Revision 6e2df0581f569038719cf2bc2b3baa3fcc83cab4 authored by Peter Zijlstra on 08 November 2019, 10:11:52 UTC, committed by Peter Zijlstra on 08 November 2019, 21:34:14 UTC
Commit 67692435c411 ("sched: Rework pick_next_task() slow-path") inadvertly introduced a race because it changed a previously unexplored dependency between dropping the rq->lock and sched_class::put_prev_task(). The comments about dropping rq->lock, in for example newidle_balance(), only mentions the task being current and ->on_cpu being set. But when we look at the 'change' pattern (in for example sched_setnuma()): queued = task_on_rq_queued(p); /* p->on_rq == TASK_ON_RQ_QUEUED */ running = task_current(rq, p); /* rq->curr == p */ if (queued) dequeue_task(...); if (running) put_prev_task(...); /* change task properties */ if (queued) enqueue_task(...); if (running) set_next_task(...); It becomes obvious that if we do this after put_prev_task() has already been called on @p, things go sideways. This is exactly what the commit in question allows to happen when it does: prev->sched_class->put_prev_task(rq, prev, rf); if (!rq->nr_running) newidle_balance(rq, rf); The newidle_balance() call will drop rq->lock after we've called put_prev_task() and that allows the above 'change' pattern to interleave and mess up the state. Furthermore, it turns out we lost the RT-pull when we put the last DL task. Fix both problems by extracting the balancing from put_prev_task() and doing a multi-class balance() pass before put_prev_task(). Fixes: 67692435c411 ("sched: Rework pick_next_task() slow-path") Reported-by: Quentin Perret <qperret@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Tested-by: Quentin Perret <qperret@google.com> Tested-by: Valentin Schneider <valentin.schneider@arm.com>
1 parent e3b8b6a
cache.c
// SPDX-License-Identifier: GPL-2.0-only
/*
* Ceph cache definitions.
*
* Copyright (C) 2013 by Adfin Solutions, Inc. All Rights Reserved.
* Written by Milosz Tanski (milosz@adfin.com)
*/
#include <linux/ceph/ceph_debug.h>
#include "super.h"
#include "cache.h"
struct ceph_aux_inode {
u64 version;
u64 mtime_sec;
u64 mtime_nsec;
};
struct fscache_netfs ceph_cache_netfs = {
.name = "ceph",
.version = 0,
};
static DEFINE_MUTEX(ceph_fscache_lock);
static LIST_HEAD(ceph_fscache_list);
struct ceph_fscache_entry {
struct list_head list;
struct fscache_cookie *fscache;
size_t uniq_len;
/* The following members must be last */
struct ceph_fsid fsid;
char uniquifier[0];
};
static const struct fscache_cookie_def ceph_fscache_fsid_object_def = {
.name = "CEPH.fsid",
.type = FSCACHE_COOKIE_TYPE_INDEX,
};
int __init ceph_fscache_register(void)
{
return fscache_register_netfs(&ceph_cache_netfs);
}
void ceph_fscache_unregister(void)
{
fscache_unregister_netfs(&ceph_cache_netfs);
}
int ceph_fscache_register_fs(struct ceph_fs_client* fsc)
{
const struct ceph_fsid *fsid = &fsc->client->fsid;
const char *fscache_uniq = fsc->mount_options->fscache_uniq;
size_t uniq_len = fscache_uniq ? strlen(fscache_uniq) : 0;
struct ceph_fscache_entry *ent;
int err = 0;
mutex_lock(&ceph_fscache_lock);
list_for_each_entry(ent, &ceph_fscache_list, list) {
if (memcmp(&ent->fsid, fsid, sizeof(*fsid)))
continue;
if (ent->uniq_len != uniq_len)
continue;
if (uniq_len && memcmp(ent->uniquifier, fscache_uniq, uniq_len))
continue;
pr_err("fscache cookie already registered for fsid %pU\n", fsid);
pr_err(" use fsc=%%s mount option to specify a uniquifier\n");
err = -EBUSY;
goto out_unlock;
}
ent = kzalloc(sizeof(*ent) + uniq_len, GFP_KERNEL);
if (!ent) {
err = -ENOMEM;
goto out_unlock;
}
memcpy(&ent->fsid, fsid, sizeof(*fsid));
if (uniq_len > 0) {
memcpy(&ent->uniquifier, fscache_uniq, uniq_len);
ent->uniq_len = uniq_len;
}
fsc->fscache = fscache_acquire_cookie(ceph_cache_netfs.primary_index,
&ceph_fscache_fsid_object_def,
&ent->fsid, sizeof(ent->fsid) + uniq_len,
NULL, 0,
fsc, 0, true);
if (fsc->fscache) {
ent->fscache = fsc->fscache;
list_add_tail(&ent->list, &ceph_fscache_list);
} else {
kfree(ent);
pr_err("unable to register fscache cookie for fsid %pU\n",
fsid);
/* all other fs ignore this error */
}
out_unlock:
mutex_unlock(&ceph_fscache_lock);
return err;
}
static enum fscache_checkaux ceph_fscache_inode_check_aux(
void *cookie_netfs_data, const void *data, uint16_t dlen,
loff_t object_size)
{
struct ceph_aux_inode aux;
struct ceph_inode_info* ci = cookie_netfs_data;
struct inode* inode = &ci->vfs_inode;
if (dlen != sizeof(aux) ||
i_size_read(inode) != object_size)
return FSCACHE_CHECKAUX_OBSOLETE;
memset(&aux, 0, sizeof(aux));
aux.version = ci->i_version;
aux.mtime_sec = inode->i_mtime.tv_sec;
aux.mtime_nsec = inode->i_mtime.tv_nsec;
if (memcmp(data, &aux, sizeof(aux)) != 0)
return FSCACHE_CHECKAUX_OBSOLETE;
dout("ceph inode 0x%p cached okay\n", ci);
return FSCACHE_CHECKAUX_OKAY;
}
static const struct fscache_cookie_def ceph_fscache_inode_object_def = {
.name = "CEPH.inode",
.type = FSCACHE_COOKIE_TYPE_DATAFILE,
.check_aux = ceph_fscache_inode_check_aux,
};
void ceph_fscache_register_inode_cookie(struct inode *inode)
{
struct ceph_inode_info *ci = ceph_inode(inode);
struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
struct ceph_aux_inode aux;
/* No caching for filesystem */
if (!fsc->fscache)
return;
/* Only cache for regular files that are read only */
if (!S_ISREG(inode->i_mode))
return;
inode_lock_nested(inode, I_MUTEX_CHILD);
if (!ci->fscache) {
memset(&aux, 0, sizeof(aux));
aux.version = ci->i_version;
aux.mtime_sec = inode->i_mtime.tv_sec;
aux.mtime_nsec = inode->i_mtime.tv_nsec;
ci->fscache = fscache_acquire_cookie(fsc->fscache,
&ceph_fscache_inode_object_def,
&ci->i_vino, sizeof(ci->i_vino),
&aux, sizeof(aux),
ci, i_size_read(inode), false);
}
inode_unlock(inode);
}
void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci)
{
struct fscache_cookie* cookie;
if ((cookie = ci->fscache) == NULL)
return;
ci->fscache = NULL;
fscache_uncache_all_inode_pages(cookie, &ci->vfs_inode);
fscache_relinquish_cookie(cookie, &ci->i_vino, false);
}
static bool ceph_fscache_can_enable(void *data)
{
struct inode *inode = data;
return !inode_is_open_for_write(inode);
}
void ceph_fscache_file_set_cookie(struct inode *inode, struct file *filp)
{
struct ceph_inode_info *ci = ceph_inode(inode);
if (!fscache_cookie_valid(ci->fscache))
return;
if (inode_is_open_for_write(inode)) {
dout("fscache_file_set_cookie %p %p disabling cache\n",
inode, filp);
fscache_disable_cookie(ci->fscache, &ci->i_vino, false);
fscache_uncache_all_inode_pages(ci->fscache, inode);
} else {
fscache_enable_cookie(ci->fscache, &ci->i_vino, i_size_read(inode),
ceph_fscache_can_enable, inode);
if (fscache_cookie_enabled(ci->fscache)) {
dout("fscache_file_set_cookie %p %p enabling cache\n",
inode, filp);
}
}
}
static void ceph_readpage_from_fscache_complete(struct page *page, void *data, int error)
{
if (!error)
SetPageUptodate(page);
unlock_page(page);
}
static inline bool cache_valid(struct ceph_inode_info *ci)
{
return ci->i_fscache_gen == ci->i_rdcache_gen;
}
/* Atempt to read from the fscache,
*
* This function is called from the readpage_nounlock context. DO NOT attempt to
* unlock the page here (or in the callback).
*/
int ceph_readpage_from_fscache(struct inode *inode, struct page *page)
{
struct ceph_inode_info *ci = ceph_inode(inode);
int ret;
if (!cache_valid(ci))
return -ENOBUFS;
ret = fscache_read_or_alloc_page(ci->fscache, page,
ceph_readpage_from_fscache_complete, NULL,
GFP_KERNEL);
switch (ret) {
case 0: /* Page found */
dout("page read submitted\n");
return 0;
case -ENOBUFS: /* Pages were not found, and can't be */
case -ENODATA: /* Pages were not found */
dout("page/inode not in cache\n");
return ret;
default:
dout("%s: unknown error ret = %i\n", __func__, ret);
return ret;
}
}
int ceph_readpages_from_fscache(struct inode *inode,
struct address_space *mapping,
struct list_head *pages,
unsigned *nr_pages)
{
struct ceph_inode_info *ci = ceph_inode(inode);
int ret;
if (!cache_valid(ci))
return -ENOBUFS;
ret = fscache_read_or_alloc_pages(ci->fscache, mapping, pages, nr_pages,
ceph_readpage_from_fscache_complete,
NULL, mapping_gfp_mask(mapping));
switch (ret) {
case 0: /* All pages found */
dout("all-page read submitted\n");
return 0;
case -ENOBUFS: /* Some pages were not found, and can't be */
case -ENODATA: /* some pages were not found */
dout("page/inode not in cache\n");
return ret;
default:
dout("%s: unknown error ret = %i\n", __func__, ret);
return ret;
}
}
void ceph_readpage_to_fscache(struct inode *inode, struct page *page)
{
struct ceph_inode_info *ci = ceph_inode(inode);
int ret;
if (!PageFsCache(page))
return;
if (!cache_valid(ci))
return;
ret = fscache_write_page(ci->fscache, page, i_size_read(inode),
GFP_KERNEL);
if (ret)
fscache_uncache_page(ci->fscache, page);
}
void ceph_invalidate_fscache_page(struct inode* inode, struct page *page)
{
struct ceph_inode_info *ci = ceph_inode(inode);
if (!PageFsCache(page))
return;
fscache_wait_on_page_write(ci->fscache, page);
fscache_uncache_page(ci->fscache, page);
}
void ceph_fscache_unregister_fs(struct ceph_fs_client* fsc)
{
if (fscache_cookie_valid(fsc->fscache)) {
struct ceph_fscache_entry *ent;
bool found = false;
mutex_lock(&ceph_fscache_lock);
list_for_each_entry(ent, &ceph_fscache_list, list) {
if (ent->fscache == fsc->fscache) {
list_del(&ent->list);
kfree(ent);
found = true;
break;
}
}
WARN_ON_ONCE(!found);
mutex_unlock(&ceph_fscache_lock);
__fscache_relinquish_cookie(fsc->fscache, NULL, false);
}
fsc->fscache = NULL;
}
/*
* caller should hold CEPH_CAP_FILE_{RD,CACHE}
*/
void ceph_fscache_revalidate_cookie(struct ceph_inode_info *ci)
{
if (cache_valid(ci))
return;
/* resue i_truncate_mutex. There should be no pending
* truncate while the caller holds CEPH_CAP_FILE_RD */
mutex_lock(&ci->i_truncate_mutex);
if (!cache_valid(ci)) {
if (fscache_check_consistency(ci->fscache, &ci->i_vino))
fscache_invalidate(ci->fscache);
spin_lock(&ci->i_ceph_lock);
ci->i_fscache_gen = ci->i_rdcache_gen;
spin_unlock(&ci->i_ceph_lock);
}
mutex_unlock(&ci->i_truncate_mutex);
}
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...