d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add non-blocking trylock variants for the rangelock functions.

Browse Source 
A future patch that will add a Linux compatible copy_file_range(2) syscall
needs to be able to lock the byte ranges of two files concurrently.
To do this without a risk of deadlock, a non-blocking variant of
vn_rangelock_rlock() called vn_rangelock_tryrlock() was needed.
This patch adds this, along with vn_rangelock_trywlock(), in order to
do this.
The patch also adds a couple of comments, that I hope clarify how the
algorithm used in kern_rangelock.c works.

Reviewed by:	kib, asomers (previous version)
Differential Revision:	https://reviews.freebsd.org/D20645
This commit is contained in:
Rick Macklem 2019-06-27 23:10:40 +00:00
parent 1db2626a9b
commit e368095437
Notes: svn2git 2020-12-20 02:59:44 +00:00 
svn path=/head/; revision=349476
3 changed files with 69 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

71
sys/kern/kern_rangelock.c
View File

@ -141,15 +141,33 @@ rangelock_calc_block(struct rangelock *lock)
static void
rangelock_unlock_locked(struct rangelock *lock, struct rl_q_entry *entry,
struct mtx *ilk)
struct mtx *ilk, bool do_calc_block)
{
MPASS(lock != NULL && entry != NULL && ilk != NULL);
mtx_assert(ilk, MA_OWNED);
KASSERT(entry != lock->rl_currdep, ("stuck currdep"));
if (!do_calc_block) {
/*
* This is the case where rangelock_enqueue() has been called
* with trylock == true and just inserted this entry in the
* queue.
* If rl_currdep is this entry, rl_currdep needs to
* be set to the next entry in the rl_waiters list.
* However, since this entry is the last entry in the
* list, the next entry is NULL.
*/
if (lock->rl_currdep == entry) {
KASSERT(TAILQ_NEXT(lock->rl_currdep, rl_q_link) == NULL,
("rangelock_enqueue: next entry not NULL"));
lock->rl_currdep = NULL;
}
} else
KASSERT(entry != lock->rl_currdep, ("stuck currdep"));
TAILQ_REMOVE(&lock->rl_waiters, entry, rl_q_link);
rangelock_calc_block(lock);
if (do_calc_block)
rangelock_calc_block(lock);
mtx_unlock(ilk);
if (curthread->td_rlqe == NULL)
curthread->td_rlqe = entry;
@ -164,7 +182,7 @@ rangelock_unlock(struct rangelock *lock, void *cookie, struct mtx *ilk)
MPASS(lock != NULL && cookie != NULL && ilk != NULL);
mtx_lock(ilk);
rangelock_unlock_locked(lock, cookie, ilk);
rangelock_unlock_locked(lock, cookie, ilk, true);
}
/*
@ -185,7 +203,7 @@ rangelock_unlock_range(struct rangelock *lock, void *cookie, off_t start,
mtx_lock(ilk);
if (entry->rl_q_end == end) {
rangelock_unlock_locked(lock, cookie, ilk);
rangelock_unlock_locked(lock, cookie, ilk, true);
return (NULL);
}
entry->rl_q_end = end;
@ -196,11 +214,11 @@ rangelock_unlock_range(struct rangelock *lock, void *cookie, off_t start,
/*
* Add the lock request to the queue of the pending requests for
* rangelock. Sleep until the request can be granted.
* rangelock. Sleep until the request can be granted unless trylock == true.
*/
static void *
rangelock_enqueue(struct rangelock *lock, off_t start, off_t end, int mode,
struct mtx *ilk)
struct mtx *ilk, bool trylock)
{
struct rl_q_entry *entry;
struct thread *td;
@ -226,11 +244,28 @@ rangelock_enqueue(struct rangelock *lock, off_t start, off_t end, int mode,
*/
TAILQ_INSERT_TAIL(&lock->rl_waiters, entry, rl_q_link);
/*
* If rl_currdep == NULL, there is no entry waiting for a conflicting
* range to be resolved, so set rl_currdep to this entry. If there is
* no conflicting entry for this entry, rl_currdep will be set back to
* NULL by rangelock_calc_block().
*/
if (lock->rl_currdep == NULL)
lock->rl_currdep = entry;
rangelock_calc_block(lock);
while (!(entry->rl_q_flags & RL_LOCK_GRANTED))
while (!(entry->rl_q_flags & RL_LOCK_GRANTED)) {
if (trylock) {
/*
* For this case, the range is not actually locked
* yet, but removal from the list requires the same
* steps, except for not doing a rangelock_calc_block()
* call, since rangelock_calc_block() was called above.
*/
rangelock_unlock_locked(lock, entry, ilk, false);
return (NULL);
}
msleep(entry, ilk, 0, "range", 0);
}
mtx_unlock(ilk);
return (entry);
}
@ -239,12 +274,28 @@ void *
rangelock_rlock(struct rangelock *lock, off_t start, off_t end, struct mtx *ilk)
{
return (rangelock_enqueue(lock, start, end, RL_LOCK_READ, ilk));
return (rangelock_enqueue(lock, start, end, RL_LOCK_READ, ilk, false));
}
void *
rangelock_tryrlock(struct rangelock *lock, off_t start, off_t end,
struct mtx *ilk)
{
return (rangelock_enqueue(lock, start, end, RL_LOCK_READ, ilk, true));
}
void *
rangelock_wlock(struct rangelock *lock, off_t start, off_t end, struct mtx *ilk)
{
return (rangelock_enqueue(lock, start, end, RL_LOCK_WRITE, ilk));
return (rangelock_enqueue(lock, start, end, RL_LOCK_WRITE, ilk, false));
}
void *
rangelock_trywlock(struct rangelock *lock, off_t start, off_t end,
struct mtx *ilk)
{
return (rangelock_enqueue(lock, start, end, RL_LOCK_WRITE, ilk, true));
}

4
sys/sys/rangelock.h
View File

@ -75,8 +75,12 @@ void *rangelock_unlock_range(struct rangelock *lock, void *cookie,
off_t start, off_t end, struct mtx *ilk);
void *rangelock_rlock(struct rangelock *lock, off_t start, off_t end,
struct mtx *ilk);
void *rangelock_tryrlock(struct rangelock *lock, off_t start, off_t end,
struct mtx *ilk);
void *rangelock_wlock(struct rangelock *lock, off_t start, off_t end,
struct mtx *ilk);
void *rangelock_trywlock(struct rangelock *lock, off_t start, off_t end,
struct mtx *ilk);
void rlqentry_free(struct rl_q_entry *rlqe);
#endif /* _KERNEL */

4
sys/sys/vnode.h
View File

@ -720,8 +720,12 @@ int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
VI_MTX(vp))
#define vn_rangelock_rlock(vp, start, end) \
rangelock_rlock(&(vp)->v_rl, (start), (end), VI_MTX(vp))
#define vn_rangelock_tryrlock(vp, start, end) \
rangelock_tryrlock(&(vp)->v_rl, (start), (end), VI_MTX(vp))
#define vn_rangelock_wlock(vp, start, end) \
rangelock_wlock(&(vp)->v_rl, (start), (end), VI_MTX(vp))
#define vn_rangelock_trywlock(vp, start, end) \
rangelock_trywlock(&(vp)->v_rl, (start), (end), VI_MTX(vp))
int vfs_cache_lookup(struct vop_lookup_args *ap);
void vfs_timestamp(struct timespec *);