kernel/git/stable/linux.git/fs/smb/server/oplock.c, branch master

ksmbd: snapshot previous oplock state before durable checks

2026-07-01T02:29:46+00:00

smb_grant_oplock() checks the previous oplock holder's o_fp to decide whether a durable handle should be invalidated when the oplock break cannot be delivered. prev_opinfo is obtained with opinfo_get_list(), which pins only the oplock_info. It does not pin the ksmbd_file stored in opinfo->o_fp. A concurrent last close can unlink the opinfo from ci->m_op_list under ci->m_lock and then free the ksmbd_file. The oplock_info can still be kept alive by the refcount taken by opinfo_get_list(), but o_fp may already point at freed memory by the time smb_grant_oplock() reads is_durable, conn, or tcon. Snapshot the previous holder's durable state while ci->m_lock is held, then use only the copied values after dropping the lock. This keeps the o_fp lifetime tied to the inode lock without taking an extra ksmbd_file reference. Taking such a reference is unsafe here because smb_grant_oplock() does not necessarily have the previous holder's session work, and dropping the temporary reference can otherwise become the final putter. Fixes: 26fa88dc877c ("ksmbd: invalidate durable handles on oplock break") Reported-by: Gil Portnoy Signed-off-by: Namjae Jeon Signed-off-by: Steve French

ksmbd: don't hold ci->m_lock while waiting for a lease break ack

2026-07-01T02:29:46+00:00

When a cifs.ko client caches a read-handle (RH) lease via deferred close and a conflicting open arrives, ksmbd breaks the lease and waits for the acknowledgment in wait_for_break_ack() for up to OPLOCK_WAIT_TIME (35s). __smb_break_all_levII_oplock() runs that wait while holding ci->m_lock for read. cifs.ko reacts to a handle-lease break by closing the deferred handle rather than sending a lease break acknowledgment. That close path (close_id_del_oplock() -> opinfo_del()) takes ci->m_lock for write and is exactly what would wake the waiter, but it blocks on the read lock held by the waiting thread. The break is then resolved only by the 35s timeout, so xfstests generic/001 takes ~78s with leases enabled versus ~4s with oplocks only. Collect the target opinfos (each pinned with a reference) while holding ci->m_lock, then break them after releasing it, matching how smb_grant_oplock() already breaks a conflicting lease using only a reference. The reference keeps the opinfo (and its conn and lease) alive across the unlocked window, and a close racing the break is handled by the existing OPLOCK_CLOSING state check. Apply the same fix to the parent lease break paths. Signed-off-by: Namjae Jeon Signed-off-by: Steve French

ksmbd: fix kernel-doc warnings in smb2_lease_break_noti()

2026-06-23T22:52:42+00:00

kernel test robot report missing kernel-doc descriptions for the 'wait_ack' and 'inc_epoch' parameters of smb2_lease_break_noti(): Warning: fs/smb/server/oplock.c:937 function parameter 'wait_ack' not described in 'smb2_lease_break_noti' Warning: fs/smb/server/oplock.c:937 function parameter 'inc_epoch' not described in 'smb2_lease_break_noti' Document both parameters to silence the warnings. Reported-by: kernel test robot Signed-off-by: Namjae Jeon Signed-off-by: Steve French

ksmbd: fix inconsistent indenting warnings

2026-06-23T22:52:42+00:00

Detected by Smatch. fs/smb/server/oplock.c:1446 smb_grant_oplock() warn: inconsistent indenting Reported-by: Dan Carpenter Signed-off-by: Namjae Jeon Signed-off-by: Steve French

ksmbd: treat read-control opens as stat opens only for leases

2026-06-23T01:15:06+00:00

A second open that requests only metadata-level access must not break the existing caching state. ksmbd already skips the break for such opens via fp->attrib_only (FILE_READ_ATTRIBUTES, FILE_WRITE_ATTRIBUTES and FILE_SYNCHRONIZE). An open requesting only READ_CONTROL (reading the security descriptor) must be treated differently depending on the existing caching state. smbtorture smb2.lease.statopen4 expects a read-control open NOT to break a caching lease, while smb2.oplock.statopen1 expects the same open to break a batch oplock. So READ_CONTROL is a stat open for leases but not for oplocks. Extend the stat-open break-skip in smb_grant_oplock() to also cover a read-control-only open, but only when the existing holder is a lease. The global fp->attrib_only flag (used for share-mode, rename and truncate decisions) is left unchanged so oplock behaviour is preserved. Signed-off-by: Namjae Jeon Signed-off-by: Steve French

ksmbd: break conflicting-open leases only as far as needed

2026-06-23T01:15:06+00:00

smb2.lease.oplock and smb2.lease.breaking1 hold a lease and then issue a single conflicting open on the same file. The held lease must break one step to drop write caching (RWH->RH, RW->R) and then stop, so lease_break_info.count is 1 and the lease keeps its read/handle caching. ksmbd instead cascaded the break all the way down to none (e.g. RWH->RH->R->none), so the break count was 2 or 3 and the reported lease state ended at 0. Commit "chain pending lease breaks before waking waiters" forces break_level to SMB2_OPLOCK_LEVEL_NONE for any non-lease open against a handle-caching lease, which drives oplock_break()'s retry loop down to none even when only one open is contending. Drop that break_level override so a conflicting open breaks a lease only to its own compatible level (level II, i.e. RH/R). A deeper break is still required when a truncating open is also waiting behind the same lease break. smb2.lease.breaking3 keeps a normal open pending through RWH->RH and an overwrite open pending behind it, and expects the lease to continue RH->R->none before either open completes. The overwrite waiter sets open_trunc on the lease while it blocks on the pending break, so extend the retry loop to chain another break while that truncating waiter still needs the lease at none. The per-break open_trunc snapshot stays cleared, so the cascade steps down (RH->R->none) instead of collapsing straight to none, and the normal open stays pending until the lease is fully broken. Signed-off-by: Namjae Jeon Signed-off-by: Steve French

ksmbd: break handle caching for share conflicts

2026-06-23T01:15:06+00:00

smb2.lease.break_twice first opens a file with an RHW lease and then tries a second open with restrictive sharing. That open must fail with a sharing violation, but the existing lease should be broken from RHW to RW because only handle caching conflicts with the requested sharing. ksmbd used the normal write-cache break calculation for this path, so RHW was broken to RH. The following successful open then did not generate the expected second break from RW to R. Pass share-conflict context into the lease break helper and, for lease breaks caused by sharing, drop only SMB2_LEASE_HANDLE_CACHING from the current lease state. Other break paths keep the existing write/truncate break behavior. A share-conflict break must also remain a single break. The triggering open fails with a sharing violation and is never granted, so there is no target oplock level to converge on. The lease break retry loop, however, keeps breaking while the lease level is still above req_op_level, which broke RHW all the way down to R in one open (lease_break_info.count became 2 instead of 1). Skip the again loop for share-conflict breaks so the sharing open produces exactly one RHW->RW break and the later successful open produces the separate RW->R break the test expects. Signed-off-by: Namjae Jeon Signed-off-by: Steve French

ksmbd: normalize ungrantable lease states

2026-06-23T01:15:06+00:00

smb2.lease.request verifies which SMB2 lease state combinations are granted by the server. Requests for H-only, W-only, and HW leases are valid lease state bitmasks, but they are not grantable combinations and should be returned as lease state none. ksmbd only checked that the requested bits were inside the SMB2 lease state mask. As a result it could grant H-only, W-only, or HW requests and return non-zero lease states where the client expects no lease. Keep the bitmask validation, but normalize ungrantable combinations to zero before allocating or looking up the lease. The grantable combinations remain unchanged: R, RH, RW, and RHW. Signed-off-by: Namjae Jeon Signed-off-by: Steve French

ksmbd: avoid level II oplock break notification on unlink

2026-06-23T01:15:05+00:00

smb2_util_unlink() opens the target with FILE_DELETE_ON_CLOSE and then closes that handle. Other clients can also mark a file for delete with SMB2 SET_INFO FileDispositionInformation. When these unlink paths break existing SMB2 level II oplocks, ksmbd sends an unsolicited SMB2_OPLOCK_BREAK notification to none. This races with the synchronous CREATE or SET_INFO response expected by the client, and smbtorture reports NT_STATUS_INVALID_NETWORK_RESPONSE while running smb2.oplock.exclusive2. SMB2 level II oplock breaks do not require an acknowledgment in the delete path. Keep lease handling unchanged, but drop plain SMB2 level II oplocks locally for unlink requests without sending a break notification. Normal write/truncate paths still send the level II to none notification, preserving the behavior covered by smb2.oplock.levelII500. Signed-off-by: Namjae Jeon Signed-off-by: Steve French

ksmbd: downgrade oplock after break timeout

2026-06-23T01:15:05+00:00

smb2.oplock.batch22a opens a file with a batch oplock and then issues a second open that waits for the oplock break timeout. After the timeout the second open should succeed, but the granted oplock level must be level II. When the break times out, oplock_break() returns -ENOENT after invalidating the previous opener. smb_grant_oplock() went straight to set_lev with the original requested oplock level, so the second open could be granted a new batch oplock. Downgrade the requested oplock to level II on the -ENOENT break-timeout path before granting the oplock to the new open. A break that completes because the previous owner closed its handle from the oplock break handler must be distinguished from a real timeout. smb2.oplock.batch7 closes the first handle during the break wait, and the second open is then expected to be granted the originally requested batch oplock. Return -EAGAIN from the non-lease break path when the previous opener closed during the break wait, recheck sharing in smb_grant_oplock(), and grant the requested oplock if the close removed the conflict. Real break timeouts still return -ENOENT and keep the downgrade to level II. Signed-off-by: Namjae Jeon Signed-off-by: Steve French