cephfs kernel driver netfs readahead crash

[grepgrokmore]

I've hit this several times over the last few weeks. Using hyperconverged configuration in proxmox.

It appears that when ceph switched to using netfs readahead there may have been some problems introduced. I realized that there is a cephfs mount option rasize=0 that should disable readahead. I'll try for a few days with this option enabled ...

/etc/pve/storage.cfg file needs rasize added to the options list

ie:

cephfs: cephfs
        path /mnt/pve/cephfs
        content vztmpl,backup,iso,snippets,import
        fs-name cephfs
        options noatime,rsize=8388608,wsize=8388608,read_from_replica=balance,rasize=0
        prune-backups keep-all=1

Oops: general protection fault, probably for non-canonical address 0xefff8f0ca4a53a08: 0000 [#1] SMP PTI
CPU: 43 UID: 100200 PID: 1774147 Comm: ffprobe Tainted: P           OE       6.17.13-1-pve #1 PREEMPT(voluntary)
Tainted: [P]=PROPRIETARY_MODULE, [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
Hardware name: LENOVO 30BDS0M202/1038, BIOS S05KT76A 01/16/2025
RIP: 0010:__kvmalloc_node_noprof+0x1cc/0x640
Code: cc b5 a8 e9 f5 fe ff ff 4d 85 c0 0f 84 61 ff ff ff 48 85 db 0f 84 58 ff ff ff bf ff ff ff ff 41 8b 45 28 49 8b 75 00 4c 01 c0 <48> 8b 18 48 89 c1 49 33 9d b8 00 00 00 4c 89 c0 48 0f c9 48 31 cb
RSP: 0018:ffffd53170a272a0 EFLAGS: 00010282
RAX: efff8f0ca4a53a08 RBX: fffff977e49294c0 RCX: 0000000000000000
RDX: 000000000056402b RSI: ffffffffa9cb20a0 RDI: 00000000ffffffff
RBP: ffffd53170a27300 R08: efff8f0ca4a53a00 R09: ffffd53170a27408
R10: ffffd53170a27318 R11: ffff8f0c09f56800 R12: 0000000000000010
R13: ffff8ee480042500 R14: 0000000000000cc0 R15: 0000000000000cc0
FS:  0000717f3fc2c600(0000) GS:ffff8f2395f06000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00005fcbf5bb8b00 CR3: 00000020bf494006 CR4: 00000000007726f0
PKRU: 55555554>
Call Trace:
 <TASK>
 ? __iov_iter_get_pages_alloc+0x78b/0x7d0
 ? osd_req_op_extent_init+0x25/0xa0 [libceph]
 __iov_iter_get_pages_alloc+0x78b/0x7d0
 ? __iov_iter_get_pages_alloc+0x78b/0x7d0
 iov_iter_get_pages_alloc2+0x20/0x70
 ? __ceph_alloc_sparse_ext_map+0x31/0x70 [libceph]
 ceph_netfs_issue_read+0x3df/0x790 [ceph]
 netfs_read_to_pagecache+0x18e/0x4e0 [netfs]
 netfs_readahead+0x15b/0x270 [netfs]
 read_pages+0x81/0x240
 page_cache_ra_unbounded+0x1f6/0x230
 page_cache_ra_order+0x37c/0x3b0

 

[grepgrokmore]

I should note that as far as I know, this only occurs when using encrypted inodes (fscrypt)