What is linux buffer

In Linux, Buffer refers to a buffer, which is an area used to store data transmitted between devices with unsynchronized speeds or devices with different priorities. It is a balance of processing speeds at both ends of the system (from a long time It is used when looking at the scale); it is introduced to reduce the impact of burst I/O in the short term and play the role of traffic shaping.

#The operating environment of this tutorial: linux7.3 system, Dell G3 computer.

What are buffer and cache

1. Buffer (buffer) is used when the processing speed of both ends of the system is balanced (from a long-term perspective). It is introduced to reduce the impact of burst I/O in the short term and play the role of traffic shaping.

Buffer is an area used to store data transferred between devices with unsynchronized speeds or devices with different priorities. Through the buffer, the waiting between processes can be reduced, so that when data is read from a slow device, the operation process of the fast device will not be interrupted.

For example: If we use X Thunder to download a movie, it is impossible to download a little bit and then write a little bit to the disk. Doing this will really destroy the hard disk. On the contrary, write to the buffer first, and then write to the disk all at once to reduce I/O, which is both efficient and friendly to the hard disk.

2. Cache is a compromise strategy when the processing speeds of both ends do not match. Because the speed difference between CPU and memory is getting larger and larger, people make full use of the local characteristics of data and reduce the impact of this difference by using the strategy of storage system hierarchy (memory hierarchy).

Understanding Linux Cache and Buffer

cache refers to the page cache in Linux, and buffer refers to the buffer cache, which is displayed in cat /proc/meminfo cache and buffer.

We know that when files or a single large file are frequently accessed under Linux, the physical memory will be quickly used up. When the program ends, the memory will not be released normally but will always occupy the memory as cahce. Therefore, the system often causes OOM due to this, especially in high pressure scenarios. At this time, it is very high to check the cache and buffer memory at the first time. There is currently no good solution to this kind of problem, and most of the problems encountered in the past were avoided. Therefore, this case attempts to give an idea of ​​analysis and solution.

The key to solving this problem is to understand what cache and buffer are, when and where they are consumed, and how to control cache and buffer, so this question mainly focuses on these points. The entire discussion process should start with kernel source code analysis as much as possible, then refine the APP-related interfaces and conduct practical verification, and finally summarize and give application programming suggestions.

You can check the buffer and cache status of the system through free or cat /proc/meminfo.

Full analysis of free command

1. Cache and Buffer analysis

From cat /proc/meminfo To start, let’s first look at the implementation of this interface:

static int meminfo_proc_show(struct seq_file *m, void *v)  
{  
……  
cached = global_page_state(NR_FILE_PAGES) -  
 total_swapcache_pages() - i.bufferram;  
if (cached < 0)  
 cached = 0;  
……  
 seq_printf(m,  
 "MemTotal: %8lu kB\n"  
 "MemFree: %8lu kB\n"  
 "Buffers: %8lu kB\n"  
 "Cached: %8lu kB\n"  
 ……  
 ,  
 K(i.totalram),  
 K(i.freeram),  
 K(i.bufferram),  
 K(cached),  
 ……  
 );  
……  
}

Among them, the page frame is used as the unit in the kernel, and it is converted into KB as the unit for output through the macro K. These values ​​are obtained through si_meminfo:

void si_meminfo(struct sysinfo *val) 
{ 
 val->totalram = totalram_pages; 
 val->sharedram = 0; 
 val->freeram = global_page_state(NR_FREE_PAGES); 
 val->bufferram = nr_blockdev_pages(); 
 val->totalhigh = totalhigh_pages; 
 val->freehigh = nr_free_highpages(); 
 val->mem_unit = PAGE_SIZE; 
}

The bufferram comes from nr_blockdev_pages(). This function calculates the number of page frames used by the block device, traverses all block devices, and adds the number of page frames used. It does not include the number of page frames used by ordinary files.

long nr_blockdev_pages(void)  
{  
 struct block_device *bdev;  
 long ret = 0;  
 spin_lock(&bdev_lock);  
 list_for_each_entry(bdev, &all_bdevs, bd_list) {  
 ret += bdev->bd_inode->i_mapping->nrpages;  
 }  
 spin_unlock(&bdev_lock);  
 return ret;  
}

From the above, we can conclude the sources of cache and buffer in meminfo:

• Buffer is the number of page frames occupied by the block device;
• The size of Cache is the total number of kernel The page cache minus the number of page frames occupied by the swap cache and block devices. In fact, the cache is the page cache occupied by ordinary files.

Through kernel code analysis (complex kernel code analysis is skipped here), although the implementation difference between the two is not very big, both are managed through the address_space object, but the page cache is for files Cache of data and buffer cache is a cache of block device data. Each block device will be assigned a file operation method of def_blk_ops. This is the operation method of the device. There will be a radix tree under the inode of each block device (the inode of the bdev pseudo file system). Under this radix tree will be placed Page of cached data. The number of this page will be displayed in the cat /proc/meminfobuffer column. That is to say, in the absence of a file system, the data directly operated on the block device using tools such as dd will be cached in the buffer cache. If the block device is used as a file system, then each file in the file system has an inode. This inode will be assigned operation methods such as ext3_ops. These methods are file system methods. There is also a radix tree under this inode, and here too The page of the file will be cached. The number of cached pages is counted in the cache column of cat /proc/meminfo. When operating on files at this time, most of the data will be cached in the page cache, and a small amount of the metadata of the file system files will be cached in the buffer cache.

Here, we use the cp command to copy a 50MB file operation, what will happen to the memory:

[root nfs_dir] # ll -h file_50MB.bin 
-rw-rw-r-- 1 4104 4106 50.0M Feb 24 2016 file_50MB.bin 
[root nfs_dir] # cat /proc/meminfo 
MemTotal: 90532 kB 
MemFree: 65696 kB 
Buffers: 0 kB 
Cached: 8148 kB 
…… 
[root@test nfs_dir] # cp file_50MB.bin / 
[root@test nfs_dir] # cat /proc/meminfo 
MemTotal: 90532 kB 
MemFree: 13012 kB 
Buffers: 0 kB 
Cached: 60488 kB

可以看到cp命令前后，MemFree从65696 kB减少为13012 kB，Cached从8148 kB增大为60488  kB，而Buffers却不变。那么过一段时间，Linux会自动释放掉所用的cache内存吗?一个小时后查看proc/meminfo显示cache仍然没有变化。

接着，我们看下使用dd命令对块设备写操作前后的内存变化：

[0225_19:10:44:10s][root@test nfs_dir] # cat /proc/meminfo 
[0225_19:10:44:10s]MemTotal: 90532 kB  
[0225_19:10:44:10s]MemFree: 58988 kB  
[0225_19:10:44:10s]Buffers: 0 kB  
[0225_19:10:44:10s]Cached: 4144 kB  
...... ......  
[0225_19:11:13:11s][root@test nfs_dir] # dd if=/dev/zero of=/dev/h_sda bs=10M count=2000 &  
[0225_19:11:17:11s][root@test nfs_dir] # cat /proc/meminfo  
[0225_19:11:17:11s]MemTotal: 90532 kB  
[0225_19:11:17:11s]MemFree: 11852 kB  
[0225_19:11:17:11s]Buffers: 36224 kB  
[0225_19:11:17:11s]Cached: 4148 kB  
...... ......  
[0225_19:11:21:11s][root@test nfs_dir] # cat /proc/meminfo  
[0225_19:11:21:11s]MemTotal: 90532 kB  
[0225_19:11:21:11s]MemFree: 11356 kB  
[0225_19:11:21:11s]Buffers: 36732 kB  
[0225_19:11:21:11s]Cached: 4148kB  
...... ......  
[0225_19:11:41:11s][root@test nfs_dir] # cat /proc/meminfo  
[0225_19:11:41:11s]MemTotal: 90532 kB  
[0225_19:11:41:11s]MemFree: 11864 kB  
[0225_19:11:41:11s]Buffers: 36264 kB  
[0225_19:11:41:11s]Cached: 4148 kB  
….. ……

裸写块设备前Buffs为0，裸写硬盘过程中每隔一段时间查看内存信息发现Buffers一直在增加，空闲内存越来越少，而Cached数量一直保持不变。

总结：

通过代码分析及实际操作，我们理解了buffer cache和page cache都会占用内存，但也看到了两者的差别。page  cache针对文件的cache，buffer是针对块设备数据的cache。Linux在可用内存充裕的情况下，不会主动释放page cache和buffer  cache。

2. 使用posix_fadvise控制Cache

在Linux中文件的读写一般是通过buffer io方式，以便充分利用到page cache。

Buffer  IO的特点是读的时候，先检查页缓存里面是否有需要的数据，如果没有就从设备读取，返回给用户的同时，加到缓存一份;写的时候，直接写到缓存去，再由后台的进程定期刷到磁盘去。这样的机制看起来非常的好，实际也能提高文件读写的效率。

但是当系统的IO比较密集时，就会出问题。当系统写的很多，超过了内存的某个上限时，后台的回写线程就会出来回收页面，但是一旦回收的速度小于写入的速度，就会触发OOM。最关键的是整个过程由内核参与，用户不好控制。

那么到底如何才能有效的控制cache呢?

目前主要由两种方法来规避风险：

• 走direct io;
• 走buffer io，但是定期清除无用page cache;

这里当然讨论的是第二种方式，即在buffer io方式下如何有效控制page cache。

在程序中只要知道文件的句柄，就能用：

int posix_fadvise(int fd, off_t offset, off_t len, int advice);

POSIX_FADV_DONTNEED (该文件在接下来不会再被访问)，但是曾有开发人员反馈怀疑该接口的有效性。那么该接口确实有效吗?首先，我们查看mm/fadvise.c内核代码来看posix_fadvise是如何实现的：

/*  
 * POSIX_FADV_WILLNEED could set PG_Referenced, and POSIX_FADV_NOREUSE could  
 * deactivate the pages and clear PG_Referenced.  
 */  
SYSCALL_DEFINE4(fadvise64_64, int, fd, loff_t, offset, loff_t, len, int, advice)  
{  
 … … … …  
 /* => 将指定范围内的数据从page cache中换出 */  
 case POSIX_FADV_DONTNEED:  
 /* => 如果后备设备不忙的话，先调用__filemap_fdatawrite_range把脏页面刷掉 */  
 if (!bdi_write_congested(mapping->backing_dev_info))  
 /* => WB_SYNC_NONE: 不是同步等待页面刷新完成，只是提交了 */  
 /* => 而fsync和fdatasync是用WB_SYNC_ALL参数等到完成才返回的 */  
 __filemap_fdatawrite_range(mapping, offset, endbyte,  
 WB_SYNC_NONE);  
  
 
 /* First and last FULL page! */  
 start_index = (offset+(PAGE_CACHE_SIZE-1)) >> PAGE_CACHE_SHIFT;  
 end_index = (endbyte >> PAGE_CACHE_SHIFT);  
  
 
 /* => 接下来清除页面缓存 */ 
 if (end_index >= start_index) {  
 unsigned long count = invalidate_mapping_pages(mapping,  
 start_index, end_index); 
  
 
 /*  
 * If fewer pages were invalidated than expected then  
 * it is possible that some of the pages were on 
 * a per-cpu pagevec for a remote CPU. Drain all  
 * pagevecs and try again.  
 */  
 if (count < (end_index - start_index + 1)) {  
 lru_add_drain_all();  
 invalidate_mapping_pages(mapping, start_index,  
 end_index);  
 }  
 }  
 break;  
… … … …  
}

我们可以看到如果后台系统不忙的话，会先调用__filemap_fdatawrite_range把脏页面刷掉，刷页面用的参数是是  WB_SYNC_NONE，也就是说不是同步等待页面刷新完成，提交完写脏页后立即返回了。

然后再调invalidate_mapping_pages清除页面，回收内存：

/* => 清除缓存页(除了脏页、上锁的、正在回写的或映射在页表中的)*/  
unsigned long invalidate_mapping_pages(struct address_space *mapping,  
 pgoff_t start, pgoff_t end)  
{  
 struct pagevec pvec;  
 pgoff_t index = start;  
 unsigned long ret;  
 unsigned long count = 0;  
 int i;  
  
 
 /*  
 * Note: this function may get called on a shmem/tmpfs mapping:  
 * pagevec_lookup() might then return 0 prematurely (because it  
 * got a gangful of swap entries); but it's hardly worth worrying  
 * about - it can rarely have anything to free from such a mapping  
 * (most pages are dirty), and already skips over any difficulties.  
 */  
  
 
 pagevec_init(&pvec, 0);  
 while (index <= end && pagevec_lookup(&pvec, mapping, index,  
 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {  
 mem_cgroup_uncharge_start();  
 for (i = 0; i < pagevec_count(&pvec); i++) {  
 struct page *page = pvec.pages[i];  
  
 
 /* We rely upon deletion not changing page->index */  
 index = page->index;  
 if (index > end)  
 break;  
  
 
 if (!trylock_page(page))  
 continue;  
 WARN_ON(page->index != index);  
 /* => 无效一个文件的缓存 */  
 ret = invalidate_inode_page(page);  
 unlock_page(page);  
 /*  
 * Invalidation is a hint that the page is no longer  
 * of interest and try to speed up its reclaim.  
 */  
 if (!ret)  
 deactivate_page(page);  
 count += ret;  
 }  
 pagevec_release(&pvec);  
 mem_cgroup_uncharge_end();  
 cond_resched();  
 index++;  
 }  
 return count;  
} 
   
 
/*  
 * Safely invalidate one page from its pagecache mapping.  
 * It only drops clean, unused pages. The page must be locked.  
 *  
 * Returns 1 if the page is successfully invalidated, otherwise 0.  
 */  
/* => 无效一个文件的缓存 */  
int invalidate_inode_page(struct page *page)  
{  
 struct address_space *mapping = page_mapping(page);  
 if (!mapping)  
 return 0;  
 /* => 若当前页是脏页或正在写回的页，直接返回 */  
 if (PageDirty(page) || PageWriteback(page))  
 return 0;  
 /* => 若已经被映射到页表了，则直接返回 */  
 if (page_mapped(page))  
 return 0;  
 /* => 如果满足了以上条件就调用invalidate_complete_page继续 */  
 return invalidate_complete_page(mapping, page);  
}  
从上面的代码可以看到清除相关的页面要满足二个条件： 1. 不脏且没在回写； 2. 未被使用。如果满足了这二个条件就调用invalidate_complete_page继续：  
/* => 无效一个完整的页 */  
static int  
invalidate_complete_page(struct address_space *mapping, struct page *page)  
{  
 int ret;  
  
 
 if (page->mapping != mapping)  
 return 0;  
  
 
 if (page_has_private(page) && !try_to_release_page(page, 0))  
 return 0;  
  
 
 /* => 若满足以上更多条件，则从地址空间中解除该页 */  
 ret = remove_mapping(mapping, page); 
   
 
 return ret;  
}  
  
 
/*  
 * Attempt to detach a locked page from its ->mapping. If it is dirty or if  
 * someone else has a ref on the page, abort and return 0. If it was  
 * successfully detached, return 1. Assumes the caller has a single ref on  
 * this page.  
 */  
/* => 从地址空间中解除该页 */  
int remove_mapping(struct address_space *mapping, struct page *page) 
{  
 if (__remove_mapping(mapping, page)) {  
 /*  
 * Unfreezing the refcount with 1 rather than 2 effectively  
 * drops the pagecache ref for us without requiring another  
 * atomic operation.  
 */  
 page_unfreeze_refs(page, 1);  
 return 1;  
 }  
 return 0;  
} 
   
 
/*  
 * Same as remove_mapping, but if the page is removed from the mapping, it  
 * gets returned with a refcount of 0.  
 */  
/* => 从地址空间中解除该页 */  
static int __remove_mapping(struct address_space *mapping, struct page *page)  
{  
 BUG_ON(!PageLocked(page));  
 BUG_ON(mapping != page_mapping(page));  
  
 
 spin_lock_irq(&mapping->tree_lock);  
 /*  
 * The non racy check for a busy page.  
 *  
 * Must be careful with the order of the tests. When someone has  
 * a ref to the page, it may be possible that they dirty it then  
 * drop the reference. So if PageDirty is tested before page_count 
 * here, then the following race may occur: 
 *  
 * get_user_pages(&page);  
 * [user mapping goes away]  
 * write_to(page);  
 * !PageDirty(page) [good]  
 * SetPageDirty(page); 
 * put_page(page); 
 * !page_count(page) [good, discard it]  
 *  
 * [oops, our write_to data is lost]  
 *  
 * Reversing the order of the tests ensures such a situation cannot  
 * escape unnoticed. The smp_rmb is needed to ensure the page->flags  
 * load is not satisfied before that of page->_count.  
 *  
 * Note that if SetPageDirty is always performed via set_page_dirty,  
 * and thus under tree_lock, then this ordering is not required. 
 */  
 if (!page_freeze_refs(page, 2))  
 goto cannot_free;  
 /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */  
 if (unlikely(PageDirty(page))) {  
 page_unfreeze_refs(page, 2);  
 goto cannot_free; 
 } 
  
 if (PageSwapCache(page)) {  
 swp_entry_t swap = { .val = page_private(page) }; 
 __delete_from_swap_cache(page);  
 spin_unlock_irq(&mapping->tree_lock);  
 swapcache_free(swap, page);  
 } else {  
 void (*freepage)(struct page *);  
  
 
 freepage = mapping->a_ops->freepage;  
  
  /* => 从页缓存中删除和释放该页 */ 
  __delete_from_page_cache(page); 
  spin_unlock_irq(&mapping->tree_lock); 
  mem_cgroup_uncharge_cache_page(page); 
   
  if (freepage != NULL) 
  freepage(page);  
 }  
  
  return 1;  
  
 
cannot_free:  
 spin_unlock_irq(&mapping->tree_lock);  
 return 0;  
}  
  
 /* 
 * Delete a page from the page cache and free it. Caller has to make  
 * sure the page is locked and that nobody else uses it - or that usage  
 * is safe. The caller must hold the mapping&#39;s tree_lock.  
 */  
/* => 从页缓存中删除和释放该页 */  
void __delete_from_page_cache(struct page *page)  
{  
 struct address_space *mapping = page->mapping;  
   
 trace_mm_filemap_delete_from_page_cache(page);  
 /*  
 * if we&#39;re uptodate, flush out into the cleancache, otherwise  
 * invalidate any existing cleancache entries. We can&#39;t leave  
 * stale data around in the cleancache once our page is gone  
 */  
 if (PageUptodate(page) && PageMappedToDisk(page))  
 cleancache_put_page(page);  
 else  
 cleancache_invalidate_page(mapping, page);  
  
 
 radix_tree_delete(&mapping->page_tree, page->index);  
 /* => 解除与之绑定的地址空间结构 */  
 page->mapping = NULL;  
 /* Leave page->index set: truncation lookup relies upon it */  
 /* => 减少地址空间中的页计数 */ 
 mapping->nrpages--; 
 __dec_zone_page_state(page, NR_FILE_PAGES);  
 if (PageSwapBacked(page))  
 __dec_zone_page_state(page, NR_SHMEM);  
 BUG_ON(page_mapped(page)); 
   
 
 /*  
 * Some filesystems seem to re-dirty the page even after  
 * the VM has canceled the dirty bit (eg ext3 journaling).  
 *  
 * Fix it up by doing a final dirty accounting check after  
 * having removed the page entirely.  
 */  
 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {  
 dec_zone_page_state(page, NR_FILE_DIRTY);  
 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); 
 } 
}

看到这里我们就明白了：为什么使用了posix_fadvise后相关的内存没有被释放出来：页面还脏是最关键的因素。

但是我们如何保证页面全部不脏呢?fdatasync或者fsync都是选择,或者Linux下新系统调用sync_file_range都是可用的，这几个都是使用WB_SYNC_ALL模式强制要求回写完毕才返回的。所以应该这样做：

fdatasync(fd); 
posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED);

总结：

使用posix_fadvise可以有效的清除page cache，作用范围为文件级。下面给出应用程序编程建议：

• 用于测试I/O的效率时，可以用posix_fadvise来消除cache的影响;
• 当确认访问的文件在接下来一段时间不再被访问时，很有必要调用posix_fadvise来避免占用不必要的可用内存空间。
• 若当前系统内存十分紧张时，且在读写一个很大的文件时，为避免OOM风险，可以分段边读写边清cache，但也直接导致性能的下降，毕竟空间和时间是一对矛盾体。

3. 使用vmtouch控制Cache

vmtouch是一个可移植的文件系统cahce诊断和控制工具。近来该工具被广泛使用，最典型的例子是：移动应用Instagram(照片墙)后台服务端使用了vmtouch管理控制page  cache。了解vmtouch原理及使用可以为我们后续后端设备所用。

快速安装指南：

$ git clone https://github.com/hoytech/vmtouch.git 
$ cd vmtouch 
$ make 
$ sudo make install

vmtouch用途：

• 查看一个文件(或者目录)哪些部分在内存中;
• 把文件调入内存;
• 把文件清除出内存，即释放page cache;
• 把文件锁住在内存中而不被换出到磁盘上;
• ……

vmtouch实现：

其核心分别是两个系统调用，mincore和posix_fadvise。两者具体使用方法使用man帮助都有详细的说明。posix_fadvise已在上文提到，用法在此不作说明。简单说下mincore：

NAME 
 mincore - determine whether pages are resident in memory 
 
  
SYNOPSIS 
 #include <unistd.h> 
 #include <sys/mman.h> 
  
 
 int mincore(void *addr, size_t length, unsigned char *vec); 
 
  
 Feature Test Macro Requirements for glibc (see feature_test_macros(7)): 
 
  
 mincore(): _BSD_SOURCE || _SVID_SOURCE

mincore需要调用者传入文件的地址(通常由mmap()返回)，它会把文件在内存中的情况写在vec中。

vmtouch工具用法：

Usage：vmtouch [OPTIONS] ... FILES OR DIRECTORIES ...

Options：

• -t touch pages into memory
• -e evict pages from memory
• -l lock pages in physical memory with mlock(2)
• -L lock pages in physical memory with mlockall(2)
• -d daemon mode
• -mmax file size to touch
• -puse the specified portion instead of the entire file
• -f follow symbolic links
• -h also count hardlinked copies
• -w wait until all pages are locked (only useful together with -d)
• -v verbose
• -q quiet

用法举例：

例1、 获取当前/mnt/usb目录下cache占用量

[root@test nfs_dir] # mkdir /mnt/usb && mount /dev/msc /mnt/usb/  
[root@test usb] # vmtouch .  
 Files: 57  
 Directories: 2  
 Resident Pages: 0/278786 0/1G 0%  
 Elapsed: 0.023126 seconds

例2、 当前test.bin文件的cache占用量?

[root@test usb] # vmtouch -v test.bin  
test.bin  
[ ] 0/25600  
  
 
 Files: 1  
 Directories: 0  
 Resident Pages: 0/25600 0/100M 0%  
 Elapsed: 0.001867 seconds

这时使用tail命令将部分文件读取到内存中：

[root@test usb] # busybox_v400 tail -n 10 test.bin > /dev/null

现在再来看一下：

[root@test usb] # vmtouch -v test.bin  
test.bin  
[ o] 240/25600  
  
 
 Files: 1  
 Directories: 0  
 Resident Pages: 240/25600 960K/100M 0.938%  
 Elapsed: 0.002019 seconds

可知目前文件test.bin的最后240个page驻留在内存中。

例3、 最后使用-t选项将剩下的test.bin文件全部读入内存：

[root@test usb] # vmtouch -vt test.bin  
test.bin  
[OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO] 25600/25600  
  
 
 Files: 1  
 Directories: 0  
 Touched Pages: 25600 (100M) 
 Elapsed: 39.049 seconds

例4、 再把test.bin占用的cachae全部释放：

[root@test usb] # vmtouch -ev test.bin  
Evicting test.bin  
  
 
 Files: 1  
 Directories: 0  
 Evicted Pages: 25600 (100M)  
 Elapsed: 0.01461 seconds

这时候再来看下是否真的被释放了：

[root@test usb] # vmtouch -v test.bin  
test.bin  
[ ] 0/25600  
  
 
 Files: 1  
 Directories: 0  
 Resident Pages: 0/25600 0/100M 0%  
 Elapsed: 0.001867 seconds

以上通过代码分析及实际操作总结了vmtouch工具的使用，建议APP组后续集成或借鉴vmtouch工具并灵活应用到后端设备中，必能达到有效管理和控制page  cache的目的。

4. 使用BLKFLSBUF清Buffer

通过走读块设备驱动IOCTL命令实现，发现该命令能有效的清除整个块设备所占用的buffer。

int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,  
 unsigned long arg)  
{  
 struct gendisk *disk = bdev->bd_disk;  
 struct backing_dev_info *bdi;  
 loff_t size;  
 int ret, n;  
  
 
 switch(cmd) {  
 case BLKFLSBUF:  
 if (!capable(CAP_SYS_ADMIN))  
 return -EACCES;  
  
 
 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);  
 if (!is_unrecognized_ioctl(ret))  
 return ret;  
  
 
 fsync_bdev(bdev);  
 invalidate_bdev(bdev); 
 return 0;  
case ……:  
…………  
}  
  
 
/* Invalidate clean unused buffers and pagecache. */  
void invalidate_bdev(struct block_device *bdev)  
{  
 struct address_space *mapping = bdev->bd_inode->i_mapping;  
  
 
 if (mapping->nrpages == 0)  
 return;  
  
 
 invalidate_bh_lrus();  
 lru_add_drain_all(); /* make sure all lru add caches are flushed */  
 invalidate_mapping_pages(mapping, 0, -1);  
 /* 99% of the time, we don't need to flush the cleancache on the bdev.  
 * But, for the strange corners, lets be cautious  
 */  
 cleancache_invalidate_inode(mapping);  
}  
EXPORT_SYMBOL(invalidate_bdev);

光代码不够，现在让我们看下对/dev/h_sda这个块设备执行BLKFLSBUF的IOCTL命令前后的实际内存变化：

[0225_19:10:25:10s][root@test nfs_dir] # cat /proc/meminfo  
[0225_19:10:25:10s]MemTotal: 90532 kB  
[0225_19:10:25:10s]MemFree: 12296 kB  
[0225_19:10:25:10s]Buffers: 46076 kB  
[0225_19:10:25:10s]Cached: 4136 kB  
…………  
[0225_19:10:42:10s][root@test nfs_dir] # /mnt/nfs_dir/a.out  
[0225_19:10:42:10s]ioctl cmd BLKFLSBUF ok!  
[0225_19:10:44:10s][root@test nfs_dir] # cat /proc/meminfo  
[0225_19:10:44:10s]MemTotal: 90532 kB  
[0225_19:10:44:10s]MemFree: 58988 kB  
[0225_19:10:44:10s]Buffers: 0 kB  
…………  
[0225_19:10:44:10s]Cached: 4144 kB

执行的效果如代码中看到的，Buffers已被全部清除了，MemFree一下增长了约46MB，可以知道原先的Buffer已被回收并转化为可用的内存。整个过程Cache几乎没有变化，仅增加的8K  cache内存可以推断用于a.out本身及其他库文件的加载。

上述a.out的示例如下：

#include <stdio.h>  
#include <fcntl.h>  
#include <errno.h> 
#include <sys/ioctl.h> 
#define BLKFLSBUF _IO(0x12, 97) 
int main(int argc, char* argv[])  
{  
 int fd = -1;  
 fd = open("/dev/h_sda", O_RDWR); 
 if (fd < 0) 
 { 
 return -1; 
 }  
 if (ioctl(fd, BLKFLSBUF, 0))  
 {  
 printf("ioctl cmd BLKFLSBUF failed, errno:%d\n", errno);  
 } 
 close(fd); 
printf("ioctl cmd BLKFLSBUF ok!\n"); 
 return 0;  
}

综上，使用块设备命令BLKFLSBUF能有效的清除块设备上的所有buffer，且清除后的buffer能立即被释放变为可用内存。

利用这一点，联系后端业务场景，给出应用程序编程建议：

• 每次关闭一个块设备文件描述符前，必须要调用BLKFLSBUF命令，确保buffer中的脏数据及时刷入块设备，避免意外断电导致数据丢失，同时也起到及时释放回收buffer的目的。
• 当操作一个较大的块设备时，必要时可以调用BLKFLSBUF命令。怎样算较大的块设备?一般理解为当前Linux系统可用的物理内存小于操作的块设备大小。

5. 使用drop_caches控制Cache和Buffer

/proc是一个虚拟文件系统,我们可以通过对它的读写操作作为与kernel实体间进行通信的一种手段.也就是说可以通过修改/proc中的文件来对当前kernel的行为做出调整。关于Cache和Buffer的控制，我们可以通过echo  1 > /proc/sys/vm/drop_caches进行操作。

首先来看下内核源码实现：

int drop_caches_sysctl_handler(ctl_table *table, int write,  
 void __user *buffer, size_t *length, loff_t *ppos)  
{  
 int ret;  
  
 
 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);  
 if (ret)  
 return ret;  
 if (write) {  
 /* => echo 1 > /proc/sys/vm/drop_caches 清理页缓存 */  
 if (sysctl_drop_caches & 1)  
 /* => 遍历所有的超级块，清理所有的缓存 */  
 iterate_supers(drop_pagecache_sb, NULL);  
 if (sysctl_drop_caches & 2)  
 drop_slab();  
 }  
 return 0;  
}  
  
 
/**  
 * iterate_supers - call function for all active superblocks  
 * @f: function to call  
 * @arg: argument to pass to it  
 *  
 * Scans the superblock list and calls given function, passing it  
 * locked superblock and given argument.  
 */  
void iterate_supers(void (*f)(struct super_block *, void *), void *arg)  
{  
 struct super_block *sb, *p = NULL;  
  
 
 spin_lock(&sb_lock);  
 list_for_each_entry(sb, &super_blocks, s_list) {  
 if (hlist_unhashed(&sb->s_instances))  
 continue;  
 sb->s_count++;  
 spin_unlock(&sb_lock);  
  
 
 down_read(&sb->s_umount); 
 if (sb->s_root && (sb->s_flags & MS_BORN))  
 f(sb, arg);  
 up_read(&sb->s_umount); 
  
 
 spin_lock(&sb_lock);  
 if (p) 
 __put_super(p);  
 p = sb;  
 }  
 if (p)  
 __put_super(p);  
 spin_unlock(&sb_lock); 
} 
  
 
/* => 清理文件系统(包括bdev伪文件系统)的页缓存 */  
static void drop_pagecache_sb(struct super_block *sb, void *unused)  
{  
 struct inode *inode, *toput_inode = NULL;  
  
 
 spin_lock(&inode_sb_list_lock);  
 /* => 遍历所有的inode */  
 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {  
 spin_lock(&inode->i_lock);  
 /* 
 * => 若当前状态为(I_FREEING|I_WILL_FREE|I_NEW) 或  
 * => 若没有缓存页  
 * => 则跳过  
 */  
 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||  
 (inode->i_mapping->nrpages == 0)) { 
 spin_unlock(&inode->i_lock);  
 continue;  
 }  
 __iget(inode);  
 spin_unlock(&inode->i_lock);  
 spin_unlock(&inode_sb_list_lock);  
 /* => 清除缓存页(除了脏页、上锁的、正在回写的或映射在页表中的)*/  
 invalidate_mapping_pages(inode->i_mapping, 0, -1);  
 iput(toput_inode);  
 toput_inode = inode;  
 spin_lock(&inode_sb_list_lock);  
 } 
 spin_unlock(&inode_sb_list_lock);  
 iput(toput_inode);  
}

综上，echo 1 >  /proc/sys/vm/drop_caches会清除所有inode的缓存页，这里的inode包括VFS的inode、所有文件系统inode(也包括bdev伪文件系统块设备的inode的缓存页)。所以该命令执行后，就会将整个系统的page  cache和buffer cache全部清除，当然前提是这些cache都是非脏的、没有正被使用的。

接下来看下实际效果：

[root@test usb] # cat /proc/meminfo 
MemTotal: 90516 kB 
MemFree: 12396 kB 
Buffers: 96 kB 
Cached: 60756 kB 
[root@test usb] # busybox_v400 sync 
[root@test usb] # busybox_v400 sync 
[root@test usb] # busybox_v400 sync 
[root@test usb] # echo 1 > /proc/sys/vm/drop_caches 
[root@test usb] # cat /proc/meminfo 
MemTotal: 90516 kB 
MemFree: 68820 kB 
Buffers: 12 kB 
Cached: 4464 kB

可以看到Buffers和Cached都降了下来，在drop_caches前建议执行sync命令，以确保数据的完整性。sync  命令会将所有未写的系统缓冲区写到磁盘中，包含已修改的 i-node、已延迟的块 I/O 和读写映射文件等。

上面的设置虽然简单但是比较粗暴，使cache的作用基本无法发挥，尤其在系统压力比较大时进行drop  cache处理容易产生问题。因为drop_cache是全局在清内存，清的过程会加页面锁，导致有些进程等页面锁时超时，导致问题发生。因此，需要根据系统的状况进行适当的调节寻找最佳的方案。

6. 经验总结

以上分别讨论了Cache和Buffer分别从哪里来?什么时候消耗在哪里?如何分别控制Cache和Buffer这三个问题。最后还介绍了vmtouch工具的使用。

要深入理解Linux的Cache和Buffer牵涉大量内核核心机制(VFS、内存管理、块设备驱动、页高速缓存、文件访问、页框回写)，需要制定计划在后续工作中不断理解和消化。

相关推荐：《Linux视频教程》

The above is the detailed content of What is linux buffer. For more information, please follow other related articles on the PHP Chinese website!