Description of problem: arm64架构下, 64K页时 failover, domain 测试失败 Version-Release number of selected component (if applicable): openanolis kernel release/release-5.10.134-15.y How reproducible: domain-id 测试: 使用https://github.com/lostjeffle/demand-read-cachefilesd/tree/main项目, 在run.sh脚本里添加domain-id的挂载选项: mount -t erofs none -o fsid=${_bootstrap} -o device=${_datablob} -o domain_id="test1" ${mntdir2} 然后执行run.sh ./run.sh ../src ../mntdir/ ../fscachedir/ ../mntdir2 输出如下: mkfs.erofs 1.6 Build completed. [HEADER] id 0, opcode 0 [OPEN] volume key erofs,test (volume_key_size 11), cookie key test.img (cookie_key_size 8), object id 1, fd 4, flags 0 Writing cmd: copen 0,4096 [HEADER] id 0, opcode 2 [READ] object_id 1, fd 4, src_path test.img, off 0, len 10000 read src image failed, ret 4096, 0 (Success) ^C 内核输出 $ dmesg|tail [ 34.675418] FS-Cache: Loaded [ 34.944592] FS-Cache: Netfs 'erofs' registered for caching [ 34.972669] erofs: (device erofs): mounted with root inode @ nid 128. [ 72.415045] erofs: (device erofs): erofs_fscache_meta_readpage: erofs_fscache_meta_readpage: -105 [ 72.415808] erofs: (device erofs): erofs_read_superblock: cannot read erofs superblock [ 72.437722] erofs: (device erofs): erofs_fscache_meta_readpage: erofs_fscache_meta_readpage: -105 [ 72.438407] erofs: (device erofs): erofs_read_superblock: cannot read erofs superblock [ 96.285052] CacheFiles: Loaded [ 98.723166] FS-Cache: Cache "test" added (type cachefiles) [ 98.723170] CacheFiles: File cache on dm-0 registered failover测试: 测试程序: https://github.com/userzj/demand-read-cachefilesd/commits/failover-test $ cd demand-read-cachefilesd-failover-test/test $ ./test-private-01.sh mount: /mnt: can't read superblock on none. cachefilesd mount failed $ dmesg|tail [ 6.058068] Adding 4222912k swap on /dev/mapper/uos-swap. Priority:-2 extents:1 across:4222912k FS [ 6.063951] xfs filesystem being remounted at / supports timestamps until 2038 (0x7fffffff) [ 6.077982] systemd-journald[627]: Received client request to flush runtime journal. [ 6.114386] XFS (sda2): Mounting V5 Filesystem [ 6.175611] XFS (sda2): Ending clean mount [ 6.176231] xfs filesystem being mounted at /boot supports timestamps until 2038 (0x7fffffff) [ 30.900020] FS-Cache: Loaded [ 31.171706] FS-Cache: Netfs 'erofs' registered for caching [ 31.172114] erofs: (device erofs): erofs_fscache_meta_readpage: erofs_fscache_meta_readpage: -105 [ 31.172847] erofs: (device erofs): erofs_read_superblock: cannot read erofs superblock
在龙芯16k页大小下，这2项也会测试失败。
目前 erofs 镜像 size 小于 PAGE_SIZE 的时候，fscache 的实现确实存在问题 1. 首先测试使用的 demand-read-cachefilesd 的实现存在问题，内核在触发按需读的时候，会向 daemon 请求读取 PAGE_SIZE 例如 64K 的数据，daemon 在接收到这个请求的时候，先从 erofs 镜像读取数据，然后将读取的数据写入 cachefiles。daemon 在读取 erofs 镜像的过程中，发现 erofs 镜像的大小（例如只有 4KB）小于内核请求的大小（例如 PAGE_SIZE 64K）的时候，daemon 没有处理这种情况，就会报错 hang 在那边 2. 另外就算 daemon 修复了上述问题，当 erofs 镜像 size 小于 PAGE_SIZE 的时候，fscache 的目前的实现都是 break 的，需要在内核层面进行修复（linux 内核主线也存在该问题） 由于目前 fscache 尚未实际场景使用，而且上述问题只在特定场景存在，由于当前人力原因，建议先标记为已知问题，待后续追踪。