xfs

  4)               |  xfs_btree_lookup [xfs]() {
  4)   0.066 us    |    xfs_refcountbt_init_ptr_from_cur [xfs]();
  4)               |    xfs_btree_lookup_get_block [xfs]() {
  4)   0.065 us    |      xfs_btree_ptr_to_daddr [xfs]();
  4)               |      xfs_btree_read_buf_block.constprop.0 [xfs]() {
  4)   0.062 us    |        xfs_btree_ptr_to_daddr [xfs]();
  4)               |        xfs_trans_read_buf_map [xfs]() {
  4)               |          xfs_buf_read_map [xfs]() {
  4)               |            xfs_buf_get_map [xfs]() {
  4)               |              xfs_perag_get [xfs]() {
  4)   0.059 us    |                __rcu_read_lock();
  4)   0.058 us    |                __rcu_read_unlock();
  4)   0.277 us    |              }
  4)   0.059 us    |              __rcu_read_lock();
  4)   0.056 us    |              __rcu_read_unlock();
  4)               |              xfs_buf_find_lock [xfs]() {
  4)               |                xfs_buf_lock [xfs]() {
  4)               |                  down() {
  4)   0.057 us    |                    __cond_resched();
  4)   0.074 us    |                    _raw_spin_lock_irqsave();
  4)   0.067 us    |                    _raw_spin_unlock_irqrestore();
  4)   0.460 us    |                  }
  4)   0.566 us    |                }
  4)   0.680 us    |              }
  4)   0.062 us    |              xfs_perag_put [xfs]();
  4)   1.598 us    |            }
  4)   1.713 us    |          }
  4)   1.823 us    |        }
  4)               |        xfs_btree_set_refs [xfs]() {
  4)   0.059 us    |          xfs_buf_set_ref [xfs]();
  4)   0.176 us    |        }
  4)   2.293 us    |      }
  4)   0.155 us    |      xfs_btree_setbuf [xfs]();
  4)   2.833 us    |    }
  4)               |    xfs_lookup_get_search_key [xfs]() {
  4)   0.318 us    |      xfs_btree_key_offset [xfs]();
  4)   0.428 us    |    }

看上去还是会访问磁盘的

@[
    __xfs_buf_submit+233
    xfs_buf_read_map+301
    xfs_buf_readahead_map+60
    xfs_btree_readahead_sblock.isra.0+141
    xfs_btree_increment+68
    xfs_refcount_find_shared+479
    xfs_reflink_find_shared+137
    xfs_reflink_trim_around_shared+217
    xfs_buffered_write_iomap_begin+1102
    iomap_iter+381
    iomap_file_buffered_write+147
    xfs_file_buffered_write+135
    io_write+284
    io_issue_sqe+96
    io_submit_sqes+490
    __do_sys_io_uring_enter+1489
    do_syscall_64+59
    entry_SYSCALL_64_after_hwframe+110
]: 2

才意识到文件系统也是存在对下屏蔽错误的操作:

/sys/fs/xfs/nvme1n1/error/metadata/EIO/max_retries
/sys/fs/xfs/nvme1n1/error/metadata/ENODEV/max_retries
/sys/fs/xfs/nvme1n1/error/metadata/default/max_retries
/sys/fs/xfs/nvme1n1/error/metadata/ENOSPC/max_retries

原来 xfs 的工具

xfs_admin      xfs_db         xfs_freeze     xfs_info       xfs_mdrestore  xfs_ncheck     xfs_rtcp
xfs_spaceman
xfs_bmap       xfsdist.bt     xfs_fsr        xfs_io         xfs_metadump   xfs_quota      xfs_scrub
xfs_copy       xfs_estimate   xfs_growfs     xfs_logprint   xfs_mkfile     xfs_repair     xfs_scrub_all

xfs_info .
meta-data=/dev/nvme1n1p2         isize=512    agcount=4, agsize=48837846 blks
         =                       sectsz=512   attr=2, projid32bit=1
         =                       crc=1        finobt=1, sparse=1, rmapbt=1
         =                       reflink=1    bigtime=1 inobtcount=1 nrext64=1
data     =                       bsize=4096   blocks=195351382, imaxpct=25
         =                       sunit=0      swidth=0 blks
naming   =version 2              bsize=4096   ascii-ci=0, ftype=1
log      =internal log           bsize=4096   blocks=95386, version=2
         =                       sectsz=512   sunit=0 blks, lazy-count=1

无法降级，而且还默认打开，太坑了

https://man7.org/linux/man-pages/man8/xfs_admin.8.html

nrext64 Upgrade a filesystem to support large per-inode extent counters. The maximum data fork extent count will be 2^48 - 1, while the maximum attribute fork extent count will be 2^32 - 1. The filesystem cannot be downgraded after this feature is enabled. Once enabled, the filesystem will not be mountable by older kernels. This feature was added to Linux 5.19.

本站所有文章转发 CSDN 将按侵权追究法律责任，其它情况随意。