[LU-16812] ldiskfs osd_punch() should check new inode size does not exceed maximum size

• Type: Improvement
• Resolution: Unresolved
• Priority: Minor
• Severity: 3
• Assignee: Dongyang Li
• Reporter: Dongyang Li

Description

On osd we don't check for new inode size before ldiskfs_truncate(), we could overflow the block counters in ext4_es_remove_extent().

Issue Links

• is duplicated by: LU-14384 data "written" to last stripe of very large ldiskfs object can not be "read"
• is related to: LU-16537 osd-ldiskfs prevents writing of last block of very large ldiskfs file

Comments

[Andreas Dilger]

It seems kind of wrong that we allow creating a file with size=MAX_LFS_FILESIZE but then we cannot open it? It would probably make more sense that we would return an error for the truncate operation for files larger than LOV_MAX_STRIPE_COUNT * ocd_maxbytes at the client.

Something like this is already in https://review.whamcloud.com/41365 "LU-14384 tests: write and read last object stripe"

[Gerrit Updater]

"Li Dongyang <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/50898
Subject: LU-16812 osd-ldiskfs: check inode new size before truncate
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: b68d6ea613477d39ecdadc8d8131554a1a5bf27d