linux-vfs : inode number

最新推荐文章于 2025-05-31 03:41:52 发布

原创最新推荐文章于 2025-05-31 03:41:52 发布 · 476 阅读

0 ·

CC 4.0 BY-SA版权

本文介绍了Linux内核3.10中inode编号的分配机制。为了减少共享资源的竞争，每个CPU都拥有一个范围内的inode编号，并且只有当分配达到一定数量时才会更新共享资源。这种方法在一定程度上减少了溢出率。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

http://blog.chinaunix.net/uid-20481436-id-1941493.html

linux-3.10/fs/inode.c
856 /*
857  * Each cpu owns a range of LAST_INO_BATCH numbers.
858  * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
859  * to renew the exhausted range.
860  *
861  * This does not significantly increase overflow rate because every CPU can
862  * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
863  * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
864  * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
865  * overflow rate by 2x, which does not seem too significant.
866  *
867  * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
868  * error if st_ino won't fit in target struct field. Use 32bit counter
869  * here to attempt to avoid that.
870  */
871 #define LAST_INO_BATCH 1024
872 static DEFINE_PER_CPU(unsigned int, last_ino);
873 
874 unsigned int get_next_ino(void)
875 {
876         unsigned int *p = &get_cpu_var(last_ino);
877         unsigned int res = *p;
878 
879 #ifdef CONFIG_SMP
880         if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
881                 static atomic_t shared_last_ino;
882                 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
883 
884                 res = next - LAST_INO_BATCH;
885         }
886 #endif
887 
888         *p = ++res;
889         put_cpu_var(last_ino);
890         return res;
891 }
892 EXPORT_SYMBOL(get_next_ino);