Introduce a new fadvise flag to drop page cache pages of a single filesystem. At the moment it is possible to drop page cache pages via /proc/sys/vm/drop_pagecache or via posix_fadvise(POSIX_FADV_DONTNEED). The first method drops the whole page cache while the second can be used to drop page cache pages of a single file descriptor. However, there's not a simple way to drop all the pages of a filesystem (we could scan all the file descriptors and use posix_fadvise(POSIX_FADV_DONTNEED), but this solution obviously doesn't scale well). This functionality requires root privilege to avoid potential DoS in the system (i.e., a hard loop of posix_fadvise(POSIX_FADV_DONTNEED_FS) on the root filesystem). A practical example: # ls -lh /mnt/sda/zero /mnt/sdb/zero -rw-r--r-- 1 root root 16M 2011-04-20 10:20 /mnt/sda/zero -rw-r--r-- 1 root root 16M 2011-04-20 10:20 /mnt/sdb/zero $ grep ^Cached /proc/meminfo Cached: 5660 kB $ md5sum /mnt/sda/zero /mnt/sdb/zero 2c7ab85a893283e98c931e9511add182 /mnt/sda/zero 2c7ab85a893283e98c931e9511add182 /mnt/sdb/zero $ grep ^Cached /proc/meminfo Cached: 38544 kB $ sudo ./drop-pagecache /mnt/sda/ $ grep ^Cached /proc/meminfo Cached: 22440 kB $ sudo ./drop-pagecache /mnt/sdb/ $ grep ^Cached /proc/meminfo Cached: 5056 kB A previous RFC about this topic can be found here: http://marc.info/?l=linux-kernel&m=130385374902114&w=2 ChangeLog (v1 -> v2): * use the same value for POSIX_FADV_DONTNEED_FS on all architectures * check CAP_SYS_ADMIN capability instead of checking the EUID value Signed-off-by: Andrea Righi <andrea@xxxxxxxxxxxxxxx> --- fs/drop_caches.c | 2 +- include/linux/fadvise.h | 1 + include/linux/mm.h | 2 ++ mm/fadvise.c | 7 +++++++ 4 files changed, 11 insertions(+), 1 deletions(-) diff --git a/fs/drop_caches.c b/fs/drop_caches.c index 98b77c8..59d6caa 100644 --- a/fs/drop_caches.c +++ b/fs/drop_caches.c @@ -13,7 +13,7 @@ /* A global variable is a bit ugly, but it keeps the code simple */ int sysctl_drop_caches; -static void drop_pagecache_sb(struct super_block *sb, void *unused) +void drop_pagecache_sb(struct super_block *sb, void *unused) { struct inode *inode, *toput_inode = NULL; diff --git a/include/linux/fadvise.h b/include/linux/fadvise.h index e8e7471..ab39117 100644 --- a/include/linux/fadvise.h +++ b/include/linux/fadvise.h @@ -17,5 +17,6 @@ #define POSIX_FADV_DONTNEED 4 /* Don't need these pages. */ #define POSIX_FADV_NOREUSE 5 /* Data will be accessed once. */ #endif +#define POSIX_FADV_DONTNEED_FS 8 /* Don't need these filesystem pages. */ #endif /* FADVISE_H_INCLUDED */ diff --git a/include/linux/mm.h b/include/linux/mm.h index 692dbae..004cdbc 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -21,6 +21,7 @@ struct anon_vma; struct file_ra_state; struct user_struct; struct writeback_control; +struct super_block; #ifndef CONFIG_DISCONTIGMEM /* Don't use mapnrs, do it properly */ extern unsigned long max_mapnr; @@ -1602,6 +1603,7 @@ int in_gate_area_no_mm(unsigned long addr); #define in_gate_area(mm, addr) ({(void)mm; in_gate_area_no_mm(addr);}) #endif /* __HAVE_ARCH_GATE_AREA */ +void drop_pagecache_sb(struct super_block *sb, void *unused); int drop_caches_sysctl_handler(struct ctl_table *, int, void __user *, size_t *, loff_t *); unsigned long shrink_slab(unsigned long scanned, gfp_t gfp_mask, diff --git a/mm/fadvise.c b/mm/fadvise.c index 8d723c9..15155e7 100644 --- a/mm/fadvise.c +++ b/mm/fadvise.c @@ -57,6 +57,7 @@ SYSCALL_DEFINE(fadvise64_64)(int fd, loff_t offset, loff_t len, int advice) case POSIX_FADV_WILLNEED: case POSIX_FADV_NOREUSE: case POSIX_FADV_DONTNEED: + case POSIX_FADV_DONTNEED_FS: /* no bad return value, but ignore advice */ break; default: @@ -127,6 +128,12 @@ SYSCALL_DEFINE(fadvise64_64)(int fd, loff_t offset, loff_t len, int advice) invalidate_mapping_pages(mapping, start_index, end_index); break; + case POSIX_FADV_DONTNEED_FS: + if (capable(CAP_SYS_ADMIN)) + drop_pagecache_sb(file->f_dentry->d_sb, NULL); + else + ret = -EPERM; + break; default: ret = -EINVAL; } -- 1.7.1 -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
