From 11d763f0b0afc2cf5f92f4adae5dbbbbef712f8f Mon Sep 17 00:00:00 2001 From: Andreas Gruenbacher Date: Tue, 9 Dec 2025 22:59:12 +0000 Subject: [PATCH] gfs2: Retries missing in gfs2_{rename,exchange} Fix a bug in gfs2's asynchronous glock handling for rename and exchange operations. The original async implementation from commit ad26967b9afa ("gfs2: Use async glocks for rename") mentioned that retries were needed but never implemented them, causing operations to fail with -ESTALE instead of retrying on timeout. Also makes the waiting interruptible. In addition, the timeouts used were too high for situations in which timing out is a rare but expected scenario. Switch to shorter timeouts with randomization and exponentional backoff. Fixes: ad26967b9afa ("gfs2: Use async glocks for rename") Signed-off-by: Andreas Gruenbacher --- fs/gfs2/glock.c | 36 +++++++++++++++++++++++++++--------- fs/gfs2/glock.h | 3 ++- fs/gfs2/inode.c | 18 ++++++++++++++---- 3 files changed, 43 insertions(+), 14 deletions(-) diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c index 47e8809575d50..f358448d7b3d7 100644 --- a/fs/gfs2/glock.c +++ b/fs/gfs2/glock.c @@ -1284,31 +1284,45 @@ static int glocks_pending(unsigned int num_gh, struct gfs2_holder *ghs) * gfs2_glock_async_wait - wait on multiple asynchronous glock acquisitions * @num_gh: the number of holders in the array * @ghs: the glock holder array + * @retries: number of retries attempted so far * * Returns: 0 on success, meaning all glocks have been granted and are held. * -ESTALE if the request timed out, meaning all glocks were released, * and the caller should retry the operation. */ -int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs) +int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs, + unsigned int retries) { struct gfs2_sbd *sdp = ghs[0].gh_gl->gl_name.ln_sbd; - int i, ret = 0, timeout = 0; unsigned long start_time = jiffies; + int i, ret = 0; + long timeout; might_sleep(); - /* - * Total up the (minimum hold time * 2) of all glocks and use that to - * determine the max amount of time we should wait. - */ - for (i = 0; i < num_gh; i++) - timeout += ghs[i].gh_gl->gl_hold_time << 1; - if (!wait_event_timeout(sdp->sd_async_glock_wait, + timeout = GL_GLOCK_MIN_HOLD; + if (retries) { + unsigned int max_shift; + long incr; + + /* Add a random delay and increase the timeout exponentially. */ + max_shift = BITS_PER_LONG - 2 - __fls(GL_GLOCK_HOLD_INCR); + incr = min(GL_GLOCK_HOLD_INCR << min(retries - 1, max_shift), + 10 * HZ - GL_GLOCK_MIN_HOLD); + schedule_timeout_interruptible(get_random_long() % (incr / 3)); + if (signal_pending(current)) + goto interrupted; + timeout += (incr / 3) + get_random_long() % (incr / 3); + } + + if (!wait_event_interruptible_timeout(sdp->sd_async_glock_wait, !glocks_pending(num_gh, ghs), timeout)) { ret = -ESTALE; /* request timed out. */ goto out; } + if (signal_pending(current)) + goto interrupted; for (i = 0; i < num_gh; i++) { struct gfs2_holder *gh = &ghs[i]; @@ -1332,6 +1346,10 @@ out: } } return ret; + +interrupted: + ret = -EINTR; + goto out; } /** diff --git a/fs/gfs2/glock.h b/fs/gfs2/glock.h index 55d5985f32a08..dccbf36b8cb10 100644 --- a/fs/gfs2/glock.h +++ b/fs/gfs2/glock.h @@ -204,7 +204,8 @@ int gfs2_glock_poll(struct gfs2_holder *gh); int gfs2_instantiate(struct gfs2_holder *gh); int gfs2_glock_holder_ready(struct gfs2_holder *gh); int gfs2_glock_wait(struct gfs2_holder *gh); -int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs); +int gfs2_glock_async_wait(unsigned int num_gh, struct gfs2_holder *ghs, + unsigned int retries); void gfs2_glock_dq(struct gfs2_holder *gh); void gfs2_glock_dq_wait(struct gfs2_holder *gh); void gfs2_glock_dq_uninit(struct gfs2_holder *gh); diff --git a/fs/gfs2/inode.c b/fs/gfs2/inode.c index 36618e3531996..b6ed069b34872 100644 --- a/fs/gfs2/inode.c +++ b/fs/gfs2/inode.c @@ -1495,7 +1495,7 @@ static int gfs2_rename(struct inode *odir, struct dentry *odentry, unsigned int num_gh; int dir_rename = 0; struct gfs2_diradd da = { .nr_blocks = 0, .save_loc = 0, }; - unsigned int x; + unsigned int retries = 0, x; int error; gfs2_holder_mark_uninitialized(&r_gh); @@ -1545,12 +1545,17 @@ static int gfs2_rename(struct inode *odir, struct dentry *odentry, num_gh++; } +again: for (x = 0; x < num_gh; x++) { error = gfs2_glock_nq(ghs + x); if (error) goto out_gunlock; } - error = gfs2_glock_async_wait(num_gh, ghs); + error = gfs2_glock_async_wait(num_gh, ghs, retries); + if (error == -ESTALE) { + retries++; + goto again; + } if (error) goto out_gunlock; @@ -1739,7 +1744,7 @@ static int gfs2_exchange(struct inode *odir, struct dentry *odentry, struct gfs2_sbd *sdp = GFS2_SB(odir); struct gfs2_holder ghs[4], r_gh; unsigned int num_gh; - unsigned int x; + unsigned int retries = 0, x; umode_t old_mode = oip->i_inode.i_mode; umode_t new_mode = nip->i_inode.i_mode; int error; @@ -1783,13 +1788,18 @@ static int gfs2_exchange(struct inode *odir, struct dentry *odentry, gfs2_holder_init(nip->i_gl, LM_ST_EXCLUSIVE, GL_ASYNC, ghs + num_gh); num_gh++; +again: for (x = 0; x < num_gh; x++) { error = gfs2_glock_nq(ghs + x); if (error) goto out_gunlock; } - error = gfs2_glock_async_wait(num_gh, ghs); + error = gfs2_glock_async_wait(num_gh, ghs, retries); + if (error == -ESTALE) { + retries++; + goto again; + } if (error) goto out_gunlock; -- 2.47.3