remove useless check of bin match in malloc

[musl] / src / malloc / malloc.c

diff --git a/src/malloc/malloc.c b/src/malloc/malloc.c
index 39c7d05..d4de2dc 100644 (file)
--- a/src/malloc/malloc.c
+++ b/src/malloc/malloc.c
@@ -9,6 +9,10 @@
 #include "atomic.h"
 #include "pthread_impl.h"
 
+#if defined(__GNUC__) && defined(__PIC__)
+#define inline inline __attribute__((always_inline))
+#endif
+
 uintptr_t __brk(uintptr_t);
 void *__mmap(void *, size_t, int, int, int, off_t);
 int __munmap(void *, size_t);
@@ -21,7 +25,7 @@ struct chunk {
 };
 
 struct bin {
-       int lock[2];
+       volatile int lock[2];
        struct chunk *head;
        struct chunk *tail;
 };
@@ -29,10 +33,11 @@ struct bin {
 static struct {
        uintptr_t brk;
        size_t *heap;
-       uint64_t binmap;
+       volatile uint64_t binmap;
        struct bin bins[64];
-       int brk_lock[2];
-       int free_lock[2];
+       volatile int brk_lock[2];
+       volatile int free_lock[2];
+       unsigned mmap_step;
 } mal;
 
 
@@ -58,30 +63,29 @@ static struct {
 
 /* Synchronization tools */
 
-static void lock(volatile int *lk)
+static inline void lock(volatile int *lk)
 {
-       if (!libc.threads_minus_1) return;
-       while(a_swap(lk, 1)) __wait(lk, lk+1, 1, 1);
+       if (libc.threads_minus_1)
+               while(a_swap(lk, 1)) __wait(lk, lk+1, 1, 1);
 }
 
-static void unlock(volatile int *lk)
+static inline void unlock(volatile int *lk)
 {
-       if (!libc.threads_minus_1) return;
-       a_store(lk, 0);
-       if (lk[1]) __wake(lk, 1, 1);
+       if (lk[0]) {
+               a_store(lk, 0);
+               if (lk[1]) __wake(lk, 1, 1);
+       }
 }
 
-static void lock_bin(int i)
+static inline void lock_bin(int i)
 {
-       if (libc.threads_minus_1)
-               lock(mal.bins[i].lock);
+       lock(mal.bins[i].lock);
        if (!mal.bins[i].head)
                mal.bins[i].head = mal.bins[i].tail = BIN_TO_CHUNK(i);
 }
 
-static void unlock_bin(int i)
+static inline void unlock_bin(int i)
 {
-       if (!libc.threads_minus_1) return;
        unlock(mal.bins[i].lock);
 }
 
@@ -150,16 +154,51 @@ void __dump_heap(int x)
 
 static struct chunk *expand_heap(size_t n)
 {
+       static int init;
        struct chunk *w;
        uintptr_t new;
 
        lock(mal.brk_lock);
 
+       if (!init) {
+               mal.brk = __brk(0);
+#ifdef SHARED
+               mal.brk = mal.brk + PAGE_SIZE-1 & -PAGE_SIZE;
+#endif
+               mal.brk = mal.brk + 2*SIZE_ALIGN-1 & -SIZE_ALIGN;
+               mal.heap = (void *)mal.brk;
+               init = 1;
+       }
+
        if (n > SIZE_MAX - mal.brk - 2*PAGE_SIZE) goto fail;
        new = mal.brk + n + SIZE_ALIGN + PAGE_SIZE - 1 & -PAGE_SIZE;
        n = new - mal.brk;
 
-       if (__brk(new) != new) goto fail;
+       if (__brk(new) != new) {
+               size_t min = (size_t)PAGE_SIZE << mal.mmap_step/2;
+               n += -n & PAGE_SIZE-1;
+               if (n < min) n = min;
+               void *area = __mmap(0, n, PROT_READ|PROT_WRITE,
+                       MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
+               if (area == MAP_FAILED) goto fail;
+
+               mal.mmap_step++;
+               area = (char *)area + SIZE_ALIGN - OVERHEAD;
+               w = area;
+               n -= SIZE_ALIGN;
+               w->psize = 0 | C_INUSE;
+               w->csize = n | C_INUSE;
+               w = NEXT_CHUNK(w);
+               w->psize = n | C_INUSE;
+               w->csize = 0 | C_INUSE;
+
+               unlock(mal.brk_lock);
+
+               return area;
+       }
+
+       w = MEM_TO_CHUNK(mal.heap);
+       w->psize = 0 | C_INUSE;
 
        w = MEM_TO_CHUNK(new);
        w->psize = n | C_INUSE;
@@ -174,43 +213,10 @@ static struct chunk *expand_heap(size_t n)
        return w;
 fail:
        unlock(mal.brk_lock);
+       errno = ENOMEM;
        return 0;
 }
 
-static int init_malloc(size_t n)
-{
-       static int init, waiters;
-       int state;
-       struct chunk *c;
-
-       if (init == 2) return 0;
-
-       while ((state=a_swap(&init, 1)) == 1)
-               __wait(&init, &waiters, 1, 1);
-       if (state) {
-               a_store(&init, 2);
-               return 0;
-       }
-
-       mal.brk = __brk(0) + 2*SIZE_ALIGN-1 & -SIZE_ALIGN;
-
-       c = expand_heap(n);
-
-       if (!c) {
-               a_store(&init, 0);
-               if (waiters) __wake(&init, 1, 1);
-               return -1;
-       }
-
-       mal.heap = (void *)c;
-       c->psize = 0 | C_INUSE;
-       free(CHUNK_TO_MEM(c));
-
-       a_store(&init, 2);
-       if (waiters) __wake(&init, -1, 1);
-       return 1;
-}
-
 static int adjust_size(size_t *n)
 {
        /* Result of pointer difference must fit in ptrdiff_t. */
@@ -345,7 +351,6 @@ void *malloc(size_t n)
        for (;;) {
                uint64_t mask = mal.binmap & -(1ULL<<i);
                if (!mask) {
-                       if (init_malloc(n) > 0) continue;
                        c = expand_heap(n);
                        if (!c) return 0;
                        if (alloc_rev(c)) {
@@ -359,7 +364,7 @@ void *malloc(size_t n)
                j = first_set(mask);
                lock_bin(j);
                c = mal.bins[j].head;
-               if (c != BIN_TO_CHUNK(j) && j == bin_index(c->csize)) {
+               if (c != BIN_TO_CHUNK(j)) {
                        if (!pretrim(c, n, i, j)) unbin(c, j);
                        unlock_bin(j);
                        break;
@@ -410,6 +415,9 @@ void *realloc(void *p, size_t n)
 
        next = NEXT_CHUNK(self);
 
+       /* Crash on corrupted footer (likely from buffer overflow) */
+       if (next->psize != self->csize) a_crash();
+
        /* Merge adjacent chunks if we need more space. This is not
         * a waste of time even if we fail to get enough space, because our
         * subsequent call to free would otherwise have to do the merge. */
@@ -463,6 +471,9 @@ void free(void *p)
        final_size = new_size = CHUNK_SIZE(self);
        next = NEXT_CHUNK(self);
 
+       /* Crash on corrupted footer (likely from buffer overflow) */
+       if (next->psize != self->csize) a_crash();
+
        for (;;) {
                /* Replace middle of large chunks with fresh zero pages */
                if (reclaim && (self->psize & next->csize & C_INUSE)) {