[LU-16579] llite: Fix the wrong ending offset calculation

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.16.0
• Fix Version/s: Lustre 2.16.0, Lustre 2.15.3
• Type: Bug
• Priority: Major
• Reporter: Qian Yingjin
• Assignee: Qian Yingjin
• Resolution: Fixed
• Severity: 3

Description

For a single-stripe file, after applied https://review.whamcloud.com/c/fs/lustre-release/+/49226
Subject: LU-16338 readahead: clip readahead with kms,
It cause a dead loop:
https://testing.whamcloud.com/test_logs/8d73c4e7-0e6e-482c-b93c-bf7159706890/show_text

00000080:00200000:1.0:1676614736.277040:0:21215:0:(rw.c:1981:ll_readpage()) pgno:260, cnt:1032192, pos:0
00000080:00000001:1.0:1676614736.277041:0:21215:0:(vvp_io.c:1654:vvp_io_read_ahead()) Process entered
00000080:00000001:1.0:1676614736.277041:0:21215:0:(vvp_io.c:1666:vvp_io_read_ahead()) Process leaving (rc=0 : 0 : 0)
00000008:00000001:1.0:1676614736.277042:0:21215:0:(osc_io.c:83:osc_io_read_ahead()) Process entered
00000008:00000001:1.0:1676614736.277042:0:21215:0:(osc_lock.c:1281:osc_obj_dlmlock_at_pgoff()) Process entered
00000008:00000001:1.0:1676614736.277043:0:21215:0:(osc_request.c:3137:osc_match_base()) Process entered
00000008:00000001:1.0:1676614736.277043:0:21215:0:(osc_request.c:3172:osc_match_base()) Process leaving (rc=4 : 4 : 4)
00000008:00000001:1.0:1676614736.277044:0:21215:0:(osc_lock.c:1315:osc_obj_dlmlock_at_pgoff()) Process leaving (rc=18446620395345229440 : -123678364322176 : ffff8f83e3073680)
00000008:00000001:1.0:1676614736.277045:0:21215:0:(osc_io.c:112:osc_io_read_ahead()) Process leaving (rc=0 : 0 : 0)
00000080:00000001:1.0:1676614736.277045:0:21215:0:(rw.c:2012:ll_readpage()) Process leaving (rc=524289 : 524289 : 80001)
00000080:00000001:1.0:1676614736.277046:0:21215:0:(rw.c:1873:ll_readpage()) Process entered
00000080:00200000:1.0:1676614736.277046:0:21215:0:(rw.c:1981:ll_readpage()) pgno:260, cnt:1032192, pos:0
00000080:00000001:1.0:1676614736.277047:0:21215:0:(vvp_io.c:1654:vvp_io_read_ahead()) Process entered
00000080:00000001:1.0:1676614736.277047:0:21215:0:(vvp_io.c:1666:vvp_io_read_ahead()) Process leaving (rc=0 : 0 : 0)
00000008:00000001:1.0:1676614736.277048:0:21215:0:(osc_io.c:83:osc_io_read_ahead()) Process entered
00000008:00000001:1.0:1676614736.277048:0:21215:0:(osc_lock.c:1281:osc_obj_dlmlock_at_pgoff()) Process entered
00000008:00000001:1.0:1676614736.277049:0:21215:0:(osc_request.c:3137:osc_match_base()) Process entered
00000008:00000001:1.0:1676614736.277049:0:21215:0:(osc_request.c:3172:osc_match_base()) Process leaving (rc=4 : 4 : 4)
00000008:00000001:1.0:1676614736.277050:0:21215:0:(osc_lock.c:1315:osc_obj_dlmlock_at_pgoff()) Process leaving (rc=18446620395345229440 : -123678364322176 : ffff8f83e3073680)
00000008:00000001:1.0:1676614736.277051:0:21215:0:(osc_io.c:112:osc_io_read_ahead()) Process leaving (rc=0 : 0 : 0)
00000080:00000001:1.0:1676614736.277051:0:21215:0:(rw.c:2012:ll_readpage()) Process leaving (rc=524289 : 524289 : 80001)

The dead loop code lines:

if (cl_offset(clob, vmpage->index) >= iter->count + iocb->ki_pos) {
                        result = cl_io_read_ahead(env, io, vmpage->index, &ra);
                        if (result < 0 || vmpage->index > ra.cra_end_idx) {
                                cl_read_ahead_release(env, &ra);
                                unlock_page(vmpage);
                                RETURN(AOP_TRUNCATED_PAGE); ===>AOP_TRUNCATED_PAGE = 80001
                        }
                }

After investigated, I found the ending beyond offset calculation is wrong as each time finished the read of a page, it will advance the @iter->count (iter->count - read_bytes).
The wrong ending offset results in the dead loop described above.

Issue Links

• is related to: LU-16412 check truncated page in ->read page()
• is related to: LU-16338 read-ahead more than file size for a ...

Comments

[Andreas Dilger]

Yingjin, since the LU-16338 patch has not landed yet, why not just fix this as part of that patch? Is this issue caused by this patch, or just an existing bug exposed by that change?

[Gerrit Updater]

"Qian Yingjin <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/50065
Subject: LU-16579 llite: fix the wrong beyond read end calculation
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: c4afbee4c795d5305827f3ce533f3d04474a18eb

[Qian Yingjin]

I think it is an existing bug exposed by that change.

[Qian Yingjin]

Hi Patrick,

They are all not applied to ES5.2.

[Gerrit Updater]

"Andreas Dilger <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/50278
Subject: LU-16579 llite: fix the wrong beyond read end calculation
Project: fs/lustre-release
Branch: b2_15
Current Patch Set: 1
Commit: 88e538a9c6f7444fde284960d56ed62d4f17cb3a

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/50065/
Subject: LU-16579 llite: fix the wrong beyond read end calculation
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: ae356dc325877bd130ad94acc5f3610898de8a8a

[Peter Jones]

Landed for 2.16

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/50278/
Subject: LU-16579 llite: fix the wrong beyond read end calculation
Project: fs/lustre-release
Branch: b2_15
Current Patch Set:
Commit: 9c8a80bca738884e09affd66837b9e94508664d1