Root/mm/slub.c

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
9 */
10
11#include <linux/mm.h>
12#include <linux/swap.h> /* struct reclaim_state */
13#include <linux/module.h>
14#include <linux/bit_spinlock.h>
15#include <linux/interrupt.h>
16#include <linux/bitops.h>
17#include <linux/slab.h>
18#include <linux/proc_fs.h>
19#include <linux/seq_file.h>
20#include <linux/kmemtrace.h>
21#include <linux/kmemcheck.h>
22#include <linux/cpu.h>
23#include <linux/cpuset.h>
24#include <linux/mempolicy.h>
25#include <linux/ctype.h>
26#include <linux/debugobjects.h>
27#include <linux/kallsyms.h>
28#include <linux/memory.h>
29#include <linux/math64.h>
30#include <linux/fault-inject.h>
31
32/*
33 * Lock order:
34 * 1. slab_lock(page)
35 * 2. slab->list_lock
36 *
37 * The slab_lock protects operations on the object of a particular
38 * slab and its metadata in the page struct. If the slab lock
39 * has been taken then no allocations nor frees can be performed
40 * on the objects in the slab nor can the slab be added or removed
41 * from the partial or full lists since this would mean modifying
42 * the page_struct of the slab.
43 *
44 * The list_lock protects the partial and full list on each node and
45 * the partial slab counter. If taken then no new slabs may be added or
46 * removed from the lists nor make the number of partial slabs be modified.
47 * (Note that the total number of slabs is an atomic value that may be
48 * modified without taking the list lock).
49 *
50 * The list_lock is a centralized lock and thus we avoid taking it as
51 * much as possible. As long as SLUB does not have to handle partial
52 * slabs, operations can continue without any centralized lock. F.e.
53 * allocating a long series of objects that fill up slabs does not require
54 * the list lock.
55 *
56 * The lock order is sometimes inverted when we are trying to get a slab
57 * off a list. We take the list_lock and then look for a page on the list
58 * to use. While we do that objects in the slabs may be freed. We can
59 * only operate on the slab if we have also taken the slab_lock. So we use
60 * a slab_trylock() on the slab. If trylock was successful then no frees
61 * can occur anymore and we can use the slab for allocations etc. If the
62 * slab_trylock() does not succeed then frees are in progress in the slab and
63 * we must stay away from it for a while since we may cause a bouncing
64 * cacheline if we try to acquire the lock. So go onto the next slab.
65 * If all pages are busy then we may allocate a new slab instead of reusing
66 * a partial slab. A new slab has noone operating on it and thus there is
67 * no danger of cacheline contention.
68 *
69 * Interrupts are disabled during allocation and deallocation in order to
70 * make the slab allocator safe to use in the context of an irq. In addition
71 * interrupts are disabled to ensure that the processor does not change
72 * while handling per_cpu slabs, due to kernel preemption.
73 *
74 * SLUB assigns one slab for allocation to each processor.
75 * Allocations only occur from these slabs called cpu slabs.
76 *
77 * Slabs with free elements are kept on a partial list and during regular
78 * operations no list for full slabs is used. If an object in a full slab is
79 * freed then the slab will show up again on the partial lists.
80 * We track full slabs for debugging purposes though because otherwise we
81 * cannot scan all objects.
82 *
83 * Slabs are freed when they become empty. Teardown and setup is
84 * minimal so we rely on the page allocators per cpu caches for
85 * fast frees and allocs.
86 *
87 * Overloading of page flags that are otherwise used for LRU management.
88 *
89 * PageActive The slab is frozen and exempt from list processing.
90 * This means that the slab is dedicated to a purpose
91 * such as satisfying allocations for a specific
92 * processor. Objects may be freed in the slab while
93 * it is frozen but slab_free will then skip the usual
94 * list operations. It is up to the processor holding
95 * the slab to integrate the slab into the slab lists
96 * when the slab is no longer needed.
97 *
98 * One use of this flag is to mark slabs that are
99 * used for allocations. Then such a slab becomes a cpu
100 * slab. The cpu slab may be equipped with an additional
101 * freelist that allows lockless access to
102 * free objects in addition to the regular freelist
103 * that requires the slab lock.
104 *
105 * PageError Slab requires special handling due to debug
106 * options set. This moves slab handling out of
107 * the fast path and disables lockless freelists.
108 */
109
110#ifdef CONFIG_SLUB_DEBUG
111#define SLABDEBUG 1
112#else
113#define SLABDEBUG 0
114#endif
115
116/*
117 * Issues still to be resolved:
118 *
119 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
120 *
121 * - Variable sizing of the per node arrays
122 */
123
124/* Enable to test recovery from slab corruption on boot */
125#undef SLUB_RESILIENCY_TEST
126
127/*
128 * Mininum number of partial slabs. These will be left on the partial
129 * lists even if they are empty. kmem_cache_shrink may reclaim them.
130 */
131#define MIN_PARTIAL 5
132
133/*
134 * Maximum number of desirable partial slabs.
135 * The existence of more partial slabs makes kmem_cache_shrink
136 * sort the partial list by the number of objects in the.
137 */
138#define MAX_PARTIAL 10
139
140#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
141                SLAB_POISON | SLAB_STORE_USER)
142
143/*
144 * Debugging flags that require metadata to be stored in the slab. These get
145 * disabled when slub_debug=O is used and a cache's min order increases with
146 * metadata.
147 */
148#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
149
150/*
151 * Set of flags that will prevent slab merging
152 */
153#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
154        SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
155        SLAB_FAILSLAB)
156
157#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
158        SLAB_CACHE_DMA | SLAB_NOTRACK)
159
160#define OO_SHIFT 16
161#define OO_MASK ((1 << OO_SHIFT) - 1)
162#define MAX_OBJS_PER_PAGE 65535 /* since page.objects is u16 */
163
164/* Internal SLUB flags */
165#define __OBJECT_POISON 0x80000000 /* Poison object */
166#define __SYSFS_ADD_DEFERRED 0x40000000 /* Not yet visible via sysfs */
167
168static int kmem_size = sizeof(struct kmem_cache);
169
170#ifdef CONFIG_SMP
171static struct notifier_block slab_notifier;
172#endif
173
174static enum {
175    DOWN, /* No slab functionality available */
176    PARTIAL, /* kmem_cache_open() works but kmalloc does not */
177    UP, /* Everything works but does not show up in sysfs */
178    SYSFS /* Sysfs up */
179} slab_state = DOWN;
180
181/* A list of all slab caches on the system */
182static DECLARE_RWSEM(slub_lock);
183static LIST_HEAD(slab_caches);
184
185/*
186 * Tracking user of a slab.
187 */
188struct track {
189    unsigned long addr; /* Called from address */
190    int cpu; /* Was running on cpu */
191    int pid; /* Pid context */
192    unsigned long when; /* When did the operation occur */
193};
194
195enum track_item { TRACK_ALLOC, TRACK_FREE };
196
197#ifdef CONFIG_SLUB_DEBUG
198static int sysfs_slab_add(struct kmem_cache *);
199static int sysfs_slab_alias(struct kmem_cache *, const char *);
200static void sysfs_slab_remove(struct kmem_cache *);
201
202#else
203static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
204static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
205                            { return 0; }
206static inline void sysfs_slab_remove(struct kmem_cache *s)
207{
208    kfree(s);
209}
210
211#endif
212
213static inline void stat(struct kmem_cache *s, enum stat_item si)
214{
215#ifdef CONFIG_SLUB_STATS
216    __this_cpu_inc(s->cpu_slab->stat[si]);
217#endif
218}
219
220/********************************************************************
221 * Core slab cache functions
222 *******************************************************************/
223
224int slab_is_available(void)
225{
226    return slab_state >= UP;
227}
228
229static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
230{
231#ifdef CONFIG_NUMA
232    return s->node[node];
233#else
234    return &s->local_node;
235#endif
236}
237
238/* Verify that a pointer has an address that is valid within a slab page */
239static inline int check_valid_pointer(struct kmem_cache *s,
240                struct page *page, const void *object)
241{
242    void *base;
243
244    if (!object)
245        return 1;
246
247    base = page_address(page);
248    if (object < base || object >= base + page->objects * s->size ||
249        (object - base) % s->size) {
250        return 0;
251    }
252
253    return 1;
254}
255
256static inline void *get_freepointer(struct kmem_cache *s, void *object)
257{
258    return *(void **)(object + s->offset);
259}
260
261static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
262{
263    *(void **)(object + s->offset) = fp;
264}
265
266/* Loop over all objects in a slab */
267#define for_each_object(__p, __s, __addr, __objects) \
268    for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
269            __p += (__s)->size)
270
271/* Scan freelist */
272#define for_each_free_object(__p, __s, __free) \
273    for (__p = (__free); __p; __p = get_freepointer((__s), __p))
274
275/* Determine object index from a given position */
276static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
277{
278    return (p - addr) / s->size;
279}
280
281static inline struct kmem_cache_order_objects oo_make(int order,
282                        unsigned long size)
283{
284    struct kmem_cache_order_objects x = {
285        (order << OO_SHIFT) + (PAGE_SIZE << order) / size
286    };
287
288    return x;
289}
290
291static inline int oo_order(struct kmem_cache_order_objects x)
292{
293    return x.x >> OO_SHIFT;
294}
295
296static inline int oo_objects(struct kmem_cache_order_objects x)
297{
298    return x.x & OO_MASK;
299}
300
301#ifdef CONFIG_SLUB_DEBUG
302/*
303 * Debug settings:
304 */
305#ifdef CONFIG_SLUB_DEBUG_ON
306static int slub_debug = DEBUG_DEFAULT_FLAGS;
307#else
308static int slub_debug;
309#endif
310
311static char *slub_debug_slabs;
312static int disable_higher_order_debug;
313
314/*
315 * Object debugging
316 */
317static void print_section(char *text, u8 *addr, unsigned int length)
318{
319    int i, offset;
320    int newline = 1;
321    char ascii[17];
322
323    ascii[16] = 0;
324
325    for (i = 0; i < length; i++) {
326        if (newline) {
327            printk(KERN_ERR "%8s 0x%p: ", text, addr + i);
328            newline = 0;
329        }
330        printk(KERN_CONT " %02x", addr[i]);
331        offset = i % 16;
332        ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
333        if (offset == 15) {
334            printk(KERN_CONT " %s\n", ascii);
335            newline = 1;
336        }
337    }
338    if (!newline) {
339        i %= 16;
340        while (i < 16) {
341            printk(KERN_CONT " ");
342            ascii[i] = ' ';
343            i++;
344        }
345        printk(KERN_CONT " %s\n", ascii);
346    }
347}
348
349static struct track *get_track(struct kmem_cache *s, void *object,
350    enum track_item alloc)
351{
352    struct track *p;
353
354    if (s->offset)
355        p = object + s->offset + sizeof(void *);
356    else
357        p = object + s->inuse;
358
359    return p + alloc;
360}
361
362static void set_track(struct kmem_cache *s, void *object,
363            enum track_item alloc, unsigned long addr)
364{
365    struct track *p = get_track(s, object, alloc);
366
367    if (addr) {
368        p->addr = addr;
369        p->cpu = smp_processor_id();
370        p->pid = current->pid;
371        p->when = jiffies;
372    } else
373        memset(p, 0, sizeof(struct track));
374}
375
376static void init_tracking(struct kmem_cache *s, void *object)
377{
378    if (!(s->flags & SLAB_STORE_USER))
379        return;
380
381    set_track(s, object, TRACK_FREE, 0UL);
382    set_track(s, object, TRACK_ALLOC, 0UL);
383}
384
385static void print_track(const char *s, struct track *t)
386{
387    if (!t->addr)
388        return;
389
390    printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
391        s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
392}
393
394static void print_tracking(struct kmem_cache *s, void *object)
395{
396    if (!(s->flags & SLAB_STORE_USER))
397        return;
398
399    print_track("Allocated", get_track(s, object, TRACK_ALLOC));
400    print_track("Freed", get_track(s, object, TRACK_FREE));
401}
402
403static void print_page_info(struct page *page)
404{
405    printk(KERN_ERR "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
406        page, page->objects, page->inuse, page->freelist, page->flags);
407
408}
409
410static void slab_bug(struct kmem_cache *s, char *fmt, ...)
411{
412    va_list args;
413    char buf[100];
414
415    va_start(args, fmt);
416    vsnprintf(buf, sizeof(buf), fmt, args);
417    va_end(args);
418    printk(KERN_ERR "========================================"
419            "=====================================\n");
420    printk(KERN_ERR "BUG %s: %s\n", s->name, buf);
421    printk(KERN_ERR "----------------------------------------"
422            "-------------------------------------\n\n");
423}
424
425static void slab_fix(struct kmem_cache *s, char *fmt, ...)
426{
427    va_list args;
428    char buf[100];
429
430    va_start(args, fmt);
431    vsnprintf(buf, sizeof(buf), fmt, args);
432    va_end(args);
433    printk(KERN_ERR "FIX %s: %s\n", s->name, buf);
434}
435
436static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
437{
438    unsigned int off; /* Offset of last byte */
439    u8 *addr = page_address(page);
440
441    print_tracking(s, p);
442
443    print_page_info(page);
444
445    printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
446            p, p - addr, get_freepointer(s, p));
447
448    if (p > addr + 16)
449        print_section("Bytes b4", p - 16, 16);
450
451    print_section("Object", p, min_t(unsigned long, s->objsize, PAGE_SIZE));
452
453    if (s->flags & SLAB_RED_ZONE)
454        print_section("Redzone", p + s->objsize,
455            s->inuse - s->objsize);
456
457    if (s->offset)
458        off = s->offset + sizeof(void *);
459    else
460        off = s->inuse;
461
462    if (s->flags & SLAB_STORE_USER)
463        off += 2 * sizeof(struct track);
464
465    if (off != s->size)
466        /* Beginning of the filler is the free pointer */
467        print_section("Padding", p + off, s->size - off);
468
469    dump_stack();
470}
471
472static void object_err(struct kmem_cache *s, struct page *page,
473            u8 *object, char *reason)
474{
475    slab_bug(s, "%s", reason);
476    print_trailer(s, page, object);
477}
478
479static void slab_err(struct kmem_cache *s, struct page *page, char *fmt, ...)
480{
481    va_list args;
482    char buf[100];
483
484    va_start(args, fmt);
485    vsnprintf(buf, sizeof(buf), fmt, args);
486    va_end(args);
487    slab_bug(s, "%s", buf);
488    print_page_info(page);
489    dump_stack();
490}
491
492static void init_object(struct kmem_cache *s, void *object, int active)
493{
494    u8 *p = object;
495
496    if (s->flags & __OBJECT_POISON) {
497        memset(p, POISON_FREE, s->objsize - 1);
498        p[s->objsize - 1] = POISON_END;
499    }
500
501    if (s->flags & SLAB_RED_ZONE)
502        memset(p + s->objsize,
503            active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE,
504            s->inuse - s->objsize);
505}
506
507static u8 *check_bytes(u8 *start, unsigned int value, unsigned int bytes)
508{
509    while (bytes) {
510        if (*start != (u8)value)
511            return start;
512        start++;
513        bytes--;
514    }
515    return NULL;
516}
517
518static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
519                        void *from, void *to)
520{
521    slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
522    memset(from, data, to - from);
523}
524
525static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
526            u8 *object, char *what,
527            u8 *start, unsigned int value, unsigned int bytes)
528{
529    u8 *fault;
530    u8 *end;
531
532    fault = check_bytes(start, value, bytes);
533    if (!fault)
534        return 1;
535
536    end = start + bytes;
537    while (end > fault && end[-1] == value)
538        end--;
539
540    slab_bug(s, "%s overwritten", what);
541    printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
542                    fault, end - 1, fault[0], value);
543    print_trailer(s, page, object);
544
545    restore_bytes(s, what, value, fault, end);
546    return 0;
547}
548
549/*
550 * Object layout:
551 *
552 * object address
553 * Bytes of the object to be managed.
554 * If the freepointer may overlay the object then the free
555 * pointer is the first word of the object.
556 *
557 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
558 * 0xa5 (POISON_END)
559 *
560 * object + s->objsize
561 * Padding to reach word boundary. This is also used for Redzoning.
562 * Padding is extended by another word if Redzoning is enabled and
563 * objsize == inuse.
564 *
565 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
566 * 0xcc (RED_ACTIVE) for objects in use.
567 *
568 * object + s->inuse
569 * Meta data starts here.
570 *
571 * A. Free pointer (if we cannot overwrite object on free)
572 * B. Tracking data for SLAB_STORE_USER
573 * C. Padding to reach required alignment boundary or at mininum
574 * one word if debugging is on to be able to detect writes
575 * before the word boundary.
576 *
577 * Padding is done using 0x5a (POISON_INUSE)
578 *
579 * object + s->size
580 * Nothing is used beyond s->size.
581 *
582 * If slabcaches are merged then the objsize and inuse boundaries are mostly
583 * ignored. And therefore no slab options that rely on these boundaries
584 * may be used with merged slabcaches.
585 */
586
587static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
588{
589    unsigned long off = s->inuse; /* The end of info */
590
591    if (s->offset)
592        /* Freepointer is placed after the object. */
593        off += sizeof(void *);
594
595    if (s->flags & SLAB_STORE_USER)
596        /* We also have user information there */
597        off += 2 * sizeof(struct track);
598
599    if (s->size == off)
600        return 1;
601
602    return check_bytes_and_report(s, page, p, "Object padding",
603                p + off, POISON_INUSE, s->size - off);
604}
605
606/* Check the pad bytes at the end of a slab page */
607static int slab_pad_check(struct kmem_cache *s, struct page *page)
608{
609    u8 *start;
610    u8 *fault;
611    u8 *end;
612    int length;
613    int remainder;
614
615    if (!(s->flags & SLAB_POISON))
616        return 1;
617
618    start = page_address(page);
619    length = (PAGE_SIZE << compound_order(page));
620    end = start + length;
621    remainder = length % s->size;
622    if (!remainder)
623        return 1;
624
625    fault = check_bytes(end - remainder, POISON_INUSE, remainder);
626    if (!fault)
627        return 1;
628    while (end > fault && end[-1] == POISON_INUSE)
629        end--;
630
631    slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
632    print_section("Padding", end - remainder, remainder);
633
634    restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
635    return 0;
636}
637
638static int check_object(struct kmem_cache *s, struct page *page,
639                    void *object, int active)
640{
641    u8 *p = object;
642    u8 *endobject = object + s->objsize;
643
644    if (s->flags & SLAB_RED_ZONE) {
645        unsigned int red =
646            active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE;
647
648        if (!check_bytes_and_report(s, page, object, "Redzone",
649            endobject, red, s->inuse - s->objsize))
650            return 0;
651    } else {
652        if ((s->flags & SLAB_POISON) && s->objsize < s->inuse) {
653            check_bytes_and_report(s, page, p, "Alignment padding",
654                endobject, POISON_INUSE, s->inuse - s->objsize);
655        }
656    }
657
658    if (s->flags & SLAB_POISON) {
659        if (!active && (s->flags & __OBJECT_POISON) &&
660            (!check_bytes_and_report(s, page, p, "Poison", p,
661                    POISON_FREE, s->objsize - 1) ||
662             !check_bytes_and_report(s, page, p, "Poison",
663                p + s->objsize - 1, POISON_END, 1)))
664            return 0;
665        /*
666         * check_pad_bytes cleans up on its own.
667         */
668        check_pad_bytes(s, page, p);
669    }
670
671    if (!s->offset && active)
672        /*
673         * Object and freepointer overlap. Cannot check
674         * freepointer while object is allocated.
675         */
676        return 1;
677
678    /* Check free pointer validity */
679    if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
680        object_err(s, page, p, "Freepointer corrupt");
681        /*
682         * No choice but to zap it and thus lose the remainder
683         * of the free objects in this slab. May cause
684         * another error because the object count is now wrong.
685         */
686        set_freepointer(s, p, NULL);
687        return 0;
688    }
689    return 1;
690}
691
692static int check_slab(struct kmem_cache *s, struct page *page)
693{
694    int maxobj;
695
696    VM_BUG_ON(!irqs_disabled());
697
698    if (!PageSlab(page)) {
699        slab_err(s, page, "Not a valid slab page");
700        return 0;
701    }
702
703    maxobj = (PAGE_SIZE << compound_order(page)) / s->size;
704    if (page->objects > maxobj) {
705        slab_err(s, page, "objects %u > max %u",
706            s->name, page->objects, maxobj);
707        return 0;
708    }
709    if (page->inuse > page->objects) {
710        slab_err(s, page, "inuse %u > max %u",
711            s->name, page->inuse, page->objects);
712        return 0;
713    }
714    /* Slab_pad_check fixes things up after itself */
715    slab_pad_check(s, page);
716    return 1;
717}
718
719/*
720 * Determine if a certain object on a page is on the freelist. Must hold the
721 * slab lock to guarantee that the chains are in a consistent state.
722 */
723static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
724{
725    int nr = 0;
726    void *fp = page->freelist;
727    void *object = NULL;
728    unsigned long max_objects;
729
730    while (fp && nr <= page->objects) {
731        if (fp == search)
732            return 1;
733        if (!check_valid_pointer(s, page, fp)) {
734            if (object) {
735                object_err(s, page, object,
736                    "Freechain corrupt");
737                set_freepointer(s, object, NULL);
738                break;
739            } else {
740                slab_err(s, page, "Freepointer corrupt");
741                page->freelist = NULL;
742                page->inuse = page->objects;
743                slab_fix(s, "Freelist cleared");
744                return 0;
745            }
746            break;
747        }
748        object = fp;
749        fp = get_freepointer(s, object);
750        nr++;
751    }
752
753    max_objects = (PAGE_SIZE << compound_order(page)) / s->size;
754    if (max_objects > MAX_OBJS_PER_PAGE)
755        max_objects = MAX_OBJS_PER_PAGE;
756
757    if (page->objects != max_objects) {
758        slab_err(s, page, "Wrong number of objects. Found %d but "
759            "should be %d", page->objects, max_objects);
760        page->objects = max_objects;
761        slab_fix(s, "Number of objects adjusted.");
762    }
763    if (page->inuse != page->objects - nr) {
764        slab_err(s, page, "Wrong object count. Counter is %d but "
765            "counted were %d", page->inuse, page->objects - nr);
766        page->inuse = page->objects - nr;
767        slab_fix(s, "Object count adjusted.");
768    }
769    return search == NULL;
770}
771
772static void trace(struct kmem_cache *s, struct page *page, void *object,
773                                int alloc)
774{
775    if (s->flags & SLAB_TRACE) {
776        printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
777            s->name,
778            alloc ? "alloc" : "free",
779            object, page->inuse,
780            page->freelist);
781
782        if (!alloc)
783            print_section("Object", (void *)object, s->objsize);
784
785        dump_stack();
786    }
787}
788
789/*
790 * Tracking of fully allocated slabs for debugging purposes.
791 */
792static void add_full(struct kmem_cache_node *n, struct page *page)
793{
794    spin_lock(&n->list_lock);
795    list_add(&page->lru, &n->full);
796    spin_unlock(&n->list_lock);
797}
798
799static void remove_full(struct kmem_cache *s, struct page *page)
800{
801    struct kmem_cache_node *n;
802
803    if (!(s->flags & SLAB_STORE_USER))
804        return;
805
806    n = get_node(s, page_to_nid(page));
807
808    spin_lock(&n->list_lock);
809    list_del(&page->lru);
810    spin_unlock(&n->list_lock);
811}
812
813/* Tracking of the number of slabs for debugging purposes */
814static inline unsigned long slabs_node(struct kmem_cache *s, int node)
815{
816    struct kmem_cache_node *n = get_node(s, node);
817
818    return atomic_long_read(&n->nr_slabs);
819}
820
821static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
822{
823    return atomic_long_read(&n->nr_slabs);
824}
825
826static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
827{
828    struct kmem_cache_node *n = get_node(s, node);
829
830    /*
831     * May be called early in order to allocate a slab for the
832     * kmem_cache_node structure. Solve the chicken-egg
833     * dilemma by deferring the increment of the count during
834     * bootstrap (see early_kmem_cache_node_alloc).
835     */
836    if (!NUMA_BUILD || n) {
837        atomic_long_inc(&n->nr_slabs);
838        atomic_long_add(objects, &n->total_objects);
839    }
840}
841static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
842{
843    struct kmem_cache_node *n = get_node(s, node);
844
845    atomic_long_dec(&n->nr_slabs);
846    atomic_long_sub(objects, &n->total_objects);
847}
848
849/* Object debug checks for alloc/free paths */
850static void setup_object_debug(struct kmem_cache *s, struct page *page,
851                                void *object)
852{
853    if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
854        return;
855
856    init_object(s, object, 0);
857    init_tracking(s, object);
858}
859
860static int alloc_debug_processing(struct kmem_cache *s, struct page *page,
861                    void *object, unsigned long addr)
862{
863    if (!check_slab(s, page))
864        goto bad;
865
866    if (!on_freelist(s, page, object)) {
867        object_err(s, page, object, "Object already allocated");
868        goto bad;
869    }
870
871    if (!check_valid_pointer(s, page, object)) {
872        object_err(s, page, object, "Freelist Pointer check fails");
873        goto bad;
874    }
875
876    if (!check_object(s, page, object, 0))
877        goto bad;
878
879    /* Success perform special debug activities for allocs */
880    if (s->flags & SLAB_STORE_USER)
881        set_track(s, object, TRACK_ALLOC, addr);
882    trace(s, page, object, 1);
883    init_object(s, object, 1);
884    return 1;
885
886bad:
887    if (PageSlab(page)) {
888        /*
889         * If this is a slab page then lets do the best we can
890         * to avoid issues in the future. Marking all objects
891         * as used avoids touching the remaining objects.
892         */
893        slab_fix(s, "Marking all objects used");
894        page->inuse = page->objects;
895        page->freelist = NULL;
896    }
897    return 0;
898}
899
900static int free_debug_processing(struct kmem_cache *s, struct page *page,
901                    void *object, unsigned long addr)
902{
903    if (!check_slab(s, page))
904        goto fail;
905
906    if (!check_valid_pointer(s, page, object)) {
907        slab_err(s, page, "Invalid object pointer 0x%p", object);
908        goto fail;
909    }
910
911    if (on_freelist(s, page, object)) {
912        object_err(s, page, object, "Object already free");
913        goto fail;
914    }
915
916    if (!check_object(s, page, object, 1))
917        return 0;
918
919    if (unlikely(s != page->slab)) {
920        if (!PageSlab(page)) {
921            slab_err(s, page, "Attempt to free object(0x%p) "
922                "outside of slab", object);
923        } else if (!page->slab) {
924            printk(KERN_ERR
925                "SLUB <none>: no slab for object 0x%p.\n",
926                        object);
927            dump_stack();
928        } else
929            object_err(s, page, object,
930                    "page slab pointer corrupt.");
931        goto fail;
932    }
933
934    /* Special debug activities for freeing objects */
935    if (!PageSlubFrozen(page) && !page->freelist)
936        remove_full(s, page);
937    if (s->flags & SLAB_STORE_USER)
938        set_track(s, object, TRACK_FREE, addr);
939    trace(s, page, object, 0);
940    init_object(s, object, 0);
941    return 1;
942
943fail:
944    slab_fix(s, "Object at 0x%p not freed", object);
945    return 0;
946}
947
948static int __init setup_slub_debug(char *str)
949{
950    slub_debug = DEBUG_DEFAULT_FLAGS;
951    if (*str++ != '=' || !*str)
952        /*
953         * No options specified. Switch on full debugging.
954         */
955        goto out;
956
957    if (*str == ',')
958        /*
959         * No options but restriction on slabs. This means full
960         * debugging for slabs matching a pattern.
961         */
962        goto check_slabs;
963
964    if (tolower(*str) == 'o') {
965        /*
966         * Avoid enabling debugging on caches if its minimum order
967         * would increase as a result.
968         */
969        disable_higher_order_debug = 1;
970        goto out;
971    }
972
973    slub_debug = 0;
974    if (*str == '-')
975        /*
976         * Switch off all debugging measures.
977         */
978        goto out;
979
980    /*
981     * Determine which debug features should be switched on
982     */
983    for (; *str && *str != ','; str++) {
984        switch (tolower(*str)) {
985        case 'f':
986            slub_debug |= SLAB_DEBUG_FREE;
987            break;
988        case 'z':
989            slub_debug |= SLAB_RED_ZONE;
990            break;
991        case 'p':
992            slub_debug |= SLAB_POISON;
993            break;
994        case 'u':
995            slub_debug |= SLAB_STORE_USER;
996            break;
997        case 't':
998            slub_debug |= SLAB_TRACE;
999            break;
1000        case 'a':
1001            slub_debug |= SLAB_FAILSLAB;
1002            break;
1003        default:
1004            printk(KERN_ERR "slub_debug option '%c' "
1005                "unknown. skipped\n", *str);
1006        }
1007    }
1008
1009check_slabs:
1010    if (*str == ',')
1011        slub_debug_slabs = str + 1;
1012out:
1013    return 1;
1014}
1015
1016__setup("slub_debug", setup_slub_debug);
1017
1018static unsigned long kmem_cache_flags(unsigned long objsize,
1019    unsigned long flags, const char *name,
1020    void (*ctor)(void *))
1021{
1022    /*
1023     * Enable debugging if selected on the kernel commandline.
1024     */
1025    if (slub_debug && (!slub_debug_slabs ||
1026        !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))
1027        flags |= slub_debug;
1028
1029    return flags;
1030}
1031#else
1032static inline void setup_object_debug(struct kmem_cache *s,
1033            struct page *page, void *object) {}
1034
1035static inline int alloc_debug_processing(struct kmem_cache *s,
1036    struct page *page, void *object, unsigned long addr) { return 0; }
1037
1038static inline int free_debug_processing(struct kmem_cache *s,
1039    struct page *page, void *object, unsigned long addr) { return 0; }
1040
1041static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1042            { return 1; }
1043static inline int check_object(struct kmem_cache *s, struct page *page,
1044            void *object, int active) { return 1; }
1045static inline void add_full(struct kmem_cache_node *n, struct page *page) {}
1046static inline unsigned long kmem_cache_flags(unsigned long objsize,
1047    unsigned long flags, const char *name,
1048    void (*ctor)(void *))
1049{
1050    return flags;
1051}
1052#define slub_debug 0
1053
1054#define disable_higher_order_debug 0
1055
1056static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1057                            { return 0; }
1058static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1059                            { return 0; }
1060static inline void inc_slabs_node(struct kmem_cache *s, int node,
1061                            int objects) {}
1062static inline void dec_slabs_node(struct kmem_cache *s, int node,
1063                            int objects) {}
1064#endif
1065
1066/*
1067 * Slab allocation and freeing
1068 */
1069static inline struct page *alloc_slab_page(gfp_t flags, int node,
1070                    struct kmem_cache_order_objects oo)
1071{
1072    int order = oo_order(oo);
1073
1074    flags |= __GFP_NOTRACK;
1075
1076    if (node == -1)
1077        return alloc_pages(flags, order);
1078    else
1079        return alloc_pages_exact_node(node, flags, order);
1080}
1081
1082static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1083{
1084    struct page *page;
1085    struct kmem_cache_order_objects oo = s->oo;
1086    gfp_t alloc_gfp;
1087
1088    flags |= s->allocflags;
1089
1090    /*
1091     * Let the initial higher-order allocation fail under memory pressure
1092     * so we fall-back to the minimum order allocation.
1093     */
1094    alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1095
1096    page = alloc_slab_page(alloc_gfp, node, oo);
1097    if (unlikely(!page)) {
1098        oo = s->min;
1099        /*
1100         * Allocation may have failed due to fragmentation.
1101         * Try a lower order alloc if possible
1102         */
1103        page = alloc_slab_page(flags, node, oo);
1104        if (!page)
1105            return NULL;
1106
1107        stat(s, ORDER_FALLBACK);
1108    }
1109
1110    if (kmemcheck_enabled
1111        && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
1112        int pages = 1 << oo_order(oo);
1113
1114        kmemcheck_alloc_shadow(page, oo_order(oo), flags, node);
1115
1116        /*
1117         * Objects from caches that have a constructor don't get
1118         * cleared when they're allocated, so we need to do it here.
1119         */
1120        if (s->ctor)
1121            kmemcheck_mark_uninitialized_pages(page, pages);
1122        else
1123            kmemcheck_mark_unallocated_pages(page, pages);
1124    }
1125
1126    page->objects = oo_objects(oo);
1127    mod_zone_page_state(page_zone(page),
1128        (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1129        NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1130        1 << oo_order(oo));
1131
1132    return page;
1133}
1134
1135static void setup_object(struct kmem_cache *s, struct page *page,
1136                void *object)
1137{
1138    setup_object_debug(s, page, object);
1139    if (unlikely(s->ctor))
1140        s->ctor(object);
1141}
1142
1143static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1144{
1145    struct page *page;
1146    void *start;
1147    void *last;
1148    void *p;
1149
1150    BUG_ON(flags & GFP_SLAB_BUG_MASK);
1151
1152    page = allocate_slab(s,
1153        flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1154    if (!page)
1155        goto out;
1156
1157    inc_slabs_node(s, page_to_nid(page), page->objects);
1158    page->slab = s;
1159    page->flags |= 1 << PG_slab;
1160    if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON |
1161            SLAB_STORE_USER | SLAB_TRACE))
1162        __SetPageSlubDebug(page);
1163
1164    start = page_address(page);
1165
1166    if (unlikely(s->flags & SLAB_POISON))
1167        memset(start, POISON_INUSE, PAGE_SIZE << compound_order(page));
1168
1169    last = start;
1170    for_each_object(p, s, start, page->objects) {
1171        setup_object(s, page, last);
1172        set_freepointer(s, last, p);
1173        last = p;
1174    }
1175    setup_object(s, page, last);
1176    set_freepointer(s, last, NULL);
1177
1178    page->freelist = start;
1179    page->inuse = 0;
1180out:
1181    return page;
1182}
1183
1184static void __free_slab(struct kmem_cache *s, struct page *page)
1185{
1186    int order = compound_order(page);
1187    int pages = 1 << order;
1188
1189    if (unlikely(SLABDEBUG && PageSlubDebug(page))) {
1190        void *p;
1191
1192        slab_pad_check(s, page);
1193        for_each_object(p, s, page_address(page),
1194                        page->objects)
1195            check_object(s, page, p, 0);
1196        __ClearPageSlubDebug(page);
1197    }
1198
1199    kmemcheck_free_shadow(page, compound_order(page));
1200
1201    mod_zone_page_state(page_zone(page),
1202        (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1203        NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1204        -pages);
1205
1206    __ClearPageSlab(page);
1207    reset_page_mapcount(page);
1208    if (current->reclaim_state)
1209        current->reclaim_state->reclaimed_slab += pages;
1210    __free_pages(page, order);
1211}
1212
1213static void rcu_free_slab(struct rcu_head *h)
1214{
1215    struct page *page;
1216
1217    page = container_of((struct list_head *)h, struct page, lru);
1218    __free_slab(page->slab, page);
1219}
1220
1221static void free_slab(struct kmem_cache *s, struct page *page)
1222{
1223    if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
1224        /*
1225         * RCU free overloads the RCU head over the LRU
1226         */
1227        struct rcu_head *head = (void *)&page->lru;
1228
1229        call_rcu(head, rcu_free_slab);
1230    } else
1231        __free_slab(s, page);
1232}
1233
1234static void discard_slab(struct kmem_cache *s, struct page *page)
1235{
1236    dec_slabs_node(s, page_to_nid(page), page->objects);
1237    free_slab(s, page);
1238}
1239
1240/*
1241 * Per slab locking using the pagelock
1242 */
1243static __always_inline void slab_lock(struct page *page)
1244{
1245    bit_spin_lock(PG_locked, &page->flags);
1246}
1247
1248static __always_inline void slab_unlock(struct page *page)
1249{
1250    __bit_spin_unlock(PG_locked, &page->flags);
1251}
1252
1253static __always_inline int slab_trylock(struct page *page)
1254{
1255    int rc = 1;
1256
1257    rc = bit_spin_trylock(PG_locked, &page->flags);
1258    return rc;
1259}
1260
1261/*
1262 * Management of partially allocated slabs
1263 */
1264static void add_partial(struct kmem_cache_node *n,
1265                struct page *page, int tail)
1266{
1267    spin_lock(&n->list_lock);
1268    n->nr_partial++;
1269    if (tail)
1270        list_add_tail(&page->lru, &n->partial);
1271    else
1272        list_add(&page->lru, &n->partial);
1273    spin_unlock(&n->list_lock);
1274}
1275
1276static void remove_partial(struct kmem_cache *s, struct page *page)
1277{
1278    struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1279
1280    spin_lock(&n->list_lock);
1281    list_del(&page->lru);
1282    n->nr_partial--;
1283    spin_unlock(&n->list_lock);
1284}
1285
1286/*
1287 * Lock slab and remove from the partial list.
1288 *
1289 * Must hold list_lock.
1290 */
1291static inline int lock_and_freeze_slab(struct kmem_cache_node *n,
1292                            struct page *page)
1293{
1294    if (slab_trylock(page)) {
1295        list_del(&page->lru);
1296        n->nr_partial--;
1297        __SetPageSlubFrozen(page);
1298        return 1;
1299    }
1300    return 0;
1301}
1302
1303/*
1304 * Try to allocate a partial slab from a specific node.
1305 */
1306static struct page *get_partial_node(struct kmem_cache_node *n)
1307{
1308    struct page *page;
1309
1310    /*
1311     * Racy check. If we mistakenly see no partial slabs then we
1312     * just allocate an empty slab. If we mistakenly try to get a
1313     * partial slab and there is none available then get_partials()
1314     * will return NULL.
1315     */
1316    if (!n || !n->nr_partial)
1317        return NULL;
1318
1319    spin_lock(&n->list_lock);
1320    list_for_each_entry(page, &n->partial, lru)
1321        if (lock_and_freeze_slab(n, page))
1322            goto out;
1323    page = NULL;
1324out:
1325    spin_unlock(&n->list_lock);
1326    return page;
1327}
1328
1329/*
1330 * Get a page from somewhere. Search in increasing NUMA distances.
1331 */
1332static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
1333{
1334#ifdef CONFIG_NUMA
1335    struct zonelist *zonelist;
1336    struct zoneref *z;
1337    struct zone *zone;
1338    enum zone_type high_zoneidx = gfp_zone(flags);
1339    struct page *page;
1340
1341    /*
1342     * The defrag ratio allows a configuration of the tradeoffs between
1343     * inter node defragmentation and node local allocations. A lower
1344     * defrag_ratio increases the tendency to do local allocations
1345     * instead of attempting to obtain partial slabs from other nodes.
1346     *
1347     * If the defrag_ratio is set to 0 then kmalloc() always
1348     * returns node local objects. If the ratio is higher then kmalloc()
1349     * may return off node objects because partial slabs are obtained
1350     * from other nodes and filled up.
1351     *
1352     * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1353     * defrag_ratio = 1000) then every (well almost) allocation will
1354     * first attempt to defrag slab caches on other nodes. This means
1355     * scanning over all nodes to look for partial slabs which may be
1356     * expensive if we do it every time we are trying to find a slab
1357     * with available objects.
1358     */
1359    if (!s->remote_node_defrag_ratio ||
1360            get_cycles() % 1024 > s->remote_node_defrag_ratio)
1361        return NULL;
1362
1363    get_mems_allowed();
1364    zonelist = node_zonelist(slab_node(current->mempolicy), flags);
1365    for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1366        struct kmem_cache_node *n;
1367
1368        n = get_node(s, zone_to_nid(zone));
1369
1370        if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
1371                n->nr_partial > s->min_partial) {
1372            page = get_partial_node(n);
1373            if (page) {
1374                put_mems_allowed();
1375                return page;
1376            }
1377        }
1378    }
1379    put_mems_allowed();
1380#endif
1381    return NULL;
1382}
1383
1384/*
1385 * Get a partial page, lock it and return it.
1386 */
1387static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
1388{
1389    struct page *page;
1390    int searchnode = (node == -1) ? numa_node_id() : node;
1391
1392    page = get_partial_node(get_node(s, searchnode));
1393    if (page || (flags & __GFP_THISNODE))
1394        return page;
1395
1396    return get_any_partial(s, flags);
1397}
1398
1399/*
1400 * Move a page back to the lists.
1401 *
1402 * Must be called with the slab lock held.
1403 *
1404 * On exit the slab lock will have been dropped.
1405 */
1406static void unfreeze_slab(struct kmem_cache *s, struct page *page, int tail)
1407{
1408    struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1409
1410    __ClearPageSlubFrozen(page);
1411    if (page->inuse) {
1412
1413        if (page->freelist) {
1414            add_partial(n, page, tail);
1415            stat(s, tail ? DEACTIVATE_TO_TAIL : DEACTIVATE_TO_HEAD);
1416        } else {
1417            stat(s, DEACTIVATE_FULL);
1418            if (SLABDEBUG && PageSlubDebug(page) &&
1419                        (s->flags & SLAB_STORE_USER))
1420                add_full(n, page);
1421        }
1422        slab_unlock(page);
1423    } else {
1424        stat(s, DEACTIVATE_EMPTY);
1425        if (n->nr_partial < s->min_partial) {
1426            /*
1427             * Adding an empty slab to the partial slabs in order
1428             * to avoid page allocator overhead. This slab needs
1429             * to come after the other slabs with objects in
1430             * so that the others get filled first. That way the
1431             * size of the partial list stays small.
1432             *
1433             * kmem_cache_shrink can reclaim any empty slabs from
1434             * the partial list.
1435             */
1436            add_partial(n, page, 1);
1437            slab_unlock(page);
1438        } else {
1439            slab_unlock(page);
1440            stat(s, FREE_SLAB);
1441            discard_slab(s, page);
1442        }
1443    }
1444}
1445
1446/*
1447 * Remove the cpu slab
1448 */
1449static void deactivate_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1450{
1451    struct page *page = c->page;
1452    int tail = 1;
1453
1454    if (page->freelist)
1455        stat(s, DEACTIVATE_REMOTE_FREES);
1456    /*
1457     * Merge cpu freelist into slab freelist. Typically we get here
1458     * because both freelists are empty. So this is unlikely
1459     * to occur.
1460     */
1461    while (unlikely(c->freelist)) {
1462        void **object;
1463
1464        tail = 0; /* Hot objects. Put the slab first */
1465
1466        /* Retrieve object from cpu_freelist */
1467        object = c->freelist;
1468        c->freelist = get_freepointer(s, c->freelist);
1469
1470        /* And put onto the regular freelist */
1471        set_freepointer(s, object, page->freelist);
1472        page->freelist = object;
1473        page->inuse--;
1474    }
1475    c->page = NULL;
1476    unfreeze_slab(s, page, tail);
1477}
1478
1479static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
1480{
1481    stat(s, CPUSLAB_FLUSH);
1482    slab_lock(c->page);
1483    deactivate_slab(s, c);
1484}
1485
1486/*
1487 * Flush cpu slab.
1488 *
1489 * Called from IPI handler with interrupts disabled.
1490 */
1491static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
1492{
1493    struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
1494
1495    if (likely(c && c->page))
1496        flush_slab(s, c);
1497}
1498
1499static void flush_cpu_slab(void *d)
1500{
1501    struct kmem_cache *s = d;
1502
1503    __flush_cpu_slab(s, smp_processor_id());
1504}
1505
1506static void flush_all(struct kmem_cache *s)
1507{
1508    on_each_cpu(flush_cpu_slab, s, 1);
1509}
1510
1511/*
1512 * Check if the objects in a per cpu structure fit numa
1513 * locality expectations.
1514 */
1515static inline int node_match(struct kmem_cache_cpu *c, int node)
1516{
1517#ifdef CONFIG_NUMA
1518    if (node != -1 && c->node != node)
1519        return 0;
1520#endif
1521    return 1;
1522}
1523
1524static int count_free(struct page *page)
1525{
1526    return page->objects - page->inuse;
1527}
1528
1529static unsigned long count_partial(struct kmem_cache_node *n,
1530                    int (*get_count)(struct page *))
1531{
1532    unsigned long flags;
1533    unsigned long x = 0;
1534    struct page *page;
1535
1536    spin_lock_irqsave(&n->list_lock, flags);
1537    list_for_each_entry(page, &n->partial, lru)
1538        x += get_count(page);
1539    spin_unlock_irqrestore(&n->list_lock, flags);
1540    return x;
1541}
1542
1543static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
1544{
1545#ifdef CONFIG_SLUB_DEBUG
1546    return atomic_long_read(&n->total_objects);
1547#else
1548    return 0;
1549#endif
1550}
1551
1552static noinline void
1553slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
1554{
1555    int node;
1556
1557    printk(KERN_WARNING
1558        "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
1559        nid, gfpflags);
1560    printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, "
1561        "default order: %d, min order: %d\n", s->name, s->objsize,
1562        s->size, oo_order(s->oo), oo_order(s->min));
1563
1564    if (oo_order(s->min) > get_order(s->objsize))
1565        printk(KERN_WARNING " %s debugging increased min order, use "
1566               "slub_debug=O to disable.\n", s->name);
1567
1568    for_each_online_node(node) {
1569        struct kmem_cache_node *n = get_node(s, node);
1570        unsigned long nr_slabs;
1571        unsigned long nr_objs;
1572        unsigned long nr_free;
1573
1574        if (!n)
1575            continue;
1576
1577        nr_free = count_partial(n, count_free);
1578        nr_slabs = node_nr_slabs(n);
1579        nr_objs = node_nr_objs(n);
1580
1581        printk(KERN_WARNING
1582            " node %d: slabs: %ld, objs: %ld, free: %ld\n",
1583            node, nr_slabs, nr_objs, nr_free);
1584    }
1585}
1586
1587/*
1588 * Slow path. The lockless freelist is empty or we need to perform
1589 * debugging duties.
1590 *
1591 * Interrupts are disabled.
1592 *
1593 * Processing is still very fast if new objects have been freed to the
1594 * regular freelist. In that case we simply take over the regular freelist
1595 * as the lockless freelist and zap the regular freelist.
1596 *
1597 * If that is not working then we fall back to the partial lists. We take the
1598 * first element of the freelist as the object to allocate now and move the
1599 * rest of the freelist to the lockless freelist.
1600 *
1601 * And if we were unable to get a new slab from the partial slab lists then
1602 * we need to allocate a new slab. This is the slowest path since it involves
1603 * a call to the page allocator and the setup of a new slab.
1604 */
1605static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
1606              unsigned long addr, struct kmem_cache_cpu *c)
1607{
1608    void **object;
1609    struct page *new;
1610
1611    /* We handle __GFP_ZERO in the caller */
1612    gfpflags &= ~__GFP_ZERO;
1613
1614    if (!c->page)
1615        goto new_slab;
1616
1617    slab_lock(c->page);
1618    if (unlikely(!node_match(c, node)))
1619        goto another_slab;
1620
1621    stat(s, ALLOC_REFILL);
1622
1623load_freelist:
1624    object = c->page->freelist;
1625    if (unlikely(!object))
1626        goto another_slab;
1627    if (unlikely(SLABDEBUG && PageSlubDebug(c->page)))
1628        goto debug;
1629
1630    c->freelist = get_freepointer(s, object);
1631    c->page->inuse = c->page->objects;
1632    c->page->freelist = NULL;
1633    c->node = page_to_nid(c->page);
1634unlock_out:
1635    slab_unlock(c->page);
1636    stat(s, ALLOC_SLOWPATH);
1637    return object;
1638
1639another_slab:
1640    deactivate_slab(s, c);
1641
1642new_slab:
1643    new = get_partial(s, gfpflags, node);
1644    if (new) {
1645        c->page = new;
1646        stat(s, ALLOC_FROM_PARTIAL);
1647        goto load_freelist;
1648    }
1649
1650    if (gfpflags & __GFP_WAIT)
1651        local_irq_enable();
1652
1653    new = new_slab(s, gfpflags, node);
1654
1655    if (gfpflags & __GFP_WAIT)
1656        local_irq_disable();
1657
1658    if (new) {
1659        c = __this_cpu_ptr(s->cpu_slab);
1660        stat(s, ALLOC_SLAB);
1661        if (c->page)
1662            flush_slab(s, c);
1663        slab_lock(new);
1664        __SetPageSlubFrozen(new);
1665        c->page = new;
1666        goto load_freelist;
1667    }
1668    if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
1669        slab_out_of_memory(s, gfpflags, node);
1670    return NULL;
1671debug:
1672    if (!alloc_debug_processing(s, c->page, object, addr))
1673        goto another_slab;
1674
1675    c->page->inuse++;
1676    c->page->freelist = get_freepointer(s, object);
1677    c->node = -1;
1678    goto unlock_out;
1679}
1680
1681/*
1682 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
1683 * have the fastpath folded into their functions. So no function call
1684 * overhead for requests that can be satisfied on the fastpath.
1685 *
1686 * The fastpath works by first checking if the lockless freelist can be used.
1687 * If not then __slab_alloc is called for slow processing.
1688 *
1689 * Otherwise we can simply pick the next object from the lockless free list.
1690 */
1691static __always_inline void *slab_alloc(struct kmem_cache *s,
1692        gfp_t gfpflags, int node, unsigned long addr)
1693{
1694    void **object;
1695    struct kmem_cache_cpu *c;
1696    unsigned long flags;
1697
1698    gfpflags &= gfp_allowed_mask;
1699
1700    lockdep_trace_alloc(gfpflags);
1701    might_sleep_if(gfpflags & __GFP_WAIT);
1702
1703    if (should_failslab(s->objsize, gfpflags, s->flags))
1704        return NULL;
1705
1706    local_irq_save(flags);
1707    c = __this_cpu_ptr(s->cpu_slab);
1708    object = c->freelist;
1709    if (unlikely(!object || !node_match(c, node)))
1710
1711        object = __slab_alloc(s, gfpflags, node, addr, c);
1712
1713    else {
1714        c->freelist = get_freepointer(s, object);
1715        stat(s, ALLOC_FASTPATH);
1716    }
1717    local_irq_restore(flags);
1718
1719    if (unlikely(gfpflags & __GFP_ZERO) && object)
1720        memset(object, 0, s->objsize);
1721
1722    kmemcheck_slab_alloc(s, gfpflags, object, s->objsize);
1723    kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, gfpflags);
1724
1725    return object;
1726}
1727
1728void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
1729{
1730    void *ret = slab_alloc(s, gfpflags, -1, _RET_IP_);
1731
1732    trace_kmem_cache_alloc(_RET_IP_, ret, s->objsize, s->size, gfpflags);
1733
1734    return ret;
1735}
1736EXPORT_SYMBOL(kmem_cache_alloc);
1737
1738#ifdef CONFIG_TRACING
1739void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
1740{
1741    return slab_alloc(s, gfpflags, -1, _RET_IP_);
1742}
1743EXPORT_SYMBOL(kmem_cache_alloc_notrace);
1744#endif
1745
1746#ifdef CONFIG_NUMA
1747void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
1748{
1749    void *ret = slab_alloc(s, gfpflags, node, _RET_IP_);
1750
1751    trace_kmem_cache_alloc_node(_RET_IP_, ret,
1752                    s->objsize, s->size, gfpflags, node);
1753
1754    return ret;
1755}
1756EXPORT_SYMBOL(kmem_cache_alloc_node);
1757#endif
1758
1759#ifdef CONFIG_TRACING
1760void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
1761                    gfp_t gfpflags,
1762                    int node)
1763{
1764    return slab_alloc(s, gfpflags, node, _RET_IP_);
1765}
1766EXPORT_SYMBOL(kmem_cache_alloc_node_notrace);
1767#endif
1768
1769/*
1770 * Slow patch handling. This may still be called frequently since objects
1771 * have a longer lifetime than the cpu slabs in most processing loads.
1772 *
1773 * So we still attempt to reduce cache line usage. Just take the slab
1774 * lock and free the item. If there is no additional partial page
1775 * handling required then we can return immediately.
1776 */
1777static void __slab_free(struct kmem_cache *s, struct page *page,
1778            void *x, unsigned long addr)
1779{
1780    void *prior;
1781    void **object = (void *)x;
1782
1783    stat(s, FREE_SLOWPATH);
1784    slab_lock(page);
1785
1786    if (unlikely(SLABDEBUG && PageSlubDebug(page)))
1787        goto debug;
1788
1789checks_ok:
1790    prior = page->freelist;
1791    set_freepointer(s, object, prior);
1792    page->freelist = object;
1793    page->inuse--;
1794
1795    if (unlikely(PageSlubFrozen(page))) {
1796        stat(s, FREE_FROZEN);
1797        goto out_unlock;
1798    }
1799
1800    if (unlikely(!page->inuse))
1801        goto slab_empty;
1802
1803    /*
1804     * Objects left in the slab. If it was not on the partial list before
1805     * then add it.
1806     */
1807    if (unlikely(!prior)) {
1808        add_partial(get_node(s, page_to_nid(page)), page, 1);
1809        stat(s, FREE_ADD_PARTIAL);
1810    }
1811
1812out_unlock:
1813    slab_unlock(page);
1814    return;
1815
1816slab_empty:
1817    if (prior) {
1818        /*
1819         * Slab still on the partial list.
1820         */
1821        remove_partial(s, page);
1822        stat(s, FREE_REMOVE_PARTIAL);
1823    }
1824    slab_unlock(page);
1825    stat(s, FREE_SLAB);
1826    discard_slab(s, page);
1827    return;
1828
1829debug:
1830    if (!free_debug_processing(s, page, x, addr))
1831        goto out_unlock;
1832    goto checks_ok;
1833}
1834
1835/*
1836 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
1837 * can perform fastpath freeing without additional function calls.
1838 *
1839 * The fastpath is only possible if we are freeing to the current cpu slab
1840 * of this processor. This typically the case if we have just allocated
1841 * the item before.
1842 *
1843 * If fastpath is not possible then fall back to __slab_free where we deal
1844 * with all sorts of special processing.
1845 */
1846static __always_inline void slab_free(struct kmem_cache *s,
1847            struct page *page, void *x, unsigned long addr)
1848{
1849    void **object = (void *)x;
1850    struct kmem_cache_cpu *c;
1851    unsigned long flags;
1852
1853    kmemleak_free_recursive(x, s->flags);
1854    local_irq_save(flags);
1855    c = __this_cpu_ptr(s->cpu_slab);
1856    kmemcheck_slab_free(s, object, s->objsize);
1857    debug_check_no_locks_freed(object, s->objsize);
1858    if (!(s->flags & SLAB_DEBUG_OBJECTS))
1859        debug_check_no_obj_freed(object, s->objsize);
1860    if (likely(page == c->page && c->node >= 0)) {
1861        set_freepointer(s, object, c->freelist);
1862        c->freelist = object;
1863        stat(s, FREE_FASTPATH);
1864    } else
1865        __slab_free(s, page, x, addr);
1866
1867    local_irq_restore(flags);
1868}
1869
1870void kmem_cache_free(struct kmem_cache *s, void *x)
1871{
1872    struct page *page;
1873
1874    page = virt_to_head_page(x);
1875
1876    slab_free(s, page, x, _RET_IP_);
1877
1878    trace_kmem_cache_free(_RET_IP_, x);
1879}
1880EXPORT_SYMBOL(kmem_cache_free);
1881
1882/* Figure out on which slab page the object resides */
1883static struct page *get_object_page(const void *x)
1884{
1885    struct page *page = virt_to_head_page(x);
1886
1887    if (!PageSlab(page))
1888        return NULL;
1889
1890    return page;
1891}
1892
1893/*
1894 * Object placement in a slab is made very easy because we always start at
1895 * offset 0. If we tune the size of the object to the alignment then we can
1896 * get the required alignment by putting one properly sized object after
1897 * another.
1898 *
1899 * Notice that the allocation order determines the sizes of the per cpu
1900 * caches. Each processor has always one slab available for allocations.
1901 * Increasing the allocation order reduces the number of times that slabs
1902 * must be moved on and off the partial lists and is therefore a factor in
1903 * locking overhead.
1904 */
1905
1906/*
1907 * Mininum / Maximum order of slab pages. This influences locking overhead
1908 * and slab fragmentation. A higher order reduces the number of partial slabs
1909 * and increases the number of allocations possible without having to
1910 * take the list_lock.
1911 */
1912static int slub_min_order;
1913static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
1914static int slub_min_objects;
1915
1916/*
1917 * Merge control. If this is set then no merging of slab caches will occur.
1918 * (Could be removed. This was introduced to pacify the merge skeptics.)
1919 */
1920static int slub_nomerge;
1921
1922/*
1923 * Calculate the order of allocation given an slab object size.
1924 *
1925 * The order of allocation has significant impact on performance and other
1926 * system components. Generally order 0 allocations should be preferred since
1927 * order 0 does not cause fragmentation in the page allocator. Larger objects
1928 * be problematic to put into order 0 slabs because there may be too much
1929 * unused space left. We go to a higher order if more than 1/16th of the slab
1930 * would be wasted.
1931 *
1932 * In order to reach satisfactory performance we must ensure that a minimum
1933 * number of objects is in one slab. Otherwise we may generate too much
1934 * activity on the partial lists which requires taking the list_lock. This is
1935 * less a concern for large slabs though which are rarely used.
1936 *
1937 * slub_max_order specifies the order where we begin to stop considering the
1938 * number of objects in a slab as critical. If we reach slub_max_order then
1939 * we try to keep the page order as low as possible. So we accept more waste
1940 * of space in favor of a small page order.
1941 *
1942 * Higher order allocations also allow the placement of more objects in a
1943 * slab and thereby reduce object handling overhead. If the user has
1944 * requested a higher mininum order then we start with that one instead of
1945 * the smallest order which will fit the object.
1946 */
1947static inline int slab_order(int size, int min_objects,
1948                int max_order, int fract_leftover)
1949{
1950    int order;
1951    int rem;
1952    int min_order = slub_min_order;
1953
1954    if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE)
1955        return get_order(size * MAX_OBJS_PER_PAGE) - 1;
1956
1957    for (order = max(min_order,
1958                fls(min_objects * size - 1) - PAGE_SHIFT);
1959            order <= max_order; order++) {
1960
1961        unsigned long slab_size = PAGE_SIZE << order;
1962
1963        if (slab_size < min_objects * size)
1964            continue;
1965
1966        rem = slab_size % size;
1967
1968        if (rem <= slab_size / fract_leftover)
1969            break;
1970
1971    }
1972
1973    return order;
1974}
1975
1976static inline int calculate_order(int size)
1977{
1978    int order;
1979    int min_objects;
1980    int fraction;
1981    int max_objects;
1982
1983    /*
1984     * Attempt to find best configuration for a slab. This
1985     * works by first attempting to generate a layout with
1986     * the best configuration and backing off gradually.
1987     *
1988     * First we reduce the acceptable waste in a slab. Then
1989     * we reduce the minimum objects required in a slab.
1990     */
1991    min_objects = slub_min_objects;
1992    if (!min_objects)
1993        min_objects = 4 * (fls(nr_cpu_ids) + 1);
1994    max_objects = (PAGE_SIZE << slub_max_order)/size;
1995    min_objects = min(min_objects, max_objects);
1996
1997    while (min_objects > 1) {
1998        fraction = 16;
1999        while (fraction >= 4) {
2000            order = slab_order(size, min_objects,
2001                        slub_max_order, fraction);
2002            if (order <= slub_max_order)
2003                return order;
2004            fraction /= 2;
2005        }
2006        min_objects--;
2007    }
2008
2009    /*
2010     * We were unable to place multiple objects in a slab. Now
2011     * lets see if we can place a single object there.
2012     */
2013    order = slab_order(size, 1, slub_max_order, 1);
2014    if (order <= slub_max_order)
2015        return order;
2016
2017    /*
2018     * Doh this slab cannot be placed using slub_max_order.
2019     */
2020    order = slab_order(size, 1, MAX_ORDER, 1);
2021    if (order < MAX_ORDER)
2022        return order;
2023    return -ENOSYS;
2024}
2025
2026/*
2027 * Figure out what the alignment of the objects will be.
2028 */
2029static unsigned long calculate_alignment(unsigned long flags,
2030        unsigned long align, unsigned long size)
2031{
2032    /*
2033     * If the user wants hardware cache aligned objects then follow that
2034     * suggestion if the object is sufficiently large.
2035     *
2036     * The hardware cache alignment cannot override the specified
2037     * alignment though. If that is greater then use it.
2038     */
2039    if (flags & SLAB_HWCACHE_ALIGN) {
2040        unsigned long ralign = cache_line_size();
2041        while (size <= ralign / 2)
2042            ralign /= 2;
2043        align = max(align, ralign);
2044    }
2045
2046    if (align < ARCH_SLAB_MINALIGN)
2047        align = ARCH_SLAB_MINALIGN;
2048
2049    return ALIGN(align, sizeof(void *));
2050}
2051
2052static void
2053init_kmem_cache_node(struct kmem_cache_node *n, struct kmem_cache *s)
2054{
2055    n->nr_partial = 0;
2056    spin_lock_init(&n->list_lock);
2057    INIT_LIST_HEAD(&n->partial);
2058#ifdef CONFIG_SLUB_DEBUG
2059    atomic_long_set(&n->nr_slabs, 0);
2060    atomic_long_set(&n->total_objects, 0);
2061    INIT_LIST_HEAD(&n->full);
2062#endif
2063}
2064
2065static DEFINE_PER_CPU(struct kmem_cache_cpu, kmalloc_percpu[KMALLOC_CACHES]);
2066
2067static inline int alloc_kmem_cache_cpus(struct kmem_cache *s, gfp_t flags)
2068{
2069    if (s < kmalloc_caches + KMALLOC_CACHES && s >= kmalloc_caches)
2070        /*
2071         * Boot time creation of the kmalloc array. Use static per cpu data
2072         * since the per cpu allocator is not available yet.
2073         */
2074        s->cpu_slab = kmalloc_percpu + (s - kmalloc_caches);
2075    else
2076        s->cpu_slab = alloc_percpu(struct kmem_cache_cpu);
2077
2078    if (!s->cpu_slab)
2079        return 0;
2080
2081    return 1;
2082}
2083
2084#ifdef CONFIG_NUMA
2085/*
2086 * No kmalloc_node yet so do it by hand. We know that this is the first
2087 * slab on the node for this slabcache. There are no concurrent accesses
2088 * possible.
2089 *
2090 * Note that this function only works on the kmalloc_node_cache
2091 * when allocating for the kmalloc_node_cache. This is used for bootstrapping
2092 * memory on a fresh node that has no slab structures yet.
2093 */
2094static void early_kmem_cache_node_alloc(gfp_t gfpflags, int node)
2095{
2096    struct page *page;
2097    struct kmem_cache_node *n;
2098    unsigned long flags;
2099
2100    BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node));
2101
2102    page = new_slab(kmalloc_caches, gfpflags, node);
2103
2104    BUG_ON(!page);
2105    if (page_to_nid(page) != node) {
2106        printk(KERN_ERR "SLUB: Unable to allocate memory from "
2107                "node %d\n", node);
2108        printk(KERN_ERR "SLUB: Allocating a useless per node structure "
2109                "in order to be able to continue\n");
2110    }
2111
2112    n = page->freelist;
2113    BUG_ON(!n);
2114    page->freelist = get_freepointer(kmalloc_caches, n);
2115    page->inuse++;
2116    kmalloc_caches->node[node] = n;
2117#ifdef CONFIG_SLUB_DEBUG
2118    init_object(kmalloc_caches, n, 1);
2119    init_tracking(kmalloc_caches, n);
2120#endif
2121    init_kmem_cache_node(n, kmalloc_caches);
2122    inc_slabs_node(kmalloc_caches, node, page->objects);
2123
2124    /*
2125     * lockdep requires consistent irq usage for each lock
2126     * so even though there cannot be a race this early in
2127     * the boot sequence, we still disable irqs.
2128     */
2129    local_irq_save(flags);
2130    add_partial(n, page, 0);
2131    local_irq_restore(flags);
2132}
2133
2134static void free_kmem_cache_nodes(struct kmem_cache *s)
2135{
2136    int node;
2137
2138    for_each_node_state(node, N_NORMAL_MEMORY) {
2139        struct kmem_cache_node *n = s->node[node];
2140        if (n)
2141            kmem_cache_free(kmalloc_caches, n);
2142        s->node[node] = NULL;
2143    }
2144}
2145
2146static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2147{
2148    int node;
2149
2150    for_each_node_state(node, N_NORMAL_MEMORY) {
2151        struct kmem_cache_node *n;
2152
2153        if (slab_state == DOWN) {
2154            early_kmem_cache_node_alloc(gfpflags, node);
2155            continue;
2156        }
2157        n = kmem_cache_alloc_node(kmalloc_caches,
2158                        gfpflags, node);
2159
2160        if (!n) {
2161            free_kmem_cache_nodes(s);
2162            return 0;
2163        }
2164
2165        s->node[node] = n;
2166        init_kmem_cache_node(n, s);
2167    }
2168    return 1;
2169}
2170#else
2171static void free_kmem_cache_nodes(struct kmem_cache *s)
2172{
2173}
2174
2175static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
2176{
2177    init_kmem_cache_node(&s->local_node, s);
2178    return 1;
2179}
2180#endif
2181
2182static void set_min_partial(struct kmem_cache *s, unsigned long min)
2183{
2184    if (min < MIN_PARTIAL)
2185        min = MIN_PARTIAL;
2186    else if (min > MAX_PARTIAL)
2187        min = MAX_PARTIAL;
2188    s->min_partial = min;
2189}
2190
2191/*
2192 * calculate_sizes() determines the order and the distribution of data within
2193 * a slab object.
2194 */
2195static int calculate_sizes(struct kmem_cache *s, int forced_order)
2196{
2197    unsigned long flags = s->flags;
2198    unsigned long size = s->objsize;
2199    unsigned long align = s->align;
2200    int order;
2201
2202    /*
2203     * Round up object size to the next word boundary. We can only
2204     * place the free pointer at word boundaries and this determines
2205     * the possible location of the free pointer.
2206     */
2207    size = ALIGN(size, sizeof(void *));
2208
2209#ifdef CONFIG_SLUB_DEBUG
2210    /*
2211     * Determine if we can poison the object itself. If the user of
2212     * the slab may touch the object after free or before allocation
2213     * then we should never poison the object itself.
2214     */
2215    if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
2216            !s->ctor)
2217        s->flags |= __OBJECT_POISON;
2218    else
2219        s->flags &= ~__OBJECT_POISON;
2220
2221
2222    /*
2223     * If we are Redzoning then check if there is some space between the
2224     * end of the object and the free pointer. If not then add an
2225     * additional word to have some bytes to store Redzone information.
2226     */
2227    if ((flags & SLAB_RED_ZONE) && size == s->objsize)
2228        size += sizeof(void *);
2229#endif
2230
2231    /*
2232     * With that we have determined the number of bytes in actual use
2233     * by the object. This is the potential offset to the free pointer.
2234     */
2235    s->inuse = size;
2236
2237    if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
2238        s->ctor)) {
2239        /*
2240         * Relocate free pointer after the object if it is not
2241         * permitted to overwrite the first word of the object on
2242         * kmem_cache_free.
2243         *
2244         * This is the case if we do RCU, have a constructor or
2245         * destructor or are poisoning the objects.
2246         */
2247        s->offset = size;
2248        size += sizeof(void *);
2249    }
2250
2251#ifdef CONFIG_SLUB_DEBUG
2252    if (flags & SLAB_STORE_USER)
2253        /*
2254         * Need to store information about allocs and frees after
2255         * the object.
2256         */
2257        size += 2 * sizeof(struct track);
2258
2259    if (flags & SLAB_RED_ZONE)
2260        /*
2261         * Add some empty padding so that we can catch
2262         * overwrites from earlier objects rather than let
2263         * tracking information or the free pointer be
2264         * corrupted if a user writes before the start
2265         * of the object.
2266         */
2267        size += sizeof(void *);
2268#endif
2269
2270    /*
2271     * Determine the alignment based on various parameters that the
2272     * user specified and the dynamic determination of cache line size
2273     * on bootup.
2274     */
2275    align = calculate_alignment(flags, align, s->objsize);
2276    s->align = align;
2277
2278    /*
2279     * SLUB stores one object immediately after another beginning from
2280     * offset 0. In order to align the objects we have to simply size
2281     * each object to conform to the alignment.
2282     */
2283    size = ALIGN(size, align);
2284    s->size = size;
2285    if (forced_order >= 0)
2286        order = forced_order;
2287    else
2288        order = calculate_order(size);
2289
2290    if (order < 0)
2291        return 0;
2292
2293    s->allocflags = 0;
2294    if (order)
2295        s->allocflags |= __GFP_COMP;
2296
2297    if (s->flags & SLAB_CACHE_DMA)
2298        s->allocflags |= SLUB_DMA;
2299
2300    if (s->flags & SLAB_RECLAIM_ACCOUNT)
2301        s->allocflags |= __GFP_RECLAIMABLE;
2302
2303    /*
2304     * Determine the number of objects per slab
2305     */
2306    s->oo = oo_make(order, size);
2307    s->min = oo_make(get_order(size), size);
2308    if (oo_objects(s->oo) > oo_objects(s->max))
2309        s->max = s->oo;
2310
2311    return !!oo_objects(s->oo);
2312
2313}
2314
2315static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags,
2316        const char *name, size_t size,
2317        size_t align, unsigned long flags,
2318        void (*ctor)(void *))
2319{
2320    memset(s, 0, kmem_size);
2321    s->name = name;
2322    s->ctor = ctor;
2323    s->objsize = size;
2324    s->align = align;
2325    s->flags = kmem_cache_flags(size, flags, name, ctor);
2326
2327    if (!calculate_sizes(s, -1))
2328        goto error;
2329    if (disable_higher_order_debug) {
2330        /*
2331         * Disable debugging flags that store metadata if the min slab
2332         * order increased.
2333         */
2334        if (get_order(s->size) > get_order(s->objsize)) {
2335            s->flags &= ~DEBUG_METADATA_FLAGS;
2336            s->offset = 0;
2337            if (!calculate_sizes(s, -1))
2338                goto error;
2339        }
2340    }
2341
2342    /*
2343     * The larger the object size is, the more pages we want on the partial
2344     * list to avoid pounding the page allocator excessively.
2345     */
2346    set_min_partial(s, ilog2(s->size));
2347    s->refcount = 1;
2348#ifdef CONFIG_NUMA
2349    s->remote_node_defrag_ratio = 1000;
2350#endif
2351    if (!init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA))
2352        goto error;
2353
2354    if (alloc_kmem_cache_cpus(s, gfpflags & ~SLUB_DMA))
2355        return 1;
2356
2357    free_kmem_cache_nodes(s);
2358error:
2359    if (flags & SLAB_PANIC)
2360        panic("Cannot create slab %s size=%lu realsize=%u "
2361            "order=%u offset=%u flags=%lx\n",
2362            s->name, (unsigned long)size, s->size, oo_order(s->oo),
2363            s->offset, flags);
2364    return 0;
2365}
2366
2367/*
2368 * Check if a given pointer is valid
2369 */
2370int kmem_ptr_validate(struct kmem_cache *s, const void *object)
2371{
2372    struct page *page;
2373
2374    if (!kern_ptr_validate(object, s->size))
2375        return 0;
2376
2377    page = get_object_page(object);
2378
2379    if (!page || s != page->slab)
2380        /* No slab or wrong slab */
2381        return 0;
2382
2383    if (!check_valid_pointer(s, page, object))
2384        return 0;
2385
2386    /*
2387     * We could also check if the object is on the slabs freelist.
2388     * But this would be too expensive and it seems that the main
2389     * purpose of kmem_ptr_valid() is to check if the object belongs
2390     * to a certain slab.
2391     */
2392    return 1;
2393}
2394EXPORT_SYMBOL(kmem_ptr_validate);
2395
2396/*
2397 * Determine the size of a slab object
2398 */
2399unsigned int kmem_cache_size(struct kmem_cache *s)
2400{
2401    return s->objsize;
2402}
2403EXPORT_SYMBOL(kmem_cache_size);
2404
2405const char *kmem_cache_name(struct kmem_cache *s)
2406{
2407    return s->name;
2408}
2409EXPORT_SYMBOL(kmem_cache_name);
2410
2411static void list_slab_objects(struct kmem_cache *s, struct page *page,
2412                            const char *text)
2413{
2414#ifdef CONFIG_SLUB_DEBUG
2415    void *addr = page_address(page);
2416    void *p;
2417    long *map = kzalloc(BITS_TO_LONGS(page->objects) * sizeof(long),
2418                GFP_ATOMIC);
2419
2420    if (!map)
2421        return;
2422    slab_err(s, page, "%s", text);
2423    slab_lock(page);
2424    for_each_free_object(p, s, page->freelist)
2425        set_bit(slab_index(p, s, addr), map);
2426
2427    for_each_object(p, s, addr, page->objects) {
2428
2429        if (!test_bit(slab_index(p, s, addr), map)) {
2430            printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n",
2431                            p, p - addr);
2432            print_tracking(s, p);
2433        }
2434    }
2435    slab_unlock(page);
2436    kfree(map);
2437#endif
2438}
2439
2440/*
2441 * Attempt to free all partial slabs on a node.
2442 */
2443static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
2444{
2445    unsigned long flags;
2446    struct page *page, *h;
2447
2448    spin_lock_irqsave(&n->list_lock, flags);
2449    list_for_each_entry_safe(page, h, &n->partial, lru) {
2450        if (!page->inuse) {
2451            list_del(&page->lru);
2452            discard_slab(s, page);
2453            n->nr_partial--;
2454        } else {
2455            list_slab_objects(s, page,
2456                "Objects remaining on kmem_cache_close()");
2457        }
2458    }
2459    spin_unlock_irqrestore(&n->list_lock, flags);
2460}
2461
2462/*
2463 * Release all resources used by a slab cache.
2464 */
2465static inline int kmem_cache_close(struct kmem_cache *s)
2466{
2467    int node;
2468
2469    flush_all(s);
2470    free_percpu(s->cpu_slab);
2471    /* Attempt to free all objects */
2472    for_each_node_state(node, N_NORMAL_MEMORY) {
2473        struct kmem_cache_node *n = get_node(s, node);
2474
2475        free_partial(s, n);
2476        if (n->nr_partial || slabs_node(s, node))
2477            return 1;
2478    }
2479    free_kmem_cache_nodes(s);
2480    return 0;
2481}
2482
2483/*
2484 * Close a cache and release the kmem_cache structure
2485 * (must be used for caches created using kmem_cache_create)
2486 */
2487void kmem_cache_destroy(struct kmem_cache *s)
2488{
2489    down_write(&slub_lock);
2490    s->refcount--;
2491    if (!s->refcount) {
2492        list_del(&s->list);
2493        up_write(&slub_lock);
2494        if (kmem_cache_close(s)) {
2495            printk(KERN_ERR "SLUB %s: %s called for cache that "
2496                "still has objects.\n", s->name, __func__);
2497            dump_stack();
2498        }
2499        if (s->flags & SLAB_DESTROY_BY_RCU)
2500            rcu_barrier();
2501        sysfs_slab_remove(s);
2502    } else
2503        up_write(&slub_lock);
2504}
2505EXPORT_SYMBOL(kmem_cache_destroy);
2506
2507/********************************************************************
2508 * Kmalloc subsystem
2509 *******************************************************************/
2510
2511struct kmem_cache kmalloc_caches[KMALLOC_CACHES] __cacheline_aligned;
2512EXPORT_SYMBOL(kmalloc_caches);
2513
2514static int __init setup_slub_min_order(char *str)
2515{
2516    get_option(&str, &slub_min_order);
2517
2518    return 1;
2519}
2520
2521__setup("slub_min_order=", setup_slub_min_order);
2522
2523static int __init setup_slub_max_order(char *str)
2524{
2525    get_option(&str, &slub_max_order);
2526    slub_max_order = min(slub_max_order, MAX_ORDER - 1);
2527
2528    return 1;
2529}
2530
2531__setup("slub_max_order=", setup_slub_max_order);
2532
2533static int __init setup_slub_min_objects(char *str)
2534{
2535    get_option(&str, &slub_min_objects);
2536
2537    return 1;
2538}
2539
2540__setup("slub_min_objects=", setup_slub_min_objects);
2541
2542static int __init setup_slub_nomerge(char *str)
2543{
2544    slub_nomerge = 1;
2545    return 1;
2546}
2547
2548__setup("slub_nomerge", setup_slub_nomerge);
2549
2550static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s,
2551        const char *name, int size, gfp_t gfp_flags)
2552{
2553    unsigned int flags = 0;
2554
2555    if (gfp_flags & SLUB_DMA)
2556        flags = SLAB_CACHE_DMA;
2557
2558    /*
2559     * This function is called with IRQs disabled during early-boot on
2560     * single CPU so there's no need to take slub_lock here.
2561     */
2562    if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN,
2563                                flags, NULL))
2564        goto panic;
2565
2566    list_add(&s->list, &slab_caches);
2567
2568    if (sysfs_slab_add(s))
2569        goto panic;
2570    return s;
2571
2572panic:
2573    panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
2574}
2575
2576#ifdef CONFIG_ZONE_DMA
2577static struct kmem_cache *kmalloc_caches_dma[SLUB_PAGE_SHIFT];
2578
2579static void sysfs_add_func(struct work_struct *w)
2580{
2581    struct kmem_cache *s;
2582
2583    down_write(&slub_lock);
2584    list_for_each_entry(s, &slab_caches, list) {
2585        if (s->flags & __SYSFS_ADD_DEFERRED) {
2586            s->flags &= ~__SYSFS_ADD_DEFERRED;
2587            sysfs_slab_add(s);
2588        }
2589    }
2590    up_write(&slub_lock);
2591}
2592
2593static DECLARE_WORK(sysfs_add_work, sysfs_add_func);
2594
2595static noinline struct kmem_cache *dma_kmalloc_cache(int index, gfp_t flags)
2596{
2597    struct kmem_cache *s;
2598    char *text;
2599    size_t realsize;
2600    unsigned long slabflags;
2601    int i;
2602
2603    s = kmalloc_caches_dma[index];
2604    if (s)
2605        return s;
2606
2607    /* Dynamically create dma cache */
2608    if (flags & __GFP_WAIT)
2609        down_write(&slub_lock);
2610    else {
2611        if (!down_write_trylock(&slub_lock))
2612            goto out;
2613    }
2614
2615    if (kmalloc_caches_dma[index])
2616        goto unlock_out;
2617
2618    realsize = kmalloc_caches[index].objsize;
2619    text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d",
2620             (unsigned int)realsize);
2621
2622    s = NULL;
2623    for (i = 0; i < KMALLOC_CACHES; i++)
2624        if (!kmalloc_caches[i].size)
2625            break;
2626
2627    BUG_ON(i >= KMALLOC_CACHES);
2628    s = kmalloc_caches + i;
2629
2630    /*
2631     * Must defer sysfs creation to a workqueue because we don't know
2632     * what context we are called from. Before sysfs comes up, we don't
2633     * need to do anything because our sysfs initcall will start by
2634     * adding all existing slabs to sysfs.
2635     */
2636    slabflags = SLAB_CACHE_DMA|SLAB_NOTRACK;
2637    if (slab_state >= SYSFS)
2638        slabflags |= __SYSFS_ADD_DEFERRED;
2639
2640    if (!text || !kmem_cache_open(s, flags, text,
2641            realsize, ARCH_KMALLOC_MINALIGN, slabflags, NULL)) {
2642        s->size = 0;
2643        kfree(text);
2644        goto unlock_out;
2645    }
2646
2647    list_add(&s->list, &slab_caches);
2648    kmalloc_caches_dma[index] = s;
2649
2650    if (slab_state >= SYSFS)
2651        schedule_work(&sysfs_add_work);
2652
2653unlock_out:
2654    up_write(&slub_lock);
2655out:
2656    return kmalloc_caches_dma[index];
2657}
2658#endif
2659
2660/*
2661 * Conversion table for small slabs sizes / 8 to the index in the
2662 * kmalloc array. This is necessary for slabs < 192 since we have non power
2663 * of two cache sizes there. The size of larger slabs can be determined using
2664 * fls.
2665 */
2666static s8 size_index[24] = {
2667    3, /* 8 */
2668    4, /* 16 */
2669    5, /* 24 */
2670    5, /* 32 */
2671    6, /* 40 */
2672    6, /* 48 */
2673    6, /* 56 */
2674    6, /* 64 */
2675    1, /* 72 */
2676    1, /* 80 */
2677    1, /* 88 */
2678    1, /* 96 */
2679    7, /* 104 */
2680    7, /* 112 */
2681    7, /* 120 */
2682    7, /* 128 */
2683    2, /* 136 */
2684    2, /* 144 */
2685    2, /* 152 */
2686    2, /* 160 */
2687    2, /* 168 */
2688    2, /* 176 */
2689    2, /* 184 */
2690    2 /* 192 */
2691};
2692
2693static inline int size_index_elem(size_t bytes)
2694{
2695    return (bytes - 1) / 8;
2696}
2697
2698static struct kmem_cache *get_slab(size_t size, gfp_t flags)
2699{
2700    int index;
2701
2702    if (size <= 192) {
2703        if (!size)
2704            return ZERO_SIZE_PTR;
2705
2706        index = size_index[size_index_elem(size)];
2707    } else
2708        index = fls(size - 1);
2709
2710#ifdef CONFIG_ZONE_DMA
2711    if (unlikely((flags & SLUB_DMA)))
2712        return dma_kmalloc_cache(index, flags);
2713
2714#endif
2715    return &kmalloc_caches[index];
2716}
2717
2718void *__kmalloc(size_t size, gfp_t flags)
2719{
2720    struct kmem_cache *s;
2721    void *ret;
2722
2723    if (unlikely(size > SLUB_MAX_SIZE))
2724        return kmalloc_large(size, flags);
2725
2726    s = get_slab(size, flags);
2727
2728    if (unlikely(ZERO_OR_NULL_PTR(s)))
2729        return s;
2730
2731    ret = slab_alloc(s, flags, -1, _RET_IP_);
2732
2733    trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
2734
2735    return ret;
2736}
2737EXPORT_SYMBOL(__kmalloc);
2738
2739static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
2740{
2741    struct page *page;
2742    void *ptr = NULL;
2743
2744    flags |= __GFP_COMP | __GFP_NOTRACK;
2745    page = alloc_pages_node(node, flags, get_order(size));
2746    if (page)
2747        ptr = page_address(page);
2748
2749    kmemleak_alloc(ptr, size, 1, flags);
2750    return ptr;
2751}
2752
2753#ifdef CONFIG_NUMA
2754void *__kmalloc_node(size_t size, gfp_t flags, int node)
2755{
2756    struct kmem_cache *s;
2757    void *ret;
2758
2759    if (unlikely(size > SLUB_MAX_SIZE)) {
2760        ret = kmalloc_large_node(size, flags, node);
2761
2762        trace_kmalloc_node(_RET_IP_, ret,
2763                   size, PAGE_SIZE << get_order(size),
2764                   flags, node);
2765
2766        return ret;
2767    }
2768
2769    s = get_slab(size, flags);
2770
2771    if (unlikely(ZERO_OR_NULL_PTR(s)))
2772        return s;
2773
2774    ret = slab_alloc(s, flags, node, _RET_IP_);
2775
2776    trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
2777
2778    return ret;
2779}
2780EXPORT_SYMBOL(__kmalloc_node);
2781#endif
2782
2783size_t ksize(const void *object)
2784{
2785    struct page *page;
2786    struct kmem_cache *s;
2787
2788    if (unlikely(object == ZERO_SIZE_PTR))
2789        return 0;
2790
2791    page = virt_to_head_page(object);
2792
2793    if (unlikely(!PageSlab(page))) {
2794        WARN_ON(!PageCompound(page));
2795        return PAGE_SIZE << compound_order(page);
2796    }
2797    s = page->slab;
2798
2799#ifdef CONFIG_SLUB_DEBUG
2800    /*
2801     * Debugging requires use of the padding between object
2802     * and whatever may come after it.
2803     */
2804    if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
2805        return s->objsize;
2806
2807#endif
2808    /*
2809     * If we have the need to store the freelist pointer
2810     * back there or track user information then we can
2811     * only use the space before that information.
2812     */
2813    if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
2814        return s->inuse;
2815    /*
2816     * Else we can use all the padding etc for the allocation
2817     */
2818    return s->size;
2819}
2820EXPORT_SYMBOL(ksize);
2821
2822void kfree(const void *x)
2823{
2824    struct page *page;
2825    void *object = (void *)x;
2826
2827    trace_kfree(_RET_IP_, x);
2828
2829    if (unlikely(ZERO_OR_NULL_PTR(x)))
2830        return;
2831
2832    page = virt_to_head_page(x);
2833    if (unlikely(!PageSlab(page))) {
2834        BUG_ON(!PageCompound(page));
2835        kmemleak_free(x);
2836        put_page(page);
2837        return;
2838    }
2839    slab_free(page->slab, page, object, _RET_IP_);
2840}
2841EXPORT_SYMBOL(kfree);
2842
2843/*
2844 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
2845 * the remaining slabs by the number of items in use. The slabs with the
2846 * most items in use come first. New allocations will then fill those up
2847 * and thus they can be removed from the partial lists.
2848 *
2849 * The slabs with the least items are placed last. This results in them
2850 * being allocated from last increasing the chance that the last objects
2851 * are freed in them.
2852 */
2853int kmem_cache_shrink(struct kmem_cache *s)
2854{
2855    int node;
2856    int i;
2857    struct kmem_cache_node *n;
2858    struct page *page;
2859    struct page *t;
2860    int objects = oo_objects(s->max);
2861    struct list_head *slabs_by_inuse =
2862        kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
2863    unsigned long flags;
2864
2865    if (!slabs_by_inuse)
2866        return -ENOMEM;
2867
2868    flush_all(s);
2869    for_each_node_state(node, N_NORMAL_MEMORY) {
2870        n = get_node(s, node);
2871
2872        if (!n->nr_partial)
2873            continue;
2874
2875        for (i = 0; i < objects; i++)
2876            INIT_LIST_HEAD(slabs_by_inuse + i);
2877
2878        spin_lock_irqsave(&n->list_lock, flags);
2879
2880        /*
2881         * Build lists indexed by the items in use in each slab.
2882         *
2883         * Note that concurrent frees may occur while we hold the
2884         * list_lock. page->inuse here is the upper limit.
2885         */
2886        list_for_each_entry_safe(page, t, &n->partial, lru) {
2887            if (!page->inuse && slab_trylock(page)) {
2888                /*
2889                 * Must hold slab lock here because slab_free
2890                 * may have freed the last object and be
2891                 * waiting to release the slab.
2892                 */
2893                list_del(&page->lru);
2894                n->nr_partial--;
2895                slab_unlock(page);
2896                discard_slab(s, page);
2897            } else {
2898                list_move(&page->lru,
2899                slabs_by_inuse + page->inuse);
2900            }
2901        }
2902
2903        /*
2904         * Rebuild the partial list with the slabs filled up most
2905         * first and the least used slabs at the end.
2906         */
2907        for (i = objects - 1; i >= 0; i--)
2908            list_splice(slabs_by_inuse + i, n->partial.prev);
2909
2910        spin_unlock_irqrestore(&n->list_lock, flags);
2911    }
2912
2913    kfree(slabs_by_inuse);
2914    return 0;
2915}
2916EXPORT_SYMBOL(kmem_cache_shrink);
2917
2918#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
2919static int slab_mem_going_offline_callback(void *arg)
2920{
2921    struct kmem_cache *s;
2922
2923    down_read(&slub_lock);
2924    list_for_each_entry(s, &slab_caches, list)
2925        kmem_cache_shrink(s);
2926    up_read(&slub_lock);
2927
2928    return 0;
2929}
2930
2931static void slab_mem_offline_callback(void *arg)
2932{
2933    struct kmem_cache_node *n;
2934    struct kmem_cache *s;
2935    struct memory_notify *marg = arg;
2936    int offline_node;
2937
2938    offline_node = marg->status_change_nid;
2939
2940    /*
2941     * If the node still has available memory. we need kmem_cache_node
2942     * for it yet.
2943     */
2944    if (offline_node < 0)
2945        return;
2946
2947    down_read(&slub_lock);
2948    list_for_each_entry(s, &slab_caches, list) {
2949        n = get_node(s, offline_node);
2950        if (n) {
2951            /*
2952             * if n->nr_slabs > 0, slabs still exist on the node
2953             * that is going down. We were unable to free them,
2954             * and offline_pages() function shouldn't call this
2955             * callback. So, we must fail.
2956             */
2957            BUG_ON(slabs_node(s, offline_node));
2958
2959            s->node[offline_node] = NULL;
2960            kmem_cache_free(kmalloc_caches, n);
2961        }
2962    }
2963    up_read(&slub_lock);
2964}
2965
2966static int slab_mem_going_online_callback(void *arg)
2967{
2968    struct kmem_cache_node *n;
2969    struct kmem_cache *s;
2970    struct memory_notify *marg = arg;
2971    int nid = marg->status_change_nid;
2972    int ret = 0;
2973
2974    /*
2975     * If the node's memory is already available, then kmem_cache_node is
2976     * already created. Nothing to do.
2977     */
2978    if (nid < 0)
2979        return 0;
2980
2981    /*
2982     * We are bringing a node online. No memory is available yet. We must
2983     * allocate a kmem_cache_node structure in order to bring the node
2984     * online.
2985     */
2986    down_read(&slub_lock);
2987    list_for_each_entry(s, &slab_caches, list) {
2988        /*
2989         * XXX: kmem_cache_alloc_node will fallback to other nodes
2990         * since memory is not yet available from the node that
2991         * is brought up.
2992         */
2993        n = kmem_cache_alloc(kmalloc_caches, GFP_KERNEL);
2994        if (!n) {
2995            ret = -ENOMEM;
2996            goto out;
2997        }
2998        init_kmem_cache_node(n, s);
2999        s->node[nid] = n;
3000    }
3001out:
3002    up_read(&slub_lock);
3003    return ret;
3004}
3005
3006static int slab_memory_callback(struct notifier_block *self,
3007                unsigned long action, void *arg)
3008{
3009    int ret = 0;
3010
3011    switch (action) {
3012    case MEM_GOING_ONLINE:
3013        ret = slab_mem_going_online_callback(arg);
3014        break;
3015    case MEM_GOING_OFFLINE:
3016        ret = slab_mem_going_offline_callback(arg);
3017        break;
3018    case MEM_OFFLINE:
3019    case MEM_CANCEL_ONLINE:
3020        slab_mem_offline_callback(arg);
3021        break;
3022    case MEM_ONLINE:
3023    case MEM_CANCEL_OFFLINE:
3024        break;
3025    }
3026    if (ret)
3027        ret = notifier_from_errno(ret);
3028    else
3029        ret = NOTIFY_OK;
3030    return ret;
3031}
3032
3033#endif /* CONFIG_MEMORY_HOTPLUG */
3034
3035/********************************************************************
3036 * Basic setup of slabs
3037 *******************************************************************/
3038
3039void __init kmem_cache_init(void)
3040{
3041    int i;
3042    int caches = 0;
3043
3044#ifdef CONFIG_NUMA
3045    /*
3046     * Must first have the slab cache available for the allocations of the
3047     * struct kmem_cache_node's. There is special bootstrap code in
3048     * kmem_cache_open for slab_state == DOWN.
3049     */
3050    create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node",
3051        sizeof(struct kmem_cache_node), GFP_NOWAIT);
3052    kmalloc_caches[0].refcount = -1;
3053    caches++;
3054
3055    hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
3056#endif
3057
3058    /* Able to allocate the per node structures */
3059    slab_state = PARTIAL;
3060
3061    /* Caches that are not of the two-to-the-power-of size */
3062    if (KMALLOC_MIN_SIZE <= 32) {
3063        create_kmalloc_cache(&kmalloc_caches[1],
3064                "kmalloc-96", 96, GFP_NOWAIT);
3065        caches++;
3066    }
3067    if (KMALLOC_MIN_SIZE <= 64) {
3068        create_kmalloc_cache(&kmalloc_caches[2],
3069                "kmalloc-192", 192, GFP_NOWAIT);
3070        caches++;
3071    }
3072
3073    for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++) {
3074        create_kmalloc_cache(&kmalloc_caches[i],
3075            "kmalloc", 1 << i, GFP_NOWAIT);
3076        caches++;
3077    }
3078
3079
3080    /*
3081     * Patch up the size_index table if we have strange large alignment
3082     * requirements for the kmalloc array. This is only the case for
3083     * MIPS it seems. The standard arches will not generate any code here.
3084     *
3085     * Largest permitted alignment is 256 bytes due to the way we
3086     * handle the index determination for the smaller caches.
3087     *
3088     * Make sure that nothing crazy happens if someone starts tinkering
3089     * around with ARCH_KMALLOC_MINALIGN
3090     */
3091    BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
3092        (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
3093
3094    for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
3095        int elem = size_index_elem(i);
3096        if (elem >= ARRAY_SIZE(size_index))
3097            break;
3098        size_index[elem] = KMALLOC_SHIFT_LOW;
3099    }
3100
3101    if (KMALLOC_MIN_SIZE == 64) {
3102        /*
3103         * The 96 byte size cache is not used if the alignment
3104         * is 64 byte.
3105         */
3106        for (i = 64 + 8; i <= 96; i += 8)
3107            size_index[size_index_elem(i)] = 7;
3108    } else if (KMALLOC_MIN_SIZE == 128) {
3109        /*
3110         * The 192 byte sized cache is not used if the alignment
3111         * is 128 byte. Redirect kmalloc to use the 256 byte cache
3112         * instead.
3113         */
3114        for (i = 128 + 8; i <= 192; i += 8)
3115            size_index[size_index_elem(i)] = 8;
3116    }
3117
3118    slab_state = UP;
3119
3120    /* Provide the correct kmalloc names now that the caches are up */
3121    for (i = KMALLOC_SHIFT_LOW; i < SLUB_PAGE_SHIFT; i++)
3122        kmalloc_caches[i]. name =
3123            kasprintf(GFP_NOWAIT, "kmalloc-%d", 1 << i);
3124
3125#ifdef CONFIG_SMP
3126    register_cpu_notifier(&slab_notifier);
3127#endif
3128#ifdef CONFIG_NUMA
3129    kmem_size = offsetof(struct kmem_cache, node) +
3130                nr_node_ids * sizeof(struct kmem_cache_node *);
3131#else
3132    kmem_size = sizeof(struct kmem_cache);
3133#endif
3134
3135    printk(KERN_INFO
3136        "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
3137        " CPUs=%d, Nodes=%d\n",
3138        caches, cache_line_size(),
3139        slub_min_order, slub_max_order, slub_min_objects,
3140        nr_cpu_ids, nr_node_ids);
3141}
3142
3143void __init kmem_cache_init_late(void)
3144{
3145}
3146
3147/*
3148 * Find a mergeable slab cache
3149 */
3150static int slab_unmergeable(struct kmem_cache *s)
3151{
3152    if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3153        return 1;
3154
3155    if (s->ctor)
3156        return 1;
3157
3158    /*
3159     * We may have set a slab to be unmergeable during bootstrap.
3160     */
3161    if (s->refcount < 0)
3162        return 1;
3163
3164    return 0;
3165}
3166
3167static struct kmem_cache *find_mergeable(size_t size,
3168        size_t align, unsigned long flags, const char *name,
3169        void (*ctor)(void *))
3170{
3171    struct kmem_cache *s;
3172
3173    if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3174        return NULL;
3175
3176    if (ctor)
3177        return NULL;
3178
3179    size = ALIGN(size, sizeof(void *));
3180    align = calculate_alignment(flags, align, size);
3181    size = ALIGN(size, align);
3182    flags = kmem_cache_flags(size, flags, name, NULL);
3183
3184    list_for_each_entry(s, &slab_caches, list) {
3185        if (slab_unmergeable(s))
3186            continue;
3187
3188        if (size > s->size)
3189            continue;
3190
3191        if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
3192                continue;
3193        /*
3194         * Check if alignment is compatible.
3195         * Courtesy of Adrian Drzewiecki
3196         */
3197        if ((s->size & ~(align - 1)) != s->size)
3198            continue;
3199
3200        if (s->size - size >= sizeof(void *))
3201            continue;
3202
3203        return s;
3204    }
3205    return NULL;
3206}
3207
3208struct kmem_cache *kmem_cache_create(const char *name, size_t size,
3209        size_t align, unsigned long flags, void (*ctor)(void *))
3210{
3211    struct kmem_cache *s;
3212
3213    if (WARN_ON(!name))
3214        return NULL;
3215
3216    down_write(&slub_lock);
3217    s = find_mergeable(size, align, flags, name, ctor);
3218    if (s) {
3219        s->refcount++;
3220        /*
3221         * Adjust the object sizes so that we clear
3222         * the complete object on kzalloc.
3223         */
3224        s->objsize = max(s->objsize, (int)size);
3225        s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
3226        up_write(&slub_lock);
3227
3228        if (sysfs_slab_alias(s, name)) {
3229            down_write(&slub_lock);
3230            s->refcount--;
3231            up_write(&slub_lock);
3232            goto err;
3233        }
3234        return s;
3235    }
3236
3237    s = kmalloc(kmem_size, GFP_KERNEL);
3238    if (s) {
3239        if (kmem_cache_open(s, GFP_KERNEL, name,
3240                size, align, flags, ctor)) {
3241            list_add(&s->list, &slab_caches);
3242            up_write(&slub_lock);
3243            if (sysfs_slab_add(s)) {
3244                down_write(&slub_lock);
3245                list_del(&s->list);
3246                up_write(&slub_lock);
3247                kfree(s);
3248                goto err;
3249            }
3250            return s;
3251        }
3252        kfree(s);
3253    }
3254    up_write(&slub_lock);
3255
3256err:
3257    if (flags & SLAB_PANIC)
3258        panic("Cannot create slabcache %s\n", name);
3259    else
3260        s = NULL;
3261    return s;
3262}
3263EXPORT_SYMBOL(kmem_cache_create);
3264
3265#ifdef CONFIG_SMP
3266/*
3267 * Use the cpu notifier to insure that the cpu slabs are flushed when
3268 * necessary.
3269 */
3270static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
3271        unsigned long action, void *hcpu)
3272{
3273    long cpu = (long)hcpu;
3274    struct kmem_cache *s;
3275    unsigned long flags;
3276
3277    switch (action) {
3278    case CPU_UP_CANCELED:
3279    case CPU_UP_CANCELED_FROZEN:
3280    case CPU_DEAD:
3281    case CPU_DEAD_FROZEN:
3282        down_read(&slub_lock);
3283        list_for_each_entry(s, &slab_caches, list) {
3284            local_irq_save(flags);
3285            __flush_cpu_slab(s, cpu);
3286            local_irq_restore(flags);
3287        }
3288        up_read(&slub_lock);
3289        break;
3290    default:
3291        break;
3292    }
3293    return NOTIFY_OK;
3294}
3295
3296static struct notifier_block __cpuinitdata slab_notifier = {
3297    .notifier_call = slab_cpuup_callback
3298};
3299
3300#endif
3301
3302void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
3303{
3304    struct kmem_cache *s;
3305    void *ret;
3306
3307    if (unlikely(size > SLUB_MAX_SIZE))
3308        return kmalloc_large(size, gfpflags);
3309
3310    s = get_slab(size, gfpflags);
3311
3312    if (unlikely(ZERO_OR_NULL_PTR(s)))
3313        return s;
3314
3315    ret = slab_alloc(s, gfpflags, -1, caller);
3316
3317    /* Honor the call site pointer we recieved. */
3318    trace_kmalloc(caller, ret, size, s->size, gfpflags);
3319
3320    return ret;
3321}
3322
3323void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
3324                    int node, unsigned long caller)
3325{
3326    struct kmem_cache *s;
3327    void *ret;
3328
3329    if (unlikely(size > SLUB_MAX_SIZE)) {
3330        ret = kmalloc_large_node(size, gfpflags, node);
3331
3332        trace_kmalloc_node(caller, ret,
3333                   size, PAGE_SIZE << get_order(size),
3334                   gfpflags, node);
3335
3336        return ret;
3337    }
3338
3339    s = get_slab(size, gfpflags);
3340
3341    if (unlikely(ZERO_OR_NULL_PTR(s)))
3342        return s;
3343
3344    ret = slab_alloc(s, gfpflags, node, caller);
3345
3346    /* Honor the call site pointer we recieved. */
3347    trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
3348
3349    return ret;
3350}
3351
3352#ifdef CONFIG_SLUB_DEBUG
3353static int count_inuse(struct page *page)
3354{
3355    return page->inuse;
3356}
3357
3358static int count_total(struct page *page)
3359{
3360    return page->objects;
3361}
3362
3363static int validate_slab(struct kmem_cache *s, struct page *page,
3364                        unsigned long *map)
3365{
3366    void *p;
3367    void *addr = page_address(page);
3368
3369    if (!check_slab(s, page) ||
3370            !on_freelist(s, page, NULL))
3371        return 0;
3372
3373    /* Now we know that a valid freelist exists */
3374    bitmap_zero(map, page->objects);
3375
3376    for_each_free_object(p, s, page->freelist) {
3377        set_bit(slab_index(p, s, addr), map);
3378        if (!check_object(s, page, p, 0))
3379            return 0;
3380    }
3381
3382    for_each_object(p, s, addr, page->objects)
3383        if (!test_bit(slab_index(p, s, addr), map))
3384            if (!check_object(s, page, p, 1))
3385                return 0;
3386    return 1;
3387}
3388
3389static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3390                        unsigned long *map)
3391{
3392    if (slab_trylock(page)) {
3393        validate_slab(s, page, map);
3394        slab_unlock(page);
3395    } else
3396        printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
3397            s->name, page);
3398
3399    if (s->flags & DEBUG_DEFAULT_FLAGS) {
3400        if (!PageSlubDebug(page))
3401            printk(KERN_ERR "SLUB %s: SlubDebug not set "
3402                "on slab 0x%p\n", s->name, page);
3403    } else {
3404        if (PageSlubDebug(page))
3405            printk(KERN_ERR "SLUB %s: SlubDebug set on "
3406                "slab 0x%p\n", s->name, page);
3407    }
3408}
3409
3410static int validate_slab_node(struct kmem_cache *s,
3411        struct kmem_cache_node *n, unsigned long *map)
3412{
3413    unsigned long count = 0;
3414    struct page *page;
3415    unsigned long flags;
3416
3417    spin_lock_irqsave(&n->list_lock, flags);
3418
3419    list_for_each_entry(page, &n->partial, lru) {
3420        validate_slab_slab(s, page, map);
3421        count++;
3422    }
3423    if (count != n->nr_partial)
3424        printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
3425            "counter=%ld\n", s->name, count, n->nr_partial);
3426
3427    if (!(s->flags & SLAB_STORE_USER))
3428        goto out;
3429
3430    list_for_each_entry(page, &n->full, lru) {
3431        validate_slab_slab(s, page, map);
3432        count++;
3433    }
3434    if (count != atomic_long_read(&n->nr_slabs))
3435        printk(KERN_ERR "SLUB: %s %ld slabs counted but "
3436            "counter=%ld\n", s->name, count,
3437            atomic_long_read(&n->nr_slabs));
3438
3439out:
3440    spin_unlock_irqrestore(&n->list_lock, flags);
3441    return count;
3442}
3443
3444static long validate_slab_cache(struct kmem_cache *s)
3445{
3446    int node;
3447    unsigned long count = 0;
3448    unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3449                sizeof(unsigned long), GFP_KERNEL);
3450
3451    if (!map)
3452        return -ENOMEM;
3453
3454    flush_all(s);
3455    for_each_node_state(node, N_NORMAL_MEMORY) {
3456        struct kmem_cache_node *n = get_node(s, node);
3457
3458        count += validate_slab_node(s, n, map);
3459    }
3460    kfree(map);
3461    return count;
3462}
3463
3464#ifdef SLUB_RESILIENCY_TEST
3465static void resiliency_test(void)
3466{
3467    u8 *p;
3468
3469    printk(KERN_ERR "SLUB resiliency testing\n");
3470    printk(KERN_ERR "-----------------------\n");
3471    printk(KERN_ERR "A. Corruption after allocation\n");
3472
3473    p = kzalloc(16, GFP_KERNEL);
3474    p[16] = 0x12;
3475    printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
3476            " 0x12->0x%p\n\n", p + 16);
3477
3478    validate_slab_cache(kmalloc_caches + 4);
3479
3480    /* Hmmm... The next two are dangerous */
3481    p = kzalloc(32, GFP_KERNEL);
3482    p[32 + sizeof(void *)] = 0x34;
3483    printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
3484            " 0x34 -> -0x%p\n", p);
3485    printk(KERN_ERR
3486        "If allocated object is overwritten then not detectable\n\n");
3487
3488    validate_slab_cache(kmalloc_caches + 5);
3489    p = kzalloc(64, GFP_KERNEL);
3490    p += 64 + (get_cycles() & 0xff) * sizeof(void *);
3491    *p = 0x56;
3492    printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
3493                                    p);
3494    printk(KERN_ERR
3495        "If allocated object is overwritten then not detectable\n\n");
3496    validate_slab_cache(kmalloc_caches + 6);
3497
3498    printk(KERN_ERR "\nB. Corruption after free\n");
3499    p = kzalloc(128, GFP_KERNEL);
3500    kfree(p);
3501    *p = 0x78;
3502    printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
3503    validate_slab_cache(kmalloc_caches + 7);
3504
3505    p = kzalloc(256, GFP_KERNEL);
3506    kfree(p);
3507    p[50] = 0x9a;
3508    printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n",
3509            p);
3510    validate_slab_cache(kmalloc_caches + 8);
3511
3512    p = kzalloc(512, GFP_KERNEL);
3513    kfree(p);
3514    p[512] = 0xab;
3515    printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
3516    validate_slab_cache(kmalloc_caches + 9);
3517}
3518#else
3519static void resiliency_test(void) {};
3520#endif
3521
3522/*
3523 * Generate lists of code addresses where slabcache objects are allocated
3524 * and freed.
3525 */
3526
3527struct location {
3528    unsigned long count;
3529    unsigned long addr;
3530    long long sum_time;
3531    long min_time;
3532    long max_time;
3533    long min_pid;
3534    long max_pid;
3535    DECLARE_BITMAP(cpus, NR_CPUS);
3536    nodemask_t nodes;
3537};
3538
3539struct loc_track {
3540    unsigned long max;
3541    unsigned long count;
3542    struct location *loc;
3543};
3544
3545static void free_loc_track(struct loc_track *t)
3546{
3547    if (t->max)
3548        free_pages((unsigned long)t->loc,
3549            get_order(sizeof(struct location) * t->max));
3550}
3551
3552static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
3553{
3554    struct location *l;
3555    int order;
3556
3557    order = get_order(sizeof(struct location) * max);
3558
3559    l = (void *)__get_free_pages(flags, order);
3560    if (!l)
3561        return 0;
3562
3563    if (t->count) {
3564        memcpy(l, t->loc, sizeof(struct location) * t->count);
3565        free_loc_track(t);
3566    }
3567    t->max = max;
3568    t->loc = l;
3569    return 1;
3570}
3571
3572static int add_location(struct loc_track *t, struct kmem_cache *s,
3573                const struct track *track)
3574{
3575    long start, end, pos;
3576    struct location *l;
3577    unsigned long caddr;
3578    unsigned long age = jiffies - track->when;
3579
3580    start = -1;
3581    end = t->count;
3582
3583    for ( ; ; ) {
3584        pos = start + (end - start + 1) / 2;
3585
3586        /*
3587         * There is nothing at "end". If we end up there
3588         * we need to add something to before end.
3589         */
3590        if (pos == end)
3591            break;
3592
3593        caddr = t->loc[pos].addr;
3594        if (track->addr == caddr) {
3595
3596            l = &t->loc[pos];
3597            l->count++;
3598            if (track->when) {
3599                l->sum_time += age;
3600                if (age < l->min_time)
3601                    l->min_time = age;
3602                if (age > l->max_time)
3603                    l->max_time = age;
3604
3605                if (track->pid < l->min_pid)
3606                    l->min_pid = track->pid;
3607                if (track->pid > l->max_pid)
3608                    l->max_pid = track->pid;
3609
3610                cpumask_set_cpu(track->cpu,
3611                        to_cpumask(l->cpus));
3612            }
3613            node_set(page_to_nid(virt_to_page(track)), l->nodes);
3614            return 1;
3615        }
3616
3617        if (track->addr < caddr)
3618            end = pos;
3619        else
3620            start = pos;
3621    }
3622
3623    /*
3624     * Not found. Insert new tracking element.
3625     */
3626    if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
3627        return 0;
3628
3629    l = t->loc + pos;
3630    if (pos < t->count)
3631        memmove(l + 1, l,
3632            (t->count - pos) * sizeof(struct location));
3633    t->count++;
3634    l->count = 1;
3635    l->addr = track->addr;
3636    l->sum_time = age;
3637    l->min_time = age;
3638    l->max_time = age;
3639    l->min_pid = track->pid;
3640    l->max_pid = track->pid;
3641    cpumask_clear(to_cpumask(l->cpus));
3642    cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
3643    nodes_clear(l->nodes);
3644    node_set(page_to_nid(virt_to_page(track)), l->nodes);
3645    return 1;
3646}
3647
3648static void process_slab(struct loc_track *t, struct kmem_cache *s,
3649        struct page *page, enum track_item alloc,
3650        long *map)
3651{
3652    void *addr = page_address(page);
3653    void *p;
3654
3655    bitmap_zero(map, page->objects);
3656    for_each_free_object(p, s, page->freelist)
3657        set_bit(slab_index(p, s, addr), map);
3658
3659    for_each_object(p, s, addr, page->objects)
3660        if (!test_bit(slab_index(p, s, addr), map))
3661            add_location(t, s, get_track(s, p, alloc));
3662}
3663
3664static int list_locations(struct kmem_cache *s, char *buf,
3665                    enum track_item alloc)
3666{
3667    int len = 0;
3668    unsigned long i;
3669    struct loc_track t = { 0, 0, NULL };
3670    int node;
3671    unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
3672                     sizeof(unsigned long), GFP_KERNEL);
3673
3674    if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
3675                     GFP_TEMPORARY)) {
3676        kfree(map);
3677        return sprintf(buf, "Out of memory\n");
3678    }
3679    /* Push back cpu slabs */
3680    flush_all(s);
3681
3682    for_each_node_state(node, N_NORMAL_MEMORY) {
3683        struct kmem_cache_node *n = get_node(s, node);
3684        unsigned long flags;
3685        struct page *page;
3686
3687        if (!atomic_long_read(&n->nr_slabs))
3688            continue;
3689
3690        spin_lock_irqsave(&n->list_lock, flags);
3691        list_for_each_entry(page, &n->partial, lru)
3692            process_slab(&t, s, page, alloc, map);
3693        list_for_each_entry(page, &n->full, lru)
3694            process_slab(&t, s, page, alloc, map);
3695        spin_unlock_irqrestore(&n->list_lock, flags);
3696    }
3697
3698    for (i = 0; i < t.count; i++) {
3699        struct location *l = &t.loc[i];
3700
3701        if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
3702            break;
3703        len += sprintf(buf + len, "%7ld ", l->count);
3704
3705        if (l->addr)
3706            len += sprint_symbol(buf + len, (unsigned long)l->addr);
3707        else
3708            len += sprintf(buf + len, "<not-available>");
3709
3710        if (l->sum_time != l->min_time) {
3711            len += sprintf(buf + len, " age=%ld/%ld/%ld",
3712                l->min_time,
3713                (long)div_u64(l->sum_time, l->count),
3714                l->max_time);
3715        } else
3716            len += sprintf(buf + len, " age=%ld",
3717                l->min_time);
3718
3719        if (l->min_pid != l->max_pid)
3720            len += sprintf(buf + len, " pid=%ld-%ld",
3721                l->min_pid, l->max_pid);
3722        else
3723            len += sprintf(buf + len, " pid=%ld",
3724                l->min_pid);
3725
3726        if (num_online_cpus() > 1 &&
3727                !cpumask_empty(to_cpumask(l->cpus)) &&
3728                len < PAGE_SIZE - 60) {
3729            len += sprintf(buf + len, " cpus=");
3730            len += cpulist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3731                         to_cpumask(l->cpus));
3732        }
3733
3734        if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
3735                len < PAGE_SIZE - 60) {
3736            len += sprintf(buf + len, " nodes=");
3737            len += nodelist_scnprintf(buf + len, PAGE_SIZE - len - 50,
3738                    l->nodes);
3739        }
3740
3741        len += sprintf(buf + len, "\n");
3742    }
3743
3744    free_loc_track(&t);
3745    kfree(map);
3746    if (!t.count)
3747        len += sprintf(buf, "No data\n");
3748    return len;
3749}
3750
3751enum slab_stat_type {
3752    SL_ALL, /* All slabs */
3753    SL_PARTIAL, /* Only partially allocated slabs */
3754    SL_CPU, /* Only slabs used for cpu caches */
3755    SL_OBJECTS, /* Determine allocated objects not slabs */
3756    SL_TOTAL /* Determine object capacity not slabs */
3757};
3758
3759#define SO_ALL (1 << SL_ALL)
3760#define SO_PARTIAL (1 << SL_PARTIAL)
3761#define SO_CPU (1 << SL_CPU)
3762#define SO_OBJECTS (1 << SL_OBJECTS)
3763#define SO_TOTAL (1 << SL_TOTAL)
3764
3765static ssize_t show_slab_objects(struct kmem_cache *s,
3766                char *buf, unsigned long flags)
3767{
3768    unsigned long total = 0;
3769    int node;
3770    int x;
3771    unsigned long *nodes;
3772    unsigned long *per_cpu;
3773
3774    nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
3775    if (!nodes)
3776        return -ENOMEM;
3777    per_cpu = nodes + nr_node_ids;
3778
3779    if (flags & SO_CPU) {
3780        int cpu;
3781
3782        for_each_possible_cpu(cpu) {
3783            struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
3784
3785            if (!c || c->node < 0)
3786                continue;
3787
3788            if (c->page) {
3789                    if (flags & SO_TOTAL)
3790                        x = c->page->objects;
3791                else if (flags & SO_OBJECTS)
3792                    x = c->page->inuse;
3793                else
3794                    x = 1;
3795
3796                total += x;
3797                nodes[c->node] += x;
3798            }
3799            per_cpu[c->node]++;
3800        }
3801    }
3802
3803    if (flags & SO_ALL) {
3804        for_each_node_state(node, N_NORMAL_MEMORY) {
3805            struct kmem_cache_node *n = get_node(s, node);
3806
3807        if (flags & SO_TOTAL)
3808            x = atomic_long_read(&n->total_objects);
3809        else if (flags & SO_OBJECTS)
3810            x = atomic_long_read(&n->total_objects) -
3811                count_partial(n, count_free);
3812
3813            else
3814                x = atomic_long_read(&n->nr_slabs);
3815            total += x;
3816            nodes[node] += x;
3817        }
3818
3819    } else if (flags & SO_PARTIAL) {
3820        for_each_node_state(node, N_NORMAL_MEMORY) {
3821            struct kmem_cache_node *n = get_node(s, node);
3822
3823            if (flags & SO_TOTAL)
3824                x = count_partial(n, count_total);
3825            else if (flags & SO_OBJECTS)
3826                x = count_partial(n, count_inuse);
3827            else
3828                x = n->nr_partial;
3829            total += x;
3830            nodes[node] += x;
3831        }
3832    }
3833    x = sprintf(buf, "%lu", total);
3834#ifdef CONFIG_NUMA
3835    for_each_node_state(node, N_NORMAL_MEMORY)
3836        if (nodes[node])
3837            x += sprintf(buf + x, " N%d=%lu",
3838                    node, nodes[node]);
3839#endif
3840    kfree(nodes);
3841    return x + sprintf(buf + x, "\n");
3842}
3843
3844static int any_slab_objects(struct kmem_cache *s)
3845{
3846    int node;
3847
3848    for_each_online_node(node) {
3849        struct kmem_cache_node *n = get_node(s, node);
3850
3851        if (!n)
3852            continue;
3853
3854        if (atomic_long_read(&n->total_objects))
3855            return 1;
3856    }
3857    return 0;
3858}
3859
3860#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
3861#define to_slab(n) container_of(n, struct kmem_cache, kobj);
3862
3863struct slab_attribute {
3864    struct attribute attr;
3865    ssize_t (*show)(struct kmem_cache *s, char *buf);
3866    ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
3867};
3868
3869#define SLAB_ATTR_RO(_name) \
3870    static struct slab_attribute _name##_attr = __ATTR_RO(_name)
3871
3872#define SLAB_ATTR(_name) \
3873    static struct slab_attribute _name##_attr = \
3874    __ATTR(_name, 0644, _name##_show, _name##_store)
3875
3876static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
3877{
3878    return sprintf(buf, "%d\n", s->size);
3879}
3880SLAB_ATTR_RO(slab_size);
3881
3882static ssize_t align_show(struct kmem_cache *s, char *buf)
3883{
3884    return sprintf(buf, "%d\n", s->align);
3885}
3886SLAB_ATTR_RO(align);
3887
3888static ssize_t object_size_show(struct kmem_cache *s, char *buf)
3889{
3890    return sprintf(buf, "%d\n", s->objsize);
3891}
3892SLAB_ATTR_RO(object_size);
3893
3894static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
3895{
3896    return sprintf(buf, "%d\n", oo_objects(s->oo));
3897}
3898SLAB_ATTR_RO(objs_per_slab);
3899
3900static ssize_t order_store(struct kmem_cache *s,
3901                const char *buf, size_t length)
3902{
3903    unsigned long order;
3904    int err;
3905
3906    err = strict_strtoul(buf, 10, &order);
3907    if (err)
3908        return err;
3909
3910    if (order > slub_max_order || order < slub_min_order)
3911        return -EINVAL;
3912
3913    calculate_sizes(s, order);
3914    return length;
3915}
3916
3917static ssize_t order_show(struct kmem_cache *s, char *buf)
3918{
3919    return sprintf(buf, "%d\n", oo_order(s->oo));
3920}
3921SLAB_ATTR(order);
3922
3923static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
3924{
3925    return sprintf(buf, "%lu\n", s->min_partial);
3926}
3927
3928static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
3929                 size_t length)
3930{
3931    unsigned long min;
3932    int err;
3933
3934    err = strict_strtoul(buf, 10, &min);
3935    if (err)
3936        return err;
3937
3938    set_min_partial(s, min);
3939    return length;
3940}
3941SLAB_ATTR(min_partial);
3942
3943static ssize_t ctor_show(struct kmem_cache *s, char *buf)
3944{
3945    if (s->ctor) {
3946        int n = sprint_symbol(buf, (unsigned long)s->ctor);
3947
3948        return n + sprintf(buf + n, "\n");
3949    }
3950    return 0;
3951}
3952SLAB_ATTR_RO(ctor);
3953
3954static ssize_t aliases_show(struct kmem_cache *s, char *buf)
3955{
3956    return sprintf(buf, "%d\n", s->refcount - 1);
3957}
3958SLAB_ATTR_RO(aliases);
3959
3960static ssize_t slabs_show(struct kmem_cache *s, char *buf)
3961{
3962    return show_slab_objects(s, buf, SO_ALL);
3963}
3964SLAB_ATTR_RO(slabs);
3965
3966static ssize_t partial_show(struct kmem_cache *s, char *buf)
3967{
3968    return show_slab_objects(s, buf, SO_PARTIAL);
3969}
3970SLAB_ATTR_RO(partial);
3971
3972static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
3973{
3974    return show_slab_objects(s, buf, SO_CPU);
3975}
3976SLAB_ATTR_RO(cpu_slabs);
3977
3978static ssize_t objects_show(struct kmem_cache *s, char *buf)
3979{
3980    return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
3981}
3982SLAB_ATTR_RO(objects);
3983
3984static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
3985{
3986    return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
3987}
3988SLAB_ATTR_RO(objects_partial);
3989
3990static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
3991{
3992    return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
3993}
3994SLAB_ATTR_RO(total_objects);
3995
3996static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
3997{
3998    return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
3999}
4000
4001static ssize_t sanity_checks_store(struct kmem_cache *s,
4002                const char *buf, size_t length)
4003{
4004    s->flags &= ~SLAB_DEBUG_FREE;
4005    if (buf[0] == '1')
4006        s->flags |= SLAB_DEBUG_FREE;
4007    return length;
4008}
4009SLAB_ATTR(sanity_checks);
4010
4011static ssize_t trace_show(struct kmem_cache *s, char *buf)
4012{
4013    return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4014}
4015
4016static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4017                            size_t length)
4018{
4019    s->flags &= ~SLAB_TRACE;
4020    if (buf[0] == '1')
4021        s->flags |= SLAB_TRACE;
4022    return length;
4023}
4024SLAB_ATTR(trace);
4025
4026#ifdef CONFIG_FAILSLAB
4027static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4028{
4029    return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4030}
4031
4032static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4033                            size_t length)
4034{
4035    s->flags &= ~SLAB_FAILSLAB;
4036    if (buf[0] == '1')
4037        s->flags |= SLAB_FAILSLAB;
4038    return length;
4039}
4040SLAB_ATTR(failslab);
4041#endif
4042
4043static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4044{
4045    return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4046}
4047
4048static ssize_t reclaim_account_store(struct kmem_cache *s,
4049                const char *buf, size_t length)
4050{
4051    s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4052    if (buf[0] == '1')
4053        s->flags |= SLAB_RECLAIM_ACCOUNT;
4054    return length;
4055}
4056SLAB_ATTR(reclaim_account);
4057
4058static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4059{
4060    return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
4061}
4062SLAB_ATTR_RO(hwcache_align);
4063
4064#ifdef CONFIG_ZONE_DMA
4065static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4066{
4067    return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4068}
4069SLAB_ATTR_RO(cache_dma);
4070#endif
4071
4072static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4073{
4074    return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4075}
4076SLAB_ATTR_RO(destroy_by_rcu);
4077
4078static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4079{
4080    return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4081}
4082
4083static ssize_t red_zone_store(struct kmem_cache *s,
4084                const char *buf, size_t length)
4085{
4086    if (any_slab_objects(s))
4087        return -EBUSY;
4088
4089    s->flags &= ~SLAB_RED_ZONE;
4090    if (buf[0] == '1')
4091        s->flags |= SLAB_RED_ZONE;
4092    calculate_sizes(s, -1);
4093    return length;
4094}
4095SLAB_ATTR(red_zone);
4096
4097static ssize_t poison_show(struct kmem_cache *s, char *buf)
4098{
4099    return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4100}
4101
4102static ssize_t poison_store(struct kmem_cache *s,
4103                const char *buf, size_t length)
4104{
4105    if (any_slab_objects(s))
4106        return -EBUSY;
4107
4108    s->flags &= ~SLAB_POISON;
4109    if (buf[0] == '1')
4110        s->flags |= SLAB_POISON;
4111    calculate_sizes(s, -1);
4112    return length;
4113}
4114SLAB_ATTR(poison);
4115
4116static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4117{
4118    return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4119}
4120
4121static ssize_t store_user_store(struct kmem_cache *s,
4122                const char *buf, size_t length)
4123{
4124    if (any_slab_objects(s))
4125        return -EBUSY;
4126
4127    s->flags &= ~SLAB_STORE_USER;
4128    if (buf[0] == '1')
4129        s->flags |= SLAB_STORE_USER;
4130    calculate_sizes(s, -1);
4131    return length;
4132}
4133SLAB_ATTR(store_user);
4134
4135static ssize_t validate_show(struct kmem_cache *s, char *buf)
4136{
4137    return 0;
4138}
4139
4140static ssize_t validate_store(struct kmem_cache *s,
4141            const char *buf, size_t length)
4142{
4143    int ret = -EINVAL;
4144
4145    if (buf[0] == '1') {
4146        ret = validate_slab_cache(s);
4147        if (ret >= 0)
4148            ret = length;
4149    }
4150    return ret;
4151}
4152SLAB_ATTR(validate);
4153
4154static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4155{
4156    return 0;
4157}
4158
4159static ssize_t shrink_store(struct kmem_cache *s,
4160            const char *buf, size_t length)
4161{
4162    if (buf[0] == '1') {
4163        int rc = kmem_cache_shrink(s);
4164
4165        if (rc)
4166            return rc;
4167    } else
4168        return -EINVAL;
4169    return length;
4170}
4171SLAB_ATTR(shrink);
4172
4173static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4174{
4175    if (!(s->flags & SLAB_STORE_USER))
4176        return -ENOSYS;
4177    return list_locations(s, buf, TRACK_ALLOC);
4178}
4179SLAB_ATTR_RO(alloc_calls);
4180
4181static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4182{
4183    if (!(s->flags & SLAB_STORE_USER))
4184        return -ENOSYS;
4185    return list_locations(s, buf, TRACK_FREE);
4186}
4187SLAB_ATTR_RO(free_calls);
4188
4189#ifdef CONFIG_NUMA
4190static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
4191{
4192    return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
4193}
4194
4195static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
4196                const char *buf, size_t length)
4197{
4198    unsigned long ratio;
4199    int err;
4200
4201    err = strict_strtoul(buf, 10, &ratio);
4202    if (err)
4203        return err;
4204
4205    if (ratio <= 100)
4206        s->remote_node_defrag_ratio = ratio * 10;
4207
4208    return length;
4209}
4210SLAB_ATTR(remote_node_defrag_ratio);
4211#endif
4212
4213#ifdef CONFIG_SLUB_STATS
4214static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4215{
4216    unsigned long sum = 0;
4217    int cpu;
4218    int len;
4219    int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4220
4221    if (!data)
4222        return -ENOMEM;
4223
4224    for_each_online_cpu(cpu) {
4225        unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
4226
4227        data[cpu] = x;
4228        sum += x;
4229    }
4230
4231    len = sprintf(buf, "%lu", sum);
4232
4233#ifdef CONFIG_SMP
4234    for_each_online_cpu(cpu) {
4235        if (data[cpu] && len < PAGE_SIZE - 20)
4236            len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
4237    }
4238#endif
4239    kfree(data);
4240    return len + sprintf(buf + len, "\n");
4241}
4242
4243static void clear_stat(struct kmem_cache *s, enum stat_item si)
4244{
4245    int cpu;
4246
4247    for_each_online_cpu(cpu)
4248        per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
4249}
4250
4251#define STAT_ATTR(si, text) \
4252static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4253{ \
4254    return show_stat(s, buf, si); \
4255} \
4256static ssize_t text##_store(struct kmem_cache *s, \
4257                const char *buf, size_t length) \
4258{ \
4259    if (buf[0] != '0') \
4260        return -EINVAL; \
4261    clear_stat(s, si); \
4262    return length; \
4263} \
4264SLAB_ATTR(text); \
4265
4266STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4267STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4268STAT_ATTR(FREE_FASTPATH, free_fastpath);
4269STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4270STAT_ATTR(FREE_FROZEN, free_frozen);
4271STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4272STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4273STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4274STAT_ATTR(ALLOC_SLAB, alloc_slab);
4275STAT_ATTR(ALLOC_REFILL, alloc_refill);
4276STAT_ATTR(FREE_SLAB, free_slab);
4277STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4278STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4279STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4280STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4281STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4282STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
4283STAT_ATTR(ORDER_FALLBACK, order_fallback);
4284#endif
4285
4286static struct attribute *slab_attrs[] = {
4287    &slab_size_attr.attr,
4288    &object_size_attr.attr,
4289    &objs_per_slab_attr.attr,
4290    &order_attr.attr,
4291    &min_partial_attr.attr,
4292    &objects_attr.attr,
4293    &objects_partial_attr.attr,
4294    &total_objects_attr.attr,
4295    &slabs_attr.attr,
4296    &partial_attr.attr,
4297    &cpu_slabs_attr.attr,
4298    &ctor_attr.attr,
4299    &aliases_attr.attr,
4300    &align_attr.attr,
4301    &sanity_checks_attr.attr,
4302    &trace_attr.attr,
4303    &hwcache_align_attr.attr,
4304    &reclaim_account_attr.attr,
4305    &destroy_by_rcu_attr.attr,
4306    &red_zone_attr.attr,
4307    &poison_attr.attr,
4308    &store_user_attr.attr,
4309    &validate_attr.attr,
4310    &shrink_attr.attr,
4311    &alloc_calls_attr.attr,
4312    &free_calls_attr.attr,
4313#ifdef CONFIG_ZONE_DMA
4314    &cache_dma_attr.attr,
4315#endif
4316#ifdef CONFIG_NUMA
4317    &remote_node_defrag_ratio_attr.attr,
4318#endif
4319#ifdef CONFIG_SLUB_STATS
4320    &alloc_fastpath_attr.attr,
4321    &alloc_slowpath_attr.attr,
4322    &free_fastpath_attr.attr,
4323    &free_slowpath_attr.attr,
4324    &free_frozen_attr.attr,
4325    &free_add_partial_attr.attr,
4326    &free_remove_partial_attr.attr,
4327    &alloc_from_partial_attr.attr,
4328    &alloc_slab_attr.attr,
4329    &alloc_refill_attr.attr,
4330    &free_slab_attr.attr,
4331    &cpuslab_flush_attr.attr,
4332    &deactivate_full_attr.attr,
4333    &deactivate_empty_attr.attr,
4334    &deactivate_to_head_attr.attr,
4335    &deactivate_to_tail_attr.attr,
4336    &deactivate_remote_frees_attr.attr,
4337    &order_fallback_attr.attr,
4338#endif
4339#ifdef CONFIG_FAILSLAB
4340    &failslab_attr.attr,
4341#endif
4342
4343    NULL
4344};
4345
4346static struct attribute_group slab_attr_group = {
4347    .attrs = slab_attrs,
4348};
4349
4350static ssize_t slab_attr_show(struct kobject *kobj,
4351                struct attribute *attr,
4352                char *buf)
4353{
4354    struct slab_attribute *attribute;
4355    struct kmem_cache *s;
4356    int err;
4357
4358    attribute = to_slab_attr(attr);
4359    s = to_slab(kobj);
4360
4361    if (!attribute->show)
4362        return -EIO;
4363
4364    err = attribute->show(s, buf);
4365
4366    return err;
4367}
4368
4369static ssize_t slab_attr_store(struct kobject *kobj,
4370                struct attribute *attr,
4371                const char *buf, size_t len)
4372{
4373    struct slab_attribute *attribute;
4374    struct kmem_cache *s;
4375    int err;
4376
4377    attribute = to_slab_attr(attr);
4378    s = to_slab(kobj);
4379
4380    if (!attribute->store)
4381        return -EIO;
4382
4383    err = attribute->store(s, buf, len);
4384
4385    return err;
4386}
4387
4388static void kmem_cache_release(struct kobject *kobj)
4389{
4390    struct kmem_cache *s = to_slab(kobj);
4391
4392    kfree(s);
4393}
4394
4395static const struct sysfs_ops slab_sysfs_ops = {
4396    .show = slab_attr_show,
4397    .store = slab_attr_store,
4398};
4399
4400static struct kobj_type slab_ktype = {
4401    .sysfs_ops = &slab_sysfs_ops,
4402    .release = kmem_cache_release
4403};
4404
4405static int uevent_filter(struct kset *kset, struct kobject *kobj)
4406{
4407    struct kobj_type *ktype = get_ktype(kobj);
4408
4409    if (ktype == &slab_ktype)
4410        return 1;
4411    return 0;
4412}
4413
4414static const struct kset_uevent_ops slab_uevent_ops = {
4415    .filter = uevent_filter,
4416};
4417
4418static struct kset *slab_kset;
4419
4420#define ID_STR_LENGTH 64
4421
4422/* Create a unique string id for a slab cache:
4423 *
4424 * Format :[flags-]size
4425 */
4426static char *create_unique_id(struct kmem_cache *s)
4427{
4428    char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
4429    char *p = name;
4430
4431    BUG_ON(!name);
4432
4433    *p++ = ':';
4434    /*
4435     * First flags affecting slabcache operations. We will only
4436     * get here for aliasable slabs so we do not need to support
4437     * too many flags. The flags here must cover all flags that
4438     * are matched during merging to guarantee that the id is
4439     * unique.
4440     */
4441    if (s->flags & SLAB_CACHE_DMA)
4442        *p++ = 'd';
4443    if (s->flags & SLAB_RECLAIM_ACCOUNT)
4444        *p++ = 'a';
4445    if (s->flags & SLAB_DEBUG_FREE)
4446        *p++ = 'F';
4447    if (!(s->flags & SLAB_NOTRACK))
4448        *p++ = 't';
4449    if (p != name + 1)
4450        *p++ = '-';
4451    p += sprintf(p, "%07d", s->size);
4452    BUG_ON(p > name + ID_STR_LENGTH - 1);
4453    return name;
4454}
4455
4456static int sysfs_slab_add(struct kmem_cache *s)
4457{
4458    int err;
4459    const char *name;
4460    int unmergeable;
4461
4462    if (slab_state < SYSFS)
4463        /* Defer until later */
4464        return 0;
4465
4466    unmergeable = slab_unmergeable(s);
4467    if (unmergeable) {
4468        /*
4469         * Slabcache can never be merged so we can use the name proper.
4470         * This is typically the case for debug situations. In that
4471         * case we can catch duplicate names easily.
4472         */
4473        sysfs_remove_link(&slab_kset->kobj, s->name);
4474        name = s->name;
4475    } else {
4476        /*
4477         * Create a unique name for the slab as a target
4478         * for the symlinks.
4479         */
4480        name = create_unique_id(s);
4481    }
4482
4483    s->kobj.kset = slab_kset;
4484    err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, name);
4485    if (err) {
4486        kobject_put(&s->kobj);
4487        return err;
4488    }
4489
4490    err = sysfs_create_group(&s->kobj, &slab_attr_group);
4491    if (err) {
4492        kobject_del(&s->kobj);
4493        kobject_put(&s->kobj);
4494        return err;
4495    }
4496    kobject_uevent(&s->kobj, KOBJ_ADD);
4497    if (!unmergeable) {
4498        /* Setup first alias */
4499        sysfs_slab_alias(s, s->name);
4500        kfree(name);
4501    }
4502    return 0;
4503}
4504
4505static void sysfs_slab_remove(struct kmem_cache *s)
4506{
4507    kobject_uevent(&s->kobj, KOBJ_REMOVE);
4508    kobject_del(&s->kobj);
4509    kobject_put(&s->kobj);
4510}
4511
4512/*
4513 * Need to buffer aliases during bootup until sysfs becomes
4514 * available lest we lose that information.
4515 */
4516struct saved_alias {
4517    struct kmem_cache *s;
4518    const char *name;
4519    struct saved_alias *next;
4520};
4521
4522static struct saved_alias *alias_list;
4523
4524static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
4525{
4526    struct saved_alias *al;
4527
4528    if (slab_state == SYSFS) {
4529        /*
4530         * If we have a leftover link then remove it.
4531         */
4532        sysfs_remove_link(&slab_kset->kobj, name);
4533        return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
4534    }
4535
4536    al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
4537    if (!al)
4538        return -ENOMEM;
4539
4540    al->s = s;
4541    al->name = name;
4542    al->next = alias_list;
4543    alias_list = al;
4544    return 0;
4545}
4546
4547static int __init slab_sysfs_init(void)
4548{
4549    struct kmem_cache *s;
4550    int err;
4551
4552    slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
4553    if (!slab_kset) {
4554        printk(KERN_ERR "Cannot register slab subsystem.\n");
4555        return -ENOSYS;
4556    }
4557
4558    slab_state = SYSFS;
4559
4560    list_for_each_entry(s, &slab_caches, list) {
4561        err = sysfs_slab_add(s);
4562        if (err)
4563            printk(KERN_ERR "SLUB: Unable to add boot slab %s"
4564                        " to sysfs\n", s->name);
4565    }
4566
4567    while (alias_list) {
4568        struct saved_alias *al = alias_list;
4569
4570        alias_list = alias_list->next;
4571        err = sysfs_slab_alias(al->s, al->name);
4572        if (err)
4573            printk(KERN_ERR "SLUB: Unable to add boot slab alias"
4574                    " %s to sysfs\n", s->name);
4575        kfree(al);
4576    }
4577
4578    resiliency_test();
4579    return 0;
4580}
4581
4582__initcall(slab_sysfs_init);
4583#endif
4584
4585/*
4586 * The /proc/slabinfo ABI
4587 */
4588#ifdef CONFIG_SLABINFO
4589static void print_slabinfo_header(struct seq_file *m)
4590{
4591    seq_puts(m, "slabinfo - version: 2.1\n");
4592    seq_puts(m, "# name <active_objs> <num_objs> <objsize> "
4593         "<objperslab> <pagesperslab>");
4594    seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
4595    seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
4596    seq_putc(m, '\n');
4597}
4598
4599static void *s_start(struct seq_file *m, loff_t *pos)
4600{
4601    loff_t n = *pos;
4602
4603    down_read(&slub_lock);
4604    if (!n)
4605        print_slabinfo_header(m);
4606
4607    return seq_list_start(&slab_caches, *pos);
4608}
4609
4610static void *s_next(struct seq_file *m, void *p, loff_t *pos)
4611{
4612    return seq_list_next(p, &slab_caches, pos);
4613}
4614
4615static void s_stop(struct seq_file *m, void *p)
4616{
4617    up_read(&slub_lock);
4618}
4619
4620static int s_show(struct seq_file *m, void *p)
4621{
4622    unsigned long nr_partials = 0;
4623    unsigned long nr_slabs = 0;
4624    unsigned long nr_inuse = 0;
4625    unsigned long nr_objs = 0;
4626    unsigned long nr_free = 0;
4627    struct kmem_cache *s;
4628    int node;
4629
4630    s = list_entry(p, struct kmem_cache, list);
4631
4632    for_each_online_node(node) {
4633        struct kmem_cache_node *n = get_node(s, node);
4634
4635        if (!n)
4636            continue;
4637
4638        nr_partials += n->nr_partial;
4639        nr_slabs += atomic_long_read(&n->nr_slabs);
4640        nr_objs += atomic_long_read(&n->total_objects);
4641        nr_free += count_partial(n, count_free);
4642    }
4643
4644    nr_inuse = nr_objs - nr_free;
4645
4646    seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", s->name, nr_inuse,
4647           nr_objs, s->size, oo_objects(s->oo),
4648           (1 << oo_order(s->oo)));
4649    seq_printf(m, " : tunables %4u %4u %4u", 0, 0, 0);
4650    seq_printf(m, " : slabdata %6lu %6lu %6lu", nr_slabs, nr_slabs,
4651           0UL);
4652    seq_putc(m, '\n');
4653    return 0;
4654}
4655
4656static const struct seq_operations slabinfo_op = {
4657    .start = s_start,
4658    .next = s_next,
4659    .stop = s_stop,
4660    .show = s_show,
4661};
4662
4663static int slabinfo_open(struct inode *inode, struct file *file)
4664{
4665    return seq_open(file, &slabinfo_op);
4666}
4667
4668static const struct file_operations proc_slabinfo_operations = {
4669    .open = slabinfo_open,
4670    .read = seq_read,
4671    .llseek = seq_lseek,
4672    .release = seq_release,
4673};
4674
4675static int __init slab_proc_init(void)
4676{
4677    proc_create("slabinfo", S_IRUGO, NULL, &proc_slabinfo_operations);
4678    return 0;
4679}
4680module_init(slab_proc_init);
4681#endif /* CONFIG_SLABINFO */
4682

Archive Download this file



interactive