
1125 | if (likely(page)) | |
1126 | flush_slab(s, page, cpu); | |
1127 | } | |
1128 | ||
1129 | static void flush_cpu_slab(void *d) | |
1130 | { | |
1131 | struct kmem_cache *s = d; | |
1132 | int cpu = smp_processor_id(); | |
1133 | ||
1134 | __flush_cpu_slab(s, cpu); | |
1135 | } | |
1136 | ||
1137 | static void flush_all(struct kmem_cache *s) | |
1138 | { | |
1139 | #ifdef CONFIG_SMP | |
1140 | on_each_cpu(flush_cpu_slab, s, 1, 1); | |
1141 | #else | |
1142 | unsigned long flags; | |
1143 | ||
1144 | local_irq_save(flags); | |
1145 | flush_cpu_slab(s); | |
1146 | local_irq_restore(flags); | |
1147 | #endif | |
1148 | } | |
1149 | ||
1150 | /* | |
1151 | * slab_alloc is optimized to only modify two cachelines on the fast path | |
1152 | * (aside from the stack): | |
1153 | * | |
1154 | * 1. The page struct | |
1155 | * 2. The first cacheline of the object to be allocated. | |
1156 | * | |
1157 | * The only cache lines that are read (apart from code) is the | |
1158 | * per cpu array in the kmem_cache struct. | |
1159 | * | |
1160 | * Fastpath is not possible if we need to get a new slab or have | |
1161 | * debugging enabled (which means all slabs are marked with PageError) | |
1162 | */ | |
77c5e2d0 CL |
1163 | static void *slab_alloc(struct kmem_cache *s, |
1164 | gfp_t gfpflags, int node, void *addr) | |
81819f0f CL |
1165 | { |
1166 | struct page *page; | |
1167 | void **object; | |
1168 | unsigned long flags; | |
1169 | int cpu; | |
1170 | ||
1171 | local_irq_save(flags); | |
1172 | cpu = smp_processor_id(); | |
1173 | page = s->cpu_slab[cpu]; | |
1174 | if (!page) | |
1175 | goto new_slab; | |
1176 | ||
1177 | slab_lock(page); | |
1178 | if (unlikely(node != -1 && page_to_nid(page) != node)) | |
1179 | goto another_slab; | |
1180 | redo: | |
1181 | object = page->freelist; | |
1182 | if (unlikely(!object)) | |
1183 | goto another_slab; | |
1184 | if (unlikely(PageError(page))) | |
1185 | goto debug; | |
1186 | ||
1187 | have_object: | |
1188 | page->inuse++; | |
1189 | page->freelist = object[page->offset]; | |
1190 | slab_unlock(page); | |
1191 | local_irq_restore(flags); | |
1192 | return object; | |
1193 | ||
1194 | another_slab: | |
1195 | deactivate_slab(s, page, cpu); | |
1196 | ||
1197 | new_slab: | |
1198 | page = get_partial(s, gfpflags, node); | |
1199 | if (likely(page)) { | |
1200 | have_slab: | |
1201 | s->cpu_slab[cpu] = page; | |
1202 | SetPageActive(page); | |
1203 | goto redo; | |
1204 | } | |
1205 | ||
1206 | page = new_slab(s, gfpflags, node); | |
1207 | if (page) { | |
1208 | cpu = smp_processor_id(); | |
1209 | if (s->cpu_slab[cpu]) { | |
1210 | /* | |
1211 | * Someone else populated the cpu_slab while we enabled | |
1212 | * interrupts, or we have got scheduled on another cpu. | |
1213 | * The page may not be on the requested node. | |
1214 | */ | |
1215 | if (node == -1 || | |
1216 | page_to_nid(s->cpu_slab[cpu]) == node) { | |
1217 | /* | |
1218 | * Current cpuslab is acceptable and we | |
1219 | * want the current one since its cache hot | |
1220 | */ | |
1221 | discard_slab(s, page); | |
1222 | page = s->cpu_slab[cpu]; | |
1223 | slab_lock(page); | |
1224 | goto redo; | |
1225 | } | |
1226 | /* Dump the current slab */ | |
1227 | flush_slab(s, s->cpu_slab[cpu], cpu); | |
1228 | } | |
1229 | slab_lock(page); | |
1230 | goto have_slab; | |
1231 | } | |
1232 | local_irq_restore(flags); | |
1233 | return NULL; | |
1234 | debug: | |
1235 | if (!alloc_object_checks(s, page, object)) | |
1236 | goto another_slab; | |
1237 | if (s->flags & SLAB_STORE_USER) | |
77c5e2d0 | 1238 | set_track(s, object, TRACK_ALLOC, addr); |
81819f0f CL |
1239 | goto have_object; |
1240 | } | |
1241 | ||
1242 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
1243 | { | |
77c5e2d0 | 1244 | return slab_alloc(s, gfpflags, -1, __builtin_return_address(0)); |
81819f0f CL |
1245 | } |
1246 | EXPORT_SYMBOL(kmem_cache_alloc); | |
1247 | ||
1248 | #ifdef CONFIG_NUMA | |
1249 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
1250 | { | |
77c5e2d0 | 1251 | return slab_alloc(s, gfpflags, node, __builtin_return_address(0)); |
81819f0f CL |
1252 | } |
1253 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
1254 | #endif | |
1255 | ||
1256 | /* | |
1257 | * The fastpath only writes the cacheline of the page struct and the first | |
1258 | * cacheline of the object. | |
1259 | * | |
1260 | * No special cachelines need to be read | |
1261 | */ | |
77c5e2d0 CL |
1262 | static void slab_free(struct kmem_cache *s, struct page *page, |
1263 | void *x, void *addr) | |
81819f0f CL |
1264 | { |
1265 | void *prior; | |
1266 | void **object = (void *)x; | |
1267 | unsigned long flags; | |
1268 | ||
1269 | local_irq_save(flags); | |
1270 | slab_lock(page); | |
1271 | ||
1272 | if (unlikely(PageError(page))) | |
1273 | goto debug; | |
1274 | checks_ok: | |
1275 | prior = object[page->offset] = page->freelist; | |
1276 | page->freelist = object; | |
1277 | page->inuse--; | |
1278 | ||
1279 | if (unlikely(PageActive(page))) | |
1280 | /* | |
1281 | * Cpu slabs are never on partial lists and are | |
1282 | * never freed. | |
1283 | */ | |
1284 | goto out_unlock; | |
1285 | ||
1286 | if (unlikely(!page->inuse)) | |
1287 | goto slab_empty; | |
1288 | ||
1289 | /* | |
1290 | * Objects left in the slab. If it | |
1291 | * was not on the partial list before | |
1292 | * then add it. | |
1293 | */ | |
1294 | if (unlikely(!prior)) | |
1295 | add_partial(s, page); | |
1296 | ||
1297 | out_unlock: | |
1298 | slab_unlock(page); | |
1299 | local_irq_restore(flags); | |
1300 | return; | |
1301 | ||
1302 | slab_empty: | |
1303 | if (prior) | |
1304 | /* | |
1305 | * Partially used slab that is on the partial list. | |
1306 | */ | |
1307 | remove_partial(s, page); | |
1308 | ||
1309 | slab_unlock(page); | |
1310 | discard_slab(s, page); | |
1311 | local_irq_restore(flags); | |
1312 | return; | |
1313 | ||
1314 | debug: | |
77c5e2d0 CL |
1315 | if (!free_object_checks(s, page, x)) |
1316 | goto out_unlock; | |
1317 | if (s->flags & SLAB_STORE_USER) | |
1318 | set_track(s, x, TRACK_FREE, addr); | |
1319 | goto checks_ok; | |
81819f0f CL |
1320 | } |
1321 | ||
1322 | void kmem_cache_free(struct kmem_cache *s, void *x) | |
1323 | { | |
77c5e2d0 | 1324 | struct page *page; |
81819f0f | 1325 | |
b49af68f | 1326 | page = virt_to_head_page(x); |
81819f0f | 1327 | |
77c5e2d0 | 1328 | slab_free(s, page, x, __builtin_return_address(0)); |
81819f0f CL |
1329 | } |
1330 | EXPORT_SYMBOL(kmem_cache_free); | |
1331 | ||
1332 | /* Figure out on which slab object the object resides */ | |
1333 | static struct page *get_object_page(const void *x) | |
1334 | { | |
b49af68f | 1335 | struct page *page = virt_to_head_page(x); |
81819f0f CL |
1336 | |
1337 | if (!PageSlab(page)) | |
1338 | return NULL; | |
1339 | ||
1340 | return page; | |
1341 | } | |
1342 | ||
1343 | /* | |
1344 | * kmem_cache_open produces objects aligned at "size" and the first object | |
1345 | * is placed at offset 0 in the slab (We have no metainformation on the | |
1346 | * slab, all slabs are in essence "off slab"). | |
1347 | * | |
1348 | * In order to get the desired alignment one just needs to align the | |
1349 | * size. | |
1350 | * | |
1351 | * Notice that the allocation order determines the sizes of the per cpu | |
1352 | * caches. Each processor has always one slab available for allocations. | |
1353 | * Increasing the allocation order reduces the number of times that slabs | |
1354 | * must be moved on and off the partial lists and therefore may influence | |
1355 | * locking overhead. | |
1356 | * | |
1357 | * The offset is used to relocate the free list link in each object. It is | |
1358 | * therefore possible to move the free list link behind the object. This | |
1359 | * is necessary for RCU to work properly and also useful for debugging. | |
1360 | */ | |
1361 | ||
1362 | /* | |
1363 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
1364 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
1365 | * and increases the number of allocations possible without having to | |
1366 | * take the list_lock. | |
1367 | */ | |
1368 | static int slub_min_order; | |
1369 | static int slub_max_order = DEFAULT_MAX_ORDER; | |
1370 | ||
1371 | /* | |
1372 | * Minimum number of objects per slab. This is necessary in order to | |
1373 | * reduce locking overhead. Similar to the queue size in SLAB. | |
1374 | */ | |
1375 | static int slub_min_objects = DEFAULT_MIN_OBJECTS; | |
1376 | ||
1377 | /* | |
1378 | * Merge control. If this is set then no merging of slab caches will occur. | |
1379 | */ | |
1380 | static int slub_nomerge; | |
1381 | ||
1382 | /* | |
1383 | * Debug settings: | |
1384 | */ | |
1385 | static int slub_debug; | |
1386 | ||
1387 | static char *slub_debug_slabs; | |
1388 | ||
1389 | /* | |
1390 | * Calculate the order of allocation given an slab object size. | |
1391 | * | |
1392 | * The order of allocation has significant impact on other elements | |
1393 | * of the system. Generally order 0 allocations should be preferred | |
1394 | * since they do not cause fragmentation in the page allocator. Larger | |
1395 | * objects may have problems with order 0 because there may be too much | |
1396 | * space left unused in a slab. We go to a higher order if more than 1/8th | |
1397 | * of the slab would be wasted. | |
1398 | * | |
1399 | * In order to reach satisfactory performance we must ensure that | |
1400 | * a minimum number of objects is in one slab. Otherwise we may | |
1401 | * generate too much activity on the partial lists. This is less a | |
1402 | * concern for large slabs though. slub_max_order specifies the order | |
1403 | * where we begin to stop considering the number of objects in a slab. | |
1404 | * | |
1405 | * Higher order allocations also allow the placement of more objects | |
1406 | * in a slab and thereby reduce object handling overhead. If the user | |
1407 | * has requested a higher mininum order then we start with that one | |
1408 | * instead of zero. | |
1409 | */ | |
1410 | static int calculate_order(int size) | |
1411 | { | |
1412 | int order; | |
1413 | int rem; | |
1414 | ||
1415 | for (order = max(slub_min_order, fls(size - 1) - PAGE_SHIFT); | |
1416 | order < MAX_ORDER; order++) { | |
1417 | unsigned long slab_size = PAGE_SIZE << order; | |
1418 | ||
1419 | if (slub_max_order > order && | |
1420 | slab_size < slub_min_objects * size) | |
1421 | continue; | |
1422 | ||
1423 | if (slab_size < size) | |
1424 | continue; | |
1425 | ||
1426 | rem = slab_size % size; | |
1427 | ||
1428 | if (rem <= (PAGE_SIZE << order) / 8) | |
1429 | break; | |
1430 | ||
1431 | } | |
1432 | if (order >= MAX_ORDER) | |
1433 | return -E2BIG; | |
1434 | return order; | |
1435 | } | |
1436 | ||
1437 | /* | |
1438 | * Function to figure out which alignment to use from the | |
1439 | * various ways of specifying it. | |
1440 | */ | |
1441 | static unsigned long calculate_alignment(unsigned long flags, | |
1442 | unsigned long align, unsigned long size) | |
1443 | { | |
1444 | /* | |
1445 | * If the user wants hardware cache aligned objects then | |
1446 | * follow that suggestion if the object is sufficiently | |
1447 | * large. | |
1448 | * | |
1449 | * The hardware cache alignment cannot override the | |
1450 | * specified alignment though. If that is greater | |
1451 | * then use it. | |
1452 | */ | |
1453 | if ((flags & (SLAB_MUST_HWCACHE_ALIGN | SLAB_HWCACHE_ALIGN)) && | |
1454 | size > L1_CACHE_BYTES / 2) | |
1455 | return max_t(unsigned long, align, L1_CACHE_BYTES); | |
1456 | ||
1457 | if (align < ARCH_SLAB_MINALIGN) | |
1458 | return ARCH_SLAB_MINALIGN; | |
1459 | ||
1460 | return ALIGN(align, sizeof(void *)); | |
1461 | } | |
1462 | ||
1463 | static void init_kmem_cache_node(struct kmem_cache_node *n) | |
1464 | { | |
1465 | n->nr_partial = 0; | |
1466 | atomic_long_set(&n->nr_slabs, 0); | |
1467 | spin_lock_init(&n->list_lock); | |
1468 | INIT_LIST_HEAD(&n->partial); | |
1469 | } | |
1470 | ||
1471 | #ifdef CONFIG_NUMA | |
1472 | /* | |
1473 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
1474 | * slab on the node for this slabcache. There are no concurrent accesses | |
1475 | * possible. | |
1476 | * | |
1477 | * Note that this function only works on the kmalloc_node_cache | |
1478 | * when allocating for the kmalloc_node_cache. | |
1479 | */ | |
1480 | static struct kmem_cache_node * __init early_kmem_cache_node_alloc(gfp_t gfpflags, | |
1481 | int node) | |
1482 | { | |
1483 | struct page *page; | |
1484 | struct kmem_cache_node *n; | |
1485 | ||
1486 | BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node)); | |
1487 | ||
1488 | page = new_slab(kmalloc_caches, gfpflags | GFP_THISNODE, node); | |
1489 | /* new_slab() disables interupts */ | |
1490 | local_irq_enable(); | |
1491 | ||
1492 | BUG_ON(!page); | |
1493 | n = page->freelist; | |
1494 | BUG_ON(!n); | |
1495 | page->freelist = get_freepointer(kmalloc_caches, n); | |
1496 | page->inuse++; | |
1497 | kmalloc_caches->node[node] = n; | |
1498 | init_object(kmalloc_caches, n, 1); | |
1499 | init_kmem_cache_node(n); | |
1500 | atomic_long_inc(&n->nr_slabs); | |
1501 | add_partial(kmalloc_caches, page); | |
1502 | return n; | |
1503 | } | |
1504 | ||
1505 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
1506 | { | |
1507 | int node; | |
1508 | ||
1509 | for_each_online_node(node) { | |
1510 | struct kmem_cache_node *n = s->node[node]; | |
1511 | if (n && n != &s->local_node) | |
1512 | kmem_cache_free(kmalloc_caches, n); | |
1513 | s->node[node] = NULL; | |
1514 | } | |
1515 | } | |
1516 | ||
1517 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
1518 | { | |
1519 | int node; | |
1520 | int local_node; | |
1521 | ||
1522 | if (slab_state >= UP) | |
1523 | local_node = page_to_nid(virt_to_page(s)); | |
1524 | else | |
1525 | local_node = 0; | |
1526 | ||
1527 | for_each_online_node(node) { | |
1528 | struct kmem_cache_node *n; | |
1529 | ||
1530 | if (local_node == node) | |
1531 | n = &s->local_node; | |
1532 | else { | |
1533 | if (slab_state == DOWN) { | |
1534 | n = early_kmem_cache_node_alloc(gfpflags, | |
1535 | node); | |
1536 | continue; | |
1537 | } | |
1538 | n = kmem_cache_alloc_node(kmalloc_caches, | |
1539 | gfpflags, node); | |
1540 | ||
1541 | if (!n) { | |
1542 | free_kmem_cache_nodes(s); | |
1543 | return 0; | |
1544 | } | |
1545 | ||
1546 | } | |
1547 | s->node[node] = n; | |
1548 | init_kmem_cache_node(n); | |
1549 | } | |
1550 | return 1; | |
1551 | } | |
1552 | #else | |
1553 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
1554 | { | |
1555 | } | |
1556 | ||
1557 | static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags) | |
1558 | { | |
1559 | init_kmem_cache_node(&s->local_node); | |
1560 | return 1; | |
1561 | } | |
1562 | #endif | |
1563 | ||
1564 | /* | |
1565 | * calculate_sizes() determines the order and the distribution of data within | |
1566 | * a slab object. | |
1567 | */ | |
1568 | static int calculate_sizes(struct kmem_cache *s) | |
1569 | { | |
1570 | unsigned long flags = s->flags; | |
1571 | unsigned long size = s->objsize; | |
1572 | unsigned long align = s->align; | |
1573 | ||
1574 | /* | |
1575 | * Determine if we can poison the object itself. If the user of | |
1576 | * the slab may touch the object after free or before allocation | |
1577 | * then we should never poison the object itself. | |
1578 | */ | |
1579 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
1580 | !s->ctor && !s->dtor) | |
1581 | s->flags |= __OBJECT_POISON; | |
1582 | else | |
1583 | s->flags &= ~__OBJECT_POISON; | |
1584 | ||
1585 | /* | |
1586 | * Round up object size to the next word boundary. We can only | |
1587 | * place the free pointer at word boundaries and this determines | |
1588 | * the possible location of the free pointer. | |
1589 | */ | |
1590 | size = ALIGN(size, sizeof(void *)); | |
1591 | ||
1592 | /* | |
1593 | * If we are redzoning then check if there is some space between the | |
1594 | * end of the object and the free pointer. If not then add an | |
1595 | * additional word, so that we can establish a redzone between | |
1596 | * the object and the freepointer to be able to check for overwrites. | |
1597 | */ | |
1598 | if ((flags & SLAB_RED_ZONE) && size == s->objsize) | |
1599 | size += sizeof(void *); | |
1600 | ||
1601 | /* | |
1602 | * With that we have determined how much of the slab is in actual | |
1603 | * use by the object. This is the potential offset to the free | |
1604 | * pointer. | |
1605 | */ | |
1606 | s->inuse = size; | |
1607 | ||
1608 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
1609 | s->ctor || s->dtor)) { | |
1610 | /* | |
1611 | * Relocate free pointer after the object if it is not | |
1612 | * permitted to overwrite the first word of the object on | |
1613 | * kmem_cache_free. | |
1614 | * | |
1615 | * This is the case if we do RCU, have a constructor or | |
1616 | * destructor or are poisoning the objects. | |
1617 | */ | |
1618 | s->offset = size; | |
1619 | size += sizeof(void *); | |
1620 | } | |
1621 | ||
1622 | if (flags & SLAB_STORE_USER) | |
1623 | /* | |
1624 | * Need to store information about allocs and frees after | |
1625 | * the object. | |
1626 | */ | |
1627 | size += 2 * sizeof(struct track); | |
1628 | ||
1629 | if (flags & DEBUG_DEFAULT_FLAGS) | |
1630 | /* | |
1631 | * Add some empty padding so that we can catch | |
1632 | * overwrites from earlier objects rather than let | |
1633 | * tracking information or the free pointer be | |
1634 | * corrupted if an user writes before the start | |
1635 | * of the object. | |
1636 | */ | |
1637 | size += sizeof(void *); | |
1638 | /* | |
1639 | * Determine the alignment based on various parameters that the | |
1640 | * user specified (this is unecessarily complex due to the attempt | |
1641 | * to be compatible with SLAB. Should be cleaned up some day). | |
1642 | */ | |
1643 | align = calculate_alignment(flags, align, s->objsize); | |
1644 | ||
1645 | /* | |
1646 | * SLUB stores one object immediately after another beginning from | |
1647 | * offset 0. In order to align the objects we have to simply size | |
1648 | * each object to conform to the alignment. | |
1649 | */ | |
1650 | size = ALIGN(size, align); | |
1651 | s->size = size; | |
1652 | ||
1653 | s->order = calculate_order(size); | |
1654 | if (s->order < 0) | |
1655 | return 0; | |
1656 | ||
1657 | /* | |
1658 | * Determine the number of objects per slab | |
1659 | */ | |
1660 | s->objects = (PAGE_SIZE << s->order) / size; | |
1661 | ||
1662 | /* | |
1663 | * Verify that the number of objects is within permitted limits. | |
1664 | * The page->inuse field is only 16 bit wide! So we cannot have | |
1665 | * more than 64k objects per slab. | |
1666 | */ | |
1667 | if (!s->objects || s->objects > 65535) | |
1668 | return 0; | |
1669 | return 1; | |
1670 | ||
1671 | } | |
1672 | ||
1673 | static int __init finish_bootstrap(void) | |
1674 | { | |
1675 | struct list_head *h; | |
1676 | int err; | |
1677 | ||
1678 | slab_state = SYSFS; | |
1679 | ||
1680 | list_for_each(h, &slab_caches) { | |
1681 | struct kmem_cache *s = | |
1682 | container_of(h, struct kmem_cache, list); | |
1683 | ||
1684 | err = sysfs_slab_add(s); | |
1685 | BUG_ON(err); | |
1686 | } | |
1687 | return 0; | |
1688 | } | |
1689 | ||
1690 | static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags, | |
1691 | const char *name, size_t size, | |
1692 | size_t align, unsigned long flags, | |
1693 | void (*ctor)(void *, struct kmem_cache *, unsigned long), | |
1694 | void (*dtor)(void *, struct kmem_cache *, unsigned long)) | |
1695 | { | |
1696 | memset(s, 0, kmem_size); | |
1697 | s->name = name; | |
1698 | s->ctor = ctor; | |
1699 | s->dtor = dtor; | |
1700 | s->objsize = size; | |
1701 | s->flags = flags; | |
1702 | s->align = align; | |
1703 | ||
1704 | BUG_ON(flags & SLUB_UNIMPLEMENTED); | |
1705 | ||
1706 | /* | |
1707 | * The page->offset field is only 16 bit wide. This is an offset | |
1708 | * in units of words from the beginning of an object. If the slab | |
1709 | * size is bigger then we cannot move the free pointer behind the | |
1710 | * object anymore. | |
1711 | * | |
1712 | * On 32 bit platforms the limit is 256k. On 64bit platforms | |
1713 | * the limit is 512k. | |
1714 | * | |
1715 | * Debugging or ctor/dtors may create a need to move the free | |
1716 | * pointer. Fail if this happens. | |
1717 | */ | |
1718 | if (s->size >= 65535 * sizeof(void *)) { | |
1719 | BUG_ON(flags & (SLAB_RED_ZONE | SLAB_POISON | | |
1720 | SLAB_STORE_USER | SLAB_DESTROY_BY_RCU)); | |
1721 | BUG_ON(ctor || dtor); | |
1722 | } | |
1723 | else | |
1724 | /* | |
1725 | * Enable debugging if selected on the kernel commandline. | |
1726 | */ | |
1727 | if (slub_debug && (!slub_debug_slabs || | |
1728 | strncmp(slub_debug_slabs, name, | |
1729 | strlen(slub_debug_slabs)) == 0)) | |
1730 | s->flags |= slub_debug; | |
1731 | ||
1732 | if (!calculate_sizes(s)) | |
1733 | goto error; | |
1734 | ||
1735 | s->refcount = 1; | |
1736 | #ifdef CONFIG_NUMA | |
1737 | s->defrag_ratio = 100; | |
1738 | #endif | |
1739 | ||
1740 | if (init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA)) | |
1741 | return 1; | |
1742 | error: | |
1743 | if (flags & SLAB_PANIC) | |
1744 | panic("Cannot create slab %s size=%lu realsize=%u " | |
1745 | "order=%u offset=%u flags=%lx\n", | |
1746 | s->name, (unsigned long)size, s->size, s->order, | |
1747 | s->offset, flags); | |
1748 | return 0; | |
1749 | } | |
1750 | EXPORT_SYMBOL(kmem_cache_open); | |
1751 | ||
1752 | /* | |
1753 | * Check if a given pointer is valid | |
1754 | */ | |
1755 | int kmem_ptr_validate(struct kmem_cache *s, const void *object) | |
1756 | { | |
1757 | struct page * page; | |
1758 | void *addr; | |
1759 | ||
1760 | page = get_object_page(object); | |
1761 | ||
1762 | if (!page || s != page->slab) | |
1763 | /* No slab or wrong slab */ | |
1764 | return 0; | |
1765 | ||
1766 | addr = page_address(page); | |
1767 | if (object < addr || object >= addr + s->objects * s->size) | |
1768 | /* Out of bounds */ | |
1769 | return 0; | |
1770 | ||
1771 | if ((object - addr) % s->size) | |
1772 | /* Improperly aligned */ | |
1773 | return 0; | |
1774 | ||
1775 | /* | |
1776 | * We could also check if the object is on the slabs freelist. | |
1777 | * But this would be too expensive and it seems that the main | |
1778 | * purpose of kmem_ptr_valid is to check if the object belongs | |
1779 | * to a certain slab. | |
1780 | */ | |
1781 | return 1; | |
1782 | } | |
1783 | EXPORT_SYMBOL(kmem_ptr_validate); | |
1784 | ||
1785 | /* | |
1786 | * Determine the size of a slab object | |
1787 | */ | |
1788 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
1789 | { | |
1790 | return s->objsize; | |
1791 | } | |
1792 | EXPORT_SYMBOL(kmem_cache_size); | |
1793 | ||
1794 | const char *kmem_cache_name(struct kmem_cache *s) | |
1795 | { | |
1796 | return s->name; | |
1797 | } | |
1798 | EXPORT_SYMBOL(kmem_cache_name); | |
1799 | ||
1800 | /* | |
1801 | * Attempt to free all slabs on a node | |
1802 | */ | |
1803 | static int free_list(struct kmem_cache *s, struct kmem_cache_node *n, | |
1804 | struct list_head *list) | |
1805 | { | |
1806 | int slabs_inuse = 0; | |
1807 | unsigned long flags; | |
1808 | struct page *page, *h; | |
1809 | ||
1810 | spin_lock_irqsave(&n->list_lock, flags); | |
1811 | list_for_each_entry_safe(page, h, list, lru) | |
1812 | if (!page->inuse) { | |
1813 | list_del(&page->lru); | |
1814 | discard_slab(s, page); | |
1815 | } else | |
1816 | slabs_inuse++; | |
1817 | spin_unlock_irqrestore(&n->list_lock, flags); | |
1818 | return slabs_inuse; | |
1819 | } | |
1820 | ||
1821 | /* | |
1822 | * Release all resources used by slab cache | |
1823 | */ | |
1824 | static int kmem_cache_close(struct kmem_cache *s) | |
1825 | { | |
1826 | int node; | |
1827 | ||
1828 | flush_all(s); | |
1829 | ||
1830 | /* Attempt to free all objects */ | |
1831 | for_each_online_node(node) { | |
1832 | struct kmem_cache_node *n = get_node(s, node); | |
1833 | ||
1834 | free_list(s, n, &n->partial); | |
1835 | if (atomic_long_read(&n->nr_slabs)) | |
1836 | return 1; | |
1837 | } | |
1838 | free_kmem_cache_nodes(s); | |
1839 | return 0; | |
1840 | } | |
1841 | ||
1842 | /* | |
1843 | * Close a cache and release the kmem_cache structure | |
1844 | * (must be used for caches created using kmem_cache_create) | |
1845 | */ | |
1846 | void kmem_cache_destroy(struct kmem_cache *s) | |
1847 | { | |
1848 | down_write(&slub_lock); | |
1849 | s->refcount--; | |
1850 | if (!s->refcount) { | |
1851 | list_del(&s->list); | |
1852 | if (kmem_cache_close(s)) | |
1853 | WARN_ON(1); | |
1854 | sysfs_slab_remove(s); | |
1855 | kfree(s); | |
1856 | } | |
1857 | up_write(&slub_lock); | |
1858 | } | |
1859 | EXPORT_SYMBOL(kmem_cache_destroy); | |
1860 | ||
1861 | /******************************************************************** | |
1862 | * Kmalloc subsystem | |
1863 | *******************************************************************/ | |
1864 | ||
1865 | struct kmem_cache kmalloc_caches[KMALLOC_SHIFT_HIGH + 1] __cacheline_aligned; | |
1866 | EXPORT_SYMBOL(kmalloc_caches); | |
1867 | ||
1868 | #ifdef CONFIG_ZONE_DMA | |
1869 | static struct kmem_cache *kmalloc_caches_dma[KMALLOC_SHIFT_HIGH + 1]; | |
1870 | #endif | |
1871 | ||
1872 | static int __init setup_slub_min_order(char *str) | |
1873 | { | |
1874 | get_option (&str, &slub_min_order); | |
1875 | ||
1876 | return 1; | |
1877 | } | |
1878 | ||
1879 | __setup("slub_min_order=", setup_slub_min_order); | |
1880 | ||
1881 | static int __init setup_slub_max_order(char *str) | |
1882 | { | |
1883 | get_option (&str, &slub_max_order); | |
1884 | ||
1885 | return 1; | |
1886 | } | |
1887 | ||
1888 | __setup("slub_max_order=", setup_slub_max_order); | |
1889 | ||
1890 | static int __init setup_slub_min_objects(char *str) | |
1891 | { | |
1892 | get_option (&str, &slub_min_objects); | |
1893 | ||
1894 | return 1; | |
1895 | } | |
1896 | ||
1897 | __setup("slub_min_objects=", setup_slub_min_objects); | |
1898 | ||
1899 | static int __init setup_slub_nomerge(char *str) | |
1900 | { | |
1901 | slub_nomerge = 1; | |
1902 | return 1; | |
1903 | } | |
1904 | ||
1905 | __setup("slub_nomerge", setup_slub_nomerge); | |
1906 | ||
1907 | static int __init setup_slub_debug(char *str) | |
1908 | { | |
1909 | if (!str || *str != '=') | |
1910 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
1911 | else { | |
1912 | str++; | |
1913 | if (*str == 0 || *str == ',') | |
1914 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
1915 | else | |
1916 | for( ;*str && *str != ','; str++) | |
1917 | switch (*str) { | |
1918 | case 'f' : case 'F' : | |
1919 | slub_debug |= SLAB_DEBUG_FREE; | |
1920 | break; | |
1921 | case 'z' : case 'Z' : | |
1922 | slub_debug |= SLAB_RED_ZONE; | |
1923 | break; | |
1924 | case 'p' : case 'P' : | |
1925 | slub_debug |= SLAB_POISON; | |
1926 | break; | |
1927 | case 'u' : case 'U' : | |
1928 | slub_debug |= SLAB_STORE_USER; | |
1929 | break; | |
1930 | case 't' : case 'T' : | |
1931 | slub_debug |= SLAB_TRACE; | |
1932 | break; | |
1933 | default: | |
1934 | printk(KERN_ERR "slub_debug option '%c' " | |
1935 | "unknown. skipped\n",*str); | |
1936 | } | |
1937 | } | |
1938 | ||
1939 | if (*str == ',') | |
1940 | slub_debug_slabs = str + 1; | |
1941 | return 1; | |
1942 | } | |
1943 | ||
1944 | __setup("slub_debug", setup_slub_debug); | |
1945 | ||
1946 | static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s, | |
1947 | const char *name, int size, gfp_t gfp_flags) | |
1948 | { | |
1949 | unsigned int flags = 0; | |
1950 | ||
1951 | if (gfp_flags & SLUB_DMA) | |
1952 | flags = SLAB_CACHE_DMA; | |
1953 | ||
1954 | down_write(&slub_lock); | |
1955 | if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN, | |
1956 | flags, NULL, NULL)) | |
1957 | goto panic; | |
1958 | ||
1959 | list_add(&s->list, &slab_caches); | |
1960 | up_write(&slub_lock); | |
1961 | if (sysfs_slab_add(s)) | |
1962 | goto panic; | |
1963 | return s; | |
1964 | ||
1965 | panic: | |
1966 | panic("Creation of kmalloc slab %s size=%d failed.\n", name, size); | |
1967 | } | |
1968 | ||
1969 | static struct kmem_cache *get_slab(size_t size, gfp_t flags) | |
1970 | { | |
1971 | int index = kmalloc_index(size); | |
1972 | ||
614410d5 | 1973 | if (!index) |
81819f0f CL |
1974 | return NULL; |
1975 | ||
1976 | /* Allocation too large? */ | |
1977 | BUG_ON(index < 0); | |
1978 | ||
1979 | #ifdef CONFIG_ZONE_DMA | |
1980 | if ((flags & SLUB_DMA)) { | |
1981 | struct kmem_cache *s; | |
1982 | struct kmem_cache *x; | |
1983 | char *text; | |
1984 | size_t realsize; | |
1985 | ||
1986 | s = kmalloc_caches_dma[index]; | |
1987 | if (s) | |
1988 | return s; | |
1989 | ||
1990 | /* Dynamically create dma cache */ | |
1991 | x = kmalloc(kmem_size, flags & ~SLUB_DMA); | |
1992 | if (!x) | |
1993 | panic("Unable to allocate memory for dma cache\n"); | |
1994 | ||
1995 | if (index <= KMALLOC_SHIFT_HIGH) | |
1996 | realsize = 1 << index; | |
1997 | else { | |
1998 | if (index == 1) | |
1999 | realsize = 96; | |
2000 | else | |
2001 | realsize = 192; | |
2002 | } | |
2003 | ||
2004 | text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d", | |
2005 | (unsigned int)realsize); | |
2006 | s = create_kmalloc_cache(x, text, realsize, flags); | |
2007 | kmalloc_caches_dma[index] = s; | |
2008 | return s; | |
2009 | } | |
2010 | #endif | |
2011 | return &kmalloc_caches[index]; | |
2012 | } | |
2013 | ||
2014 | void *__kmalloc(size_t size, gfp_t flags) | |
2015 | { | |
2016 | struct kmem_cache *s = get_slab(size, flags); | |
2017 | ||
2018 | if (s) | |
77c5e2d0 | 2019 | return slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2020 | return NULL; |
2021 | } | |
2022 | EXPORT_SYMBOL(__kmalloc); | |
2023 | ||
2024 | #ifdef CONFIG_NUMA | |
2025 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
2026 | { | |
2027 | struct kmem_cache *s = get_slab(size, flags); | |
2028 | ||
2029 | if (s) | |
77c5e2d0 | 2030 | return slab_alloc(s, flags, node, __builtin_return_address(0)); |
81819f0f CL |
2031 | return NULL; |
2032 | } | |
2033 | EXPORT_SYMBOL(__kmalloc_node); | |
2034 | #endif | |
2035 | ||
2036 | size_t ksize(const void *object) | |
2037 | { | |
2038 | struct page *page = get_object_page(object); | |
2039 | struct kmem_cache *s; | |
2040 | ||
2041 | BUG_ON(!page); | |
2042 | s = page->slab; | |
2043 | BUG_ON(!s); | |
2044 | ||
2045 | /* | |
2046 | * Debugging requires use of the padding between object | |
2047 | * and whatever may come after it. | |
2048 | */ | |
2049 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
2050 | return s->objsize; | |
2051 | ||
2052 | /* | |
2053 | * If we have the need to store the freelist pointer | |
2054 | * back there or track user information then we can | |
2055 | * only use the space before that information. | |
2056 | */ | |
2057 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
2058 | return s->inuse; | |
2059 | ||
2060 | /* | |
2061 | * Else we can use all the padding etc for the allocation | |
2062 | */ | |
2063 | return s->size; | |
2064 | } | |
2065 | EXPORT_SYMBOL(ksize); | |
2066 | ||
2067 | void kfree(const void *x) | |
2068 | { | |
2069 | struct kmem_cache *s; | |
2070 | struct page *page; | |
2071 | ||
2072 | if (!x) | |
2073 | return; | |
2074 | ||
b49af68f | 2075 | page = virt_to_head_page(x); |
81819f0f CL |
2076 | s = page->slab; |
2077 | ||
77c5e2d0 | 2078 | slab_free(s, page, (void *)x, __builtin_return_address(0)); |
81819f0f CL |
2079 | } |
2080 | EXPORT_SYMBOL(kfree); | |
2081 | ||
2082 | /** | |
2083 | * krealloc - reallocate memory. The contents will remain unchanged. | |
2084 | * | |
2085 | * @p: object to reallocate memory for. | |
2086 | * @new_size: how many bytes of memory are required. | |
2087 | * @flags: the type of memory to allocate. | |
2088 | * | |
2089 | * The contents of the object pointed to are preserved up to the | |
2090 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
2091 | * behaves exactly like kmalloc(). If @size is 0 and @p is not a | |
2092 | * %NULL pointer, the object pointed to is freed. | |
2093 | */ | |
2094 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
2095 | { | |
2096 | struct kmem_cache *new_cache; | |
2097 | void *ret; | |
2098 | struct page *page; | |
2099 | ||
2100 | if (unlikely(!p)) | |
2101 | return kmalloc(new_size, flags); | |
2102 | ||
2103 | if (unlikely(!new_size)) { | |
2104 | kfree(p); | |
2105 | return NULL; | |
2106 | } | |
2107 | ||
b49af68f | 2108 | page = virt_to_head_page(p); |
81819f0f CL |
2109 | |
2110 | new_cache = get_slab(new_size, flags); | |
2111 | ||
2112 | /* | |
2113 | * If new size fits in the current cache, bail out. | |
2114 | */ | |
2115 | if (likely(page->slab == new_cache)) | |
2116 | return (void *)p; | |
2117 | ||
2118 | ret = kmalloc(new_size, flags); | |
2119 | if (ret) { | |
2120 | memcpy(ret, p, min(new_size, ksize(p))); | |
2121 | kfree(p); | |
2122 | } | |
2123 | return ret; | |
2124 | } | |
2125 | EXPORT_SYMBOL(krealloc); | |
2126 | ||
2127 | /******************************************************************** | |
2128 | * Basic setup of slabs | |
2129 | *******************************************************************/ | |
2130 | ||
2131 | void __init kmem_cache_init(void) | |
2132 | { | |
2133 | int i; | |
2134 | ||
2135 | #ifdef CONFIG_NUMA | |
2136 | /* | |
2137 | * Must first have the slab cache available for the allocations of the | |
2138 | * struct kmalloc_cache_node's. There is special bootstrap code in | |
2139 | * kmem_cache_open for slab_state == DOWN. | |
2140 | */ | |
2141 | create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node", | |
2142 | sizeof(struct kmem_cache_node), GFP_KERNEL); | |
2143 | #endif | |
2144 | ||
2145 | /* Able to allocate the per node structures */ | |
2146 | slab_state = PARTIAL; | |
2147 | ||
2148 | /* Caches that are not of the two-to-the-power-of size */ | |
2149 | create_kmalloc_cache(&kmalloc_caches[1], | |
2150 | "kmalloc-96", 96, GFP_KERNEL); | |
2151 | create_kmalloc_cache(&kmalloc_caches[2], | |
2152 | "kmalloc-192", 192, GFP_KERNEL); | |
2153 | ||
2154 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) | |
2155 | create_kmalloc_cache(&kmalloc_caches[i], | |
2156 | "kmalloc", 1 << i, GFP_KERNEL); | |
2157 | ||
2158 | slab_state = UP; | |
2159 | ||
2160 | /* Provide the correct kmalloc names now that the caches are up */ | |
2161 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) | |
2162 | kmalloc_caches[i]. name = | |
2163 | kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i); | |
2164 | ||
2165 | #ifdef CONFIG_SMP | |
2166 | register_cpu_notifier(&slab_notifier); | |
2167 | #endif | |
2168 | ||
2169 | if (nr_cpu_ids) /* Remove when nr_cpu_ids is fixed upstream ! */ | |
2170 | kmem_size = offsetof(struct kmem_cache, cpu_slab) | |
2171 | + nr_cpu_ids * sizeof(struct page *); | |
2172 | ||
2173 | printk(KERN_INFO "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d," | |
2174 | " Processors=%d, Nodes=%d\n", | |
2175 | KMALLOC_SHIFT_HIGH, L1_CACHE_BYTES, | |
2176 | slub_min_order, slub_max_order, slub_min_objects, | |
2177 | nr_cpu_ids, nr_node_ids); | |
2178 | } | |
2179 | ||
2180 | /* | |
2181 | * Find a mergeable slab cache | |
2182 | */ | |
2183 | static int slab_unmergeable(struct kmem_cache *s) | |
2184 | { | |
2185 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
2186 | return 1; | |
2187 | ||
2188 | if (s->ctor || s->dtor) | |
2189 | return 1; | |
2190 | ||
2191 | return 0; | |
2192 | } | |
2193 | ||
2194 | static struct kmem_cache *find_mergeable(size_t size, | |
2195 | size_t align, unsigned long flags, | |
2196 | void (*ctor)(void *, struct kmem_cache *, unsigned long), | |
2197 | void (*dtor)(void *, struct kmem_cache *, unsigned long)) | |
2198 | { | |
2199 | struct list_head *h; | |
2200 | ||
2201 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
2202 | return NULL; | |
2203 | ||
2204 | if (ctor || dtor) | |
2205 | return NULL; | |
2206 | ||
2207 | size = ALIGN(size, sizeof(void *)); | |
2208 | align = calculate_alignment(flags, align, size); | |
2209 | size = ALIGN(size, align); | |
2210 | ||
2211 | list_for_each(h, &slab_caches) { | |
2212 | struct kmem_cache *s = | |
2213 | container_of(h, struct kmem_cache, list); | |
2214 | ||
2215 | if (slab_unmergeable(s)) | |
2216 | continue; | |
2217 | ||
2218 | if (size > s->size) | |
2219 | continue; | |
2220 | ||
2221 | if (((flags | slub_debug) & SLUB_MERGE_SAME) != | |
2222 | (s->flags & SLUB_MERGE_SAME)) | |
2223 | continue; | |
2224 | /* | |
2225 | * Check if alignment is compatible. | |
2226 | * Courtesy of Adrian Drzewiecki | |
2227 | */ | |
2228 | if ((s->size & ~(align -1)) != s->size) | |
2229 | continue; | |
2230 | ||
2231 | if (s->size - size >= sizeof(void *)) | |
2232 | continue; | |
2233 | ||
2234 | return s; | |
2235 | } | |
2236 | return NULL; | |
2237 | } | |
2238 | ||
2239 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
2240 | size_t align, unsigned long flags, | |
2241 | void (*ctor)(void *, struct kmem_cache *, unsigned long), | |
2242 | void (*dtor)(void *, struct kmem_cache *, unsigned long)) | |
2243 | { | |
2244 | struct kmem_cache *s; | |
2245 | ||
2246 | down_write(&slub_lock); | |
2247 | s = find_mergeable(size, align, flags, dtor, ctor); | |
2248 | if (s) { | |
2249 | s->refcount++; | |
2250 | /* | |
2251 | * Adjust the object sizes so that we clear | |
2252 | * the complete object on kzalloc. | |
2253 | */ | |
2254 | s->objsize = max(s->objsize, (int)size); | |
2255 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); | |
2256 | if (sysfs_slab_alias(s, name)) | |
2257 | goto err; | |
2258 | } else { | |
2259 | s = kmalloc(kmem_size, GFP_KERNEL); | |
2260 | if (s && kmem_cache_open(s, GFP_KERNEL, name, | |
2261 | size, align, flags, ctor, dtor)) { | |
2262 | if (sysfs_slab_add(s)) { | |
2263 | kfree(s); | |
2264 | goto err; | |
2265 | } | |
2266 | list_add(&s->list, &slab_caches); | |
2267 | } else | |
2268 | kfree(s); | |
2269 | } | |
2270 | up_write(&slub_lock); | |
2271 | return s; | |
2272 | ||
2273 | err: | |
2274 | up_write(&slub_lock); | |
2275 | if (flags & SLAB_PANIC) | |
2276 | panic("Cannot create slabcache %s\n", name); | |
2277 | else | |
2278 | s = NULL; | |
2279 | return s; | |
2280 | } | |
2281 | EXPORT_SYMBOL(kmem_cache_create); | |
2282 | ||
2283 | void *kmem_cache_zalloc(struct kmem_cache *s, gfp_t flags) | |
2284 | { | |
2285 | void *x; | |
2286 | ||
77c5e2d0 | 2287 | x = slab_alloc(s, flags, -1, __builtin_return_address(0)); |
81819f0f CL |
2288 | if (x) |
2289 | memset(x, 0, s->objsize); | |
2290 | return x; | |
2291 | } | |
2292 | EXPORT_SYMBOL(kmem_cache_zalloc); | |
2293 | ||
2294 | #ifdef CONFIG_SMP | |
2295 | static void for_all_slabs(void (*func)(struct kmem_cache *, int), int cpu) | |
2296 | { | |
2297 | struct list_head *h; | |
2298 | ||
2299 | down_read(&slub_lock); | |
2300 | list_for_each(h, &slab_caches) { | |
2301 | struct kmem_cache *s = | |
2302 | container_of(h, struct kmem_cache, list); | |
2303 | ||
2304 | func(s, cpu); | |
2305 | } | |
2306 | up_read(&slub_lock); | |
2307 | } | |
2308 | ||
2309 | /* | |
2310 | * Use the cpu notifier to insure that the slab are flushed | |
2311 | * when necessary. | |
2312 | */ | |
2313 | static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb, | |
2314 | unsigned long action, void *hcpu) | |
2315 | { | |
2316 | long cpu = (long)hcpu; | |
2317 | ||
2318 | switch (action) { | |
2319 | case CPU_UP_CANCELED: | |
2320 | case CPU_DEAD: | |
2321 | for_all_slabs(__flush_cpu_slab, cpu); | |
2322 | break; | |
2323 | default: | |
2324 | break; | |
2325 | } | |
2326 | return NOTIFY_OK; | |
2327 | } | |
2328 | ||
2329 | static struct notifier_block __cpuinitdata slab_notifier = | |
2330 | { &slab_cpuup_callback, NULL, 0 }; | |
2331 | ||
2332 | #endif | |
2333 | ||
2334 | /*************************************************************** | |
2335 | * Compatiblility definitions | |
2336 | **************************************************************/ | |
2337 | ||
2338 | int kmem_cache_shrink(struct kmem_cache *s) | |
2339 | { | |
2340 | flush_all(s); | |
2341 | return 0; | |
2342 | } | |
2343 | EXPORT_SYMBOL(kmem_cache_shrink); | |
2344 | ||
2345 | #ifdef CONFIG_NUMA | |
2346 | ||
2347 | /***************************************************************** | |
2348 | * Generic reaper used to support the page allocator | |
2349 | * (the cpu slabs are reaped by a per slab workqueue). | |
2350 | * | |
2351 | * Maybe move this to the page allocator? | |
2352 | ****************************************************************/ | |
2353 | ||
2354 | static DEFINE_PER_CPU(unsigned long, reap_node); | |
2355 | ||
2356 | static void init_reap_node(int cpu) | |
2357 | { | |
2358 | int node; | |
2359 | ||
2360 | node = next_node(cpu_to_node(cpu), node_online_map); | |
2361 | if (node == MAX_NUMNODES) | |
2362 | node = first_node(node_online_map); | |
2363 | ||
2364 | __get_cpu_var(reap_node) = node; | |
2365 | } | |
2366 | ||
2367 | static void next_reap_node(void) | |
2368 | { | |
2369 | int node = __get_cpu_var(reap_node); | |
2370 | ||
2371 | /* | |
2372 | * Also drain per cpu pages on remote zones | |
2373 | */ | |
2374 | if (node != numa_node_id()) | |
2375 | drain_node_pages(node); | |
2376 | ||
2377 | node = next_node(node, node_online_map); | |
2378 | if (unlikely(node >= MAX_NUMNODES)) | |
2379 | node = first_node(node_online_map); | |
2380 | __get_cpu_var(reap_node) = node; | |
2381 | } | |
2382 | #else | |
2383 | #define init_reap_node(cpu) do { } while (0) | |
2384 | #define next_reap_node(void) do { } while (0) | |
2385 | #endif | |
2386 | ||
2387 | #define REAPTIMEOUT_CPUC (2*HZ) | |
2388 | ||
2389 | #ifdef CONFIG_SMP | |
2390 | static DEFINE_PER_CPU(struct delayed_work, reap_work); | |
2391 | ||
2392 | static void cache_reap(struct work_struct *unused) | |
2393 | { | |
2394 | next_reap_node(); | |
2395 | refresh_cpu_vm_stats(smp_processor_id()); | |
2396 | schedule_delayed_work(&__get_cpu_var(reap_work), | |
2397 | REAPTIMEOUT_CPUC); | |
2398 | } | |
2399 | ||
2400 | static void __devinit start_cpu_timer(int cpu) | |
2401 | { | |
2402 | struct delayed_work *reap_work = &per_cpu(reap_work, cpu); | |
2403 | ||
2404 | /* | |
2405 | * When this gets called from do_initcalls via cpucache_init(), | |
2406 | * init_workqueues() has already run, so keventd will be setup | |
2407 | * at that time. | |
2408 | */ | |
2409 | if (keventd_up() && reap_work->work.func == NULL) { | |
2410 | init_reap_node(cpu); | |
2411 | INIT_DELAYED_WORK(reap_work, cache_reap); | |
2412 | schedule_delayed_work_on(cpu, reap_work, HZ + 3 * cpu); | |
2413 | } | |
2414 | } | |
2415 | ||
2416 | static int __init cpucache_init(void) | |
2417 | { | |
2418 | int cpu; | |
2419 | ||
2420 | /* | |
2421 | * Register the timers that drain pcp pages and update vm statistics | |
2422 | */ | |
2423 | for_each_online_cpu(cpu) | |
2424 | start_cpu_timer(cpu); | |
2425 | return 0; | |
2426 | } | |
2427 | __initcall(cpucache_init); | |
2428 | #endif | |
2429 | ||
2430 | #ifdef SLUB_RESILIENCY_TEST | |
2431 | static unsigned long validate_slab_cache(struct kmem_cache *s); | |
2432 | ||
2433 | static void resiliency_test(void) | |
2434 | { | |
2435 | u8 *p; | |
2436 | ||
2437 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
2438 | printk(KERN_ERR "-----------------------\n"); | |
2439 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
2440 | ||
2441 | p = kzalloc(16, GFP_KERNEL); | |
2442 | p[16] = 0x12; | |
2443 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
2444 | " 0x12->0x%p\n\n", p + 16); | |
2445 | ||
2446 | validate_slab_cache(kmalloc_caches + 4); | |
2447 | ||
2448 | /* Hmmm... The next two are dangerous */ | |
2449 | p = kzalloc(32, GFP_KERNEL); | |
2450 | p[32 + sizeof(void *)] = 0x34; | |
2451 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
2452 | " 0x34 -> -0x%p\n", p); | |
2453 | printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n"); | |
2454 | ||
2455 | validate_slab_cache(kmalloc_caches + 5); | |
2456 | p = kzalloc(64, GFP_KERNEL); | |
2457 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
2458 | *p = 0x56; | |
2459 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
2460 | p); | |
2461 | printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n"); | |
2462 | validate_slab_cache(kmalloc_caches + 6); | |
2463 | ||
2464 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
2465 | p = kzalloc(128, GFP_KERNEL); | |
2466 | kfree(p); | |
2467 | *p = 0x78; | |
2468 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
2469 | validate_slab_cache(kmalloc_caches + 7); | |
2470 | ||
2471 | p = kzalloc(256, GFP_KERNEL); | |
2472 | kfree(p); | |
2473 | p[50] = 0x9a; | |
2474 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); | |
2475 | validate_slab_cache(kmalloc_caches + 8); | |
2476 | ||
2477 | p = kzalloc(512, GFP_KERNEL); | |
2478 | kfree(p); | |
2479 | p[512] = 0xab; | |
2480 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
2481 | validate_slab_cache(kmalloc_caches + 9); | |
2482 | } | |
2483 | #else | |
2484 | static void resiliency_test(void) {}; | |
2485 | #endif | |
2486 | ||
2487 | /* | |
2488 | * These are not as efficient as kmalloc for the non debug case. | |
2489 | * We do not have the page struct available so we have to touch one | |
2490 | * cacheline in struct kmem_cache to check slab flags. | |
2491 | */ | |
2492 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller) | |
2493 | { | |
2494 | struct kmem_cache *s = get_slab(size, gfpflags); | |
81819f0f CL |
2495 | |
2496 | if (!s) | |
2497 | return NULL; | |
2498 | ||
77c5e2d0 | 2499 | return slab_alloc(s, gfpflags, -1, caller); |
81819f0f CL |
2500 | } |
2501 | ||
2502 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, | |
2503 | int node, void *caller) | |
2504 | { | |
2505 | struct kmem_cache *s = get_slab(size, gfpflags); | |
81819f0f CL |
2506 | |
2507 | if (!s) | |
2508 | return NULL; | |
2509 | ||
77c5e2d0 | 2510 | return slab_alloc(s, gfpflags, node, caller); |
81819f0f CL |
2511 | } |
2512 | ||
2513 | #ifdef CONFIG_SYSFS | |
2514 | ||
2515 | static unsigned long count_partial(struct kmem_cache_node *n) | |
2516 | { | |
2517 | unsigned long flags; | |
2518 | unsigned long x = 0; | |
2519 | struct page *page; | |
2520 | ||
2521 | spin_lock_irqsave(&n->list_lock, flags); | |
2522 | list_for_each_entry(page, &n->partial, lru) | |
2523 | x += page->inuse; | |
2524 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2525 | return x; | |
2526 | } | |
2527 | ||
2528 | enum slab_stat_type { | |
2529 | SL_FULL, | |
2530 | SL_PARTIAL, | |
2531 | SL_CPU, | |
2532 | SL_OBJECTS | |
2533 | }; | |
2534 | ||
2535 | #define SO_FULL (1 << SL_FULL) | |
2536 | #define SO_PARTIAL (1 << SL_PARTIAL) | |
2537 | #define SO_CPU (1 << SL_CPU) | |
2538 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
2539 | ||
2540 | static unsigned long slab_objects(struct kmem_cache *s, | |
2541 | char *buf, unsigned long flags) | |
2542 | { | |
2543 | unsigned long total = 0; | |
2544 | int cpu; | |
2545 | int node; | |
2546 | int x; | |
2547 | unsigned long *nodes; | |
2548 | unsigned long *per_cpu; | |
2549 | ||
2550 | nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); | |
2551 | per_cpu = nodes + nr_node_ids; | |
2552 | ||
2553 | for_each_possible_cpu(cpu) { | |
2554 | struct page *page = s->cpu_slab[cpu]; | |
2555 | int node; | |
2556 | ||
2557 | if (page) { | |
2558 | node = page_to_nid(page); | |
2559 | if (flags & SO_CPU) { | |
2560 | int x = 0; | |
2561 | ||
2562 | if (flags & SO_OBJECTS) | |
2563 | x = page->inuse; | |
2564 | else | |
2565 | x = 1; | |
2566 | total += x; | |
2567 | nodes[node] += x; | |
2568 | } | |
2569 | per_cpu[node]++; | |
2570 | } | |
2571 | } | |
2572 | ||
2573 | for_each_online_node(node) { | |
2574 | struct kmem_cache_node *n = get_node(s, node); | |
2575 | ||
2576 | if (flags & SO_PARTIAL) { | |
2577 | if (flags & SO_OBJECTS) | |
2578 | x = count_partial(n); | |
2579 | else | |
2580 | x = n->nr_partial; | |
2581 | total += x; | |
2582 | nodes[node] += x; | |
2583 | } | |
2584 | ||
2585 | if (flags & SO_FULL) { | |
2586 | int full_slabs = atomic_read(&n->nr_slabs) | |
2587 | - per_cpu[node] | |
2588 | - n->nr_partial; | |
2589 | ||
2590 | if (flags & SO_OBJECTS) | |
2591 | x = full_slabs * s->objects; | |
2592 | else | |
2593 | x = full_slabs; | |
2594 | total += x; | |
2595 | nodes[node] += x; | |
2596 | } | |
2597 | } | |
2598 | ||
2599 | x = sprintf(buf, "%lu", total); | |
2600 | #ifdef CONFIG_NUMA | |
2601 | for_each_online_node(node) | |
2602 | if (nodes[node]) | |
2603 | x += sprintf(buf + x, " N%d=%lu", | |
2604 | node, nodes[node]); | |
2605 | #endif | |
2606 | kfree(nodes); | |
2607 | return x + sprintf(buf + x, "\n"); | |
2608 | } | |
2609 | ||
2610 | static int any_slab_objects(struct kmem_cache *s) | |
2611 | { | |
2612 | int node; | |
2613 | int cpu; | |
2614 | ||
2615 | for_each_possible_cpu(cpu) | |
2616 | if (s->cpu_slab[cpu]) | |
2617 | return 1; | |
2618 | ||
2619 | for_each_node(node) { | |
2620 | struct kmem_cache_node *n = get_node(s, node); | |
2621 | ||
2622 | if (n->nr_partial || atomic_read(&n->nr_slabs)) | |
2623 | return 1; | |
2624 | } | |
2625 | return 0; | |
2626 | } | |
2627 | ||
2628 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
2629 | #define to_slab(n) container_of(n, struct kmem_cache, kobj); | |
2630 | ||
2631 | struct slab_attribute { | |
2632 | struct attribute attr; | |
2633 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
2634 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
2635 | }; | |
2636 | ||
2637 | #define SLAB_ATTR_RO(_name) \ | |
2638 | static struct slab_attribute _name##_attr = __ATTR_RO(_name) | |
2639 | ||
2640 | #define SLAB_ATTR(_name) \ | |
2641 | static struct slab_attribute _name##_attr = \ | |
2642 | __ATTR(_name, 0644, _name##_show, _name##_store) | |
2643 | ||
2644 | ||
2645 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) | |
2646 | { | |
2647 | return sprintf(buf, "%d\n", s->size); | |
2648 | } | |
2649 | SLAB_ATTR_RO(slab_size); | |
2650 | ||
2651 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
2652 | { | |
2653 | return sprintf(buf, "%d\n", s->align); | |
2654 | } | |
2655 | SLAB_ATTR_RO(align); | |
2656 | ||
2657 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
2658 | { | |
2659 | return sprintf(buf, "%d\n", s->objsize); | |
2660 | } | |
2661 | SLAB_ATTR_RO(object_size); | |
2662 | ||
2663 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
2664 | { | |
2665 | return sprintf(buf, "%d\n", s->objects); | |
2666 | } | |
2667 | SLAB_ATTR_RO(objs_per_slab); | |
2668 | ||
2669 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
2670 | { | |
2671 | return sprintf(buf, "%d\n", s->order); | |
2672 | } | |
2673 | SLAB_ATTR_RO(order); | |
2674 | ||
2675 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) | |
2676 | { | |
2677 | if (s->ctor) { | |
2678 | int n = sprint_symbol(buf, (unsigned long)s->ctor); | |
2679 | ||
2680 | return n + sprintf(buf + n, "\n"); | |
2681 | } | |
2682 | return 0; | |
2683 | } | |
2684 | SLAB_ATTR_RO(ctor); | |
2685 | ||
2686 | static ssize_t dtor_show(struct kmem_cache *s, char *buf) | |
2687 | { | |
2688 | if (s->dtor) { | |
2689 | int n = sprint_symbol(buf, (unsigned long)s->dtor); | |
2690 | ||
2691 | return n + sprintf(buf + n, "\n"); | |
2692 | } | |
2693 | return 0; | |
2694 | } | |
2695 | SLAB_ATTR_RO(dtor); | |
2696 | ||
2697 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) | |
2698 | { | |
2699 | return sprintf(buf, "%d\n", s->refcount - 1); | |
2700 | } | |
2701 | SLAB_ATTR_RO(aliases); | |
2702 | ||
2703 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | |
2704 | { | |
2705 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU); | |
2706 | } | |
2707 | SLAB_ATTR_RO(slabs); | |
2708 | ||
2709 | static ssize_t partial_show(struct kmem_cache *s, char *buf) | |
2710 | { | |
2711 | return slab_objects(s, buf, SO_PARTIAL); | |
2712 | } | |
2713 | SLAB_ATTR_RO(partial); | |
2714 | ||
2715 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
2716 | { | |
2717 | return slab_objects(s, buf, SO_CPU); | |
2718 | } | |
2719 | SLAB_ATTR_RO(cpu_slabs); | |
2720 | ||
2721 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
2722 | { | |
2723 | return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU|SO_OBJECTS); | |
2724 | } | |
2725 | SLAB_ATTR_RO(objects); | |
2726 | ||
2727 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) | |
2728 | { | |
2729 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
2730 | } | |
2731 | ||
2732 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
2733 | const char *buf, size_t length) | |
2734 | { | |
2735 | s->flags &= ~SLAB_DEBUG_FREE; | |
2736 | if (buf[0] == '1') | |
2737 | s->flags |= SLAB_DEBUG_FREE; | |
2738 | return length; | |
2739 | } | |
2740 | SLAB_ATTR(sanity_checks); | |
2741 | ||
2742 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
2743 | { | |
2744 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
2745 | } | |
2746 | ||
2747 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
2748 | size_t length) | |
2749 | { | |
2750 | s->flags &= ~SLAB_TRACE; | |
2751 | if (buf[0] == '1') | |
2752 | s->flags |= SLAB_TRACE; | |
2753 | return length; | |
2754 | } | |
2755 | SLAB_ATTR(trace); | |
2756 | ||
2757 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) | |
2758 | { | |
2759 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
2760 | } | |
2761 | ||
2762 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
2763 | const char *buf, size_t length) | |
2764 | { | |
2765 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
2766 | if (buf[0] == '1') | |
2767 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
2768 | return length; | |
2769 | } | |
2770 | SLAB_ATTR(reclaim_account); | |
2771 | ||
2772 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
2773 | { | |
2774 | return sprintf(buf, "%d\n", !!(s->flags & | |
2775 | (SLAB_HWCACHE_ALIGN|SLAB_MUST_HWCACHE_ALIGN))); | |
2776 | } | |
2777 | SLAB_ATTR_RO(hwcache_align); | |
2778 | ||
2779 | #ifdef CONFIG_ZONE_DMA | |
2780 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
2781 | { | |
2782 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
2783 | } | |
2784 | SLAB_ATTR_RO(cache_dma); | |
2785 | #endif | |
2786 | ||
2787 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
2788 | { | |
2789 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
2790 | } | |
2791 | SLAB_ATTR_RO(destroy_by_rcu); | |
2792 | ||
2793 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) | |
2794 | { | |
2795 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
2796 | } | |
2797 | ||
2798 | static ssize_t red_zone_store(struct kmem_cache *s, | |
2799 | const char *buf, size_t length) | |
2800 | { | |
2801 | if (any_slab_objects(s)) | |
2802 | return -EBUSY; | |
2803 | ||
2804 | s->flags &= ~SLAB_RED_ZONE; | |
2805 | if (buf[0] == '1') | |
2806 | s->flags |= SLAB_RED_ZONE; | |
2807 | calculate_sizes(s); | |
2808 | return length; | |
2809 | } | |
2810 | SLAB_ATTR(red_zone); | |
2811 | ||
2812 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
2813 | { | |
2814 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
2815 | } | |
2816 | ||
2817 | static ssize_t poison_store(struct kmem_cache *s, | |
2818 | const char *buf, size_t length) | |
2819 | { | |
2820 | if (any_slab_objects(s)) | |
2821 | return -EBUSY; | |
2822 | ||
2823 | s->flags &= ~SLAB_POISON; | |
2824 | if (buf[0] == '1') | |
2825 | s->flags |= SLAB_POISON; | |
2826 | calculate_sizes(s); | |
2827 | return length; | |
2828 | } | |
2829 | SLAB_ATTR(poison); | |
2830 | ||
2831 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
2832 | { | |
2833 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
2834 | } | |
2835 | ||
2836 | static ssize_t store_user_store(struct kmem_cache *s, | |
2837 | const char *buf, size_t length) | |
2838 | { | |
2839 | if (any_slab_objects(s)) | |
2840 | return -EBUSY; | |
2841 | ||
2842 | s->flags &= ~SLAB_STORE_USER; | |
2843 | if (buf[0] == '1') | |
2844 | s->flags |= SLAB_STORE_USER; | |
2845 | calculate_sizes(s); | |
2846 | return length; | |
2847 | } | |
2848 | SLAB_ATTR(store_user); | |
2849 | ||
2850 | #ifdef CONFIG_NUMA | |
2851 | static ssize_t defrag_ratio_show(struct kmem_cache *s, char *buf) | |
2852 | { | |
2853 | return sprintf(buf, "%d\n", s->defrag_ratio / 10); | |
2854 | } | |
2855 | ||
2856 | static ssize_t defrag_ratio_store(struct kmem_cache *s, | |
2857 | const char *buf, size_t length) | |
2858 | { | |
2859 | int n = simple_strtoul(buf, NULL, 10); | |
2860 | ||
2861 | if (n < 100) | |
2862 | s->defrag_ratio = n * 10; | |
2863 | return length; | |
2864 | } | |
2865 | SLAB_ATTR(defrag_ratio); | |
2866 | #endif | |
2867 | ||
2868 | static struct attribute * slab_attrs[] = { | |
2869 | &slab_size_attr.attr, | |
2870 | &object_size_attr.attr, | |
2871 | &objs_per_slab_attr.attr, | |
2872 | &order_attr.attr, | |
2873 | &objects_attr.attr, | |
2874 | &slabs_attr.attr, | |
2875 | &partial_attr.attr, | |
2876 | &cpu_slabs_attr.attr, | |
2877 | &ctor_attr.attr, | |
2878 | &dtor_attr.attr, | |
2879 | &aliases_attr.attr, | |
2880 | &align_attr.attr, | |
2881 | &sanity_checks_attr.attr, | |
2882 | &trace_attr.attr, | |
2883 | &hwcache_align_attr.attr, | |
2884 | &reclaim_account_attr.attr, | |
2885 | &destroy_by_rcu_attr.attr, | |
2886 | &red_zone_attr.attr, | |
2887 | &poison_attr.attr, | |
2888 | &store_user_attr.attr, | |
2889 | #ifdef CONFIG_ZONE_DMA | |
2890 | &cache_dma_attr.attr, | |
2891 | #endif | |
2892 | #ifdef CONFIG_NUMA | |
2893 | &defrag_ratio_attr.attr, | |
2894 | #endif | |
2895 | NULL | |
2896 | }; | |
2897 | ||
2898 | static struct attribute_group slab_attr_group = { | |
2899 | .attrs = slab_attrs, | |
2900 | }; | |
2901 | ||
2902 | static ssize_t slab_attr_show(struct kobject *kobj, | |
2903 | struct attribute *attr, | |
2904 | char *buf) | |
2905 | { | |
2906 | struct slab_attribute *attribute; | |
2907 | struct kmem_cache *s; | |
2908 | int err; | |
2909 | ||
2910 | attribute = to_slab_attr(attr); | |
2911 | s = to_slab(kobj); | |
2912 | ||
2913 | if (!attribute->show) | |
2914 | return -EIO; | |
2915 | ||
2916 | err = attribute->show(s, buf); | |
2917 | ||
2918 | return err; | |
2919 | } | |
2920 | ||
2921 | static ssize_t slab_attr_store(struct kobject *kobj, | |
2922 | struct attribute *attr, | |
2923 | const char *buf, size_t len) | |
2924 | { | |
2925 | struct slab_attribute *attribute; | |
2926 | struct kmem_cache *s; | |
2927 | int err; | |
2928 | ||
2929 | attribute = to_slab_attr(attr); | |
2930 | s = to_slab(kobj); | |
2931 | ||
2932 | if (!attribute->store) | |
2933 | return -EIO; | |
2934 | ||
2935 | err = attribute->store(s, buf, len); | |
2936 | ||
2937 | return err; | |
2938 | } | |
2939 | ||
2940 | static struct sysfs_ops slab_sysfs_ops = { | |
2941 | .show = slab_attr_show, | |
2942 | .store = slab_attr_store, | |
2943 | }; | |
2944 | ||
2945 | static struct kobj_type slab_ktype = { | |
2946 | .sysfs_ops = &slab_sysfs_ops, | |
2947 | }; | |
2948 | ||
2949 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
2950 | { | |
2951 | struct kobj_type *ktype = get_ktype(kobj); | |
2952 | ||
2953 | if (ktype == &slab_ktype) | |
2954 | return 1; | |
2955 | return 0; | |
2956 | } | |
2957 | ||
2958 | static struct kset_uevent_ops slab_uevent_ops = { | |
2959 | .filter = uevent_filter, | |
2960 | }; | |
2961 | ||
2962 | decl_subsys(slab, &slab_ktype, &slab_uevent_ops); | |
2963 | ||
2964 | #define ID_STR_LENGTH 64 | |
2965 | ||
2966 | /* Create a unique string id for a slab cache: | |
2967 | * format | |
2968 | * :[flags-]size:[memory address of kmemcache] | |
2969 | */ | |
2970 | static char *create_unique_id(struct kmem_cache *s) | |
2971 | { | |
2972 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
2973 | char *p = name; | |
2974 | ||
2975 | BUG_ON(!name); | |
2976 | ||
2977 | *p++ = ':'; | |
2978 | /* | |
2979 | * First flags affecting slabcache operations. We will only | |
2980 | * get here for aliasable slabs so we do not need to support | |
2981 | * too many flags. The flags here must cover all flags that | |
2982 | * are matched during merging to guarantee that the id is | |
2983 | * unique. | |
2984 | */ | |
2985 | if (s->flags & SLAB_CACHE_DMA) | |
2986 | *p++ = 'd'; | |
2987 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
2988 | *p++ = 'a'; | |
2989 | if (s->flags & SLAB_DEBUG_FREE) | |
2990 | *p++ = 'F'; | |
2991 | if (p != name + 1) | |
2992 | *p++ = '-'; | |
2993 | p += sprintf(p, "%07d", s->size); | |
2994 | BUG_ON(p > name + ID_STR_LENGTH - 1); | |
2995 | return name; | |
2996 | } | |
2997 | ||
2998 | static int sysfs_slab_add(struct kmem_cache *s) | |
2999 | { | |
3000 | int err; | |
3001 | const char *name; | |
3002 | int unmergeable; | |
3003 | ||
3004 | if (slab_state < SYSFS) | |
3005 | /* Defer until later */ | |
3006 | return 0; | |
3007 | ||
3008 | unmergeable = slab_unmergeable(s); | |
3009 | if (unmergeable) { | |
3010 | /* | |
3011 | * Slabcache can never be merged so we can use the name proper. | |
3012 | * This is typically the case for debug situations. In that | |
3013 | * case we can catch duplicate names easily. | |
3014 | */ | |
3015 | sysfs_remove_link(&slab_subsys.kset.kobj, s->name); | |
3016 | name = s->name; | |
3017 | } else { | |
3018 | /* | |
3019 | * Create a unique name for the slab as a target | |
3020 | * for the symlinks. | |
3021 | */ | |
3022 | name = create_unique_id(s); | |
3023 | } | |
3024 | ||
3025 | kobj_set_kset_s(s, slab_subsys); | |
3026 | kobject_set_name(&s->kobj, name); | |
3027 | kobject_init(&s->kobj); | |
3028 | err = kobject_add(&s->kobj); | |
3029 | if (err) | |
3030 | return err; | |
3031 | ||
3032 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
3033 | if (err) | |
3034 | return err; | |
3035 | kobject_uevent(&s->kobj, KOBJ_ADD); | |
3036 | if (!unmergeable) { | |
3037 | /* Setup first alias */ | |
3038 | sysfs_slab_alias(s, s->name); | |
3039 | kfree(name); | |
3040 | } | |
3041 | return 0; | |
3042 | } | |
3043 | ||
3044 | static void sysfs_slab_remove(struct kmem_cache *s) | |
3045 | { | |
3046 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
3047 | kobject_del(&s->kobj); | |
3048 | } | |
3049 | ||
3050 | /* | |
3051 | * Need to buffer aliases during bootup until sysfs becomes | |
3052 | * available lest we loose that information. | |
3053 | */ | |
3054 | struct saved_alias { | |
3055 | struct kmem_cache *s; | |
3056 | const char *name; | |
3057 | struct saved_alias *next; | |
3058 | }; | |
3059 | ||
3060 | struct saved_alias *alias_list; | |
3061 | ||
3062 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
3063 | { | |
3064 | struct saved_alias *al; | |
3065 | ||
3066 | if (slab_state == SYSFS) { | |
3067 | /* | |
3068 | * If we have a leftover link then remove it. | |
3069 | */ | |
3070 | sysfs_remove_link(&slab_subsys.kset.kobj, name); | |
3071 | return sysfs_create_link(&slab_subsys.kset.kobj, | |
3072 | &s->kobj, name); | |
3073 | } | |
3074 | ||
3075 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
3076 | if (!al) | |
3077 | return -ENOMEM; | |
3078 | ||
3079 | al->s = s; | |
3080 | al->name = name; | |
3081 | al->next = alias_list; | |
3082 | alias_list = al; | |
3083 | return 0; | |
3084 | } | |
3085 | ||
3086 | static int __init slab_sysfs_init(void) | |
3087 | { | |
3088 | int err; | |
3089 | ||
3090 | err = subsystem_register(&slab_subsys); | |
3091 | if (err) { | |
3092 | printk(KERN_ERR "Cannot register slab subsystem.\n"); | |
3093 | return -ENOSYS; | |
3094 | } | |
3095 | ||
3096 | finish_bootstrap(); | |
3097 | ||
3098 | while (alias_list) { | |
3099 | struct saved_alias *al = alias_list; | |
3100 | ||
3101 | alias_list = alias_list->next; | |
3102 | err = sysfs_slab_alias(al->s, al->name); | |
3103 | BUG_ON(err); | |
3104 | kfree(al); | |
3105 | } | |
3106 | ||
3107 | resiliency_test(); | |
3108 | return 0; | |
3109 | } | |
3110 | ||
3111 | __initcall(slab_sysfs_init); | |
3112 | #else | |
3113 | __initcall(finish_bootstrap); | |
3114 | #endif |