Browse Source

io_uring-5.12-2021-03-05

-----BEGIN PGP SIGNATURE-----
 
 iQJEBAABCAAuFiEEwPw5LcreJtl1+l5K99NY+ylx4KYFAmBCYeIQHGF4Ym9lQGtl
 cm5lbC5kawAKCRD301j7KXHgpisOD/9bSFR7gRqO9oIy6/PEveRI4PWDujjcXgRZ
 6jxQnfFUrNQsXcXIlHO4HUDG7DVX/isxdk/YVGhVfuKoco/a0XyYAALH5SVy77T+
 hDdWCIJBXgxnfAvv+xMBQDEwlz+pdaOLfOVaGMRAp3akuVTBMA+ZE940Lc81kBaU
 bTGev+BzPUsUE7n6ebPdhIQDA6LB02e7kaBZsRDwjsABJuD3o4O1jOAtZyqpPRsW
 nADvxsrlMxB3RN97iokinBXV426iAQ/nBDYVDVnWpbckD7Ti4f6r2ohku0qEdhZS
 XrTF+1mzEqdmvMLl1YQ/GGpH7ReOLHN78aj4BaG49+pryfkaFe50AHr7frGqKLms
 DWymTJnpdJSTNT0Z2GRLNrnWHa3YgeuPMdhlIPfihnZBXhZ7p6X5iNpQ69jd93P3
 zLXMJ0RKpkl6bmV+Pk4kCqUfz1BV3sUqG9euLdTq+3uBRA0/B5ktPosyH2DGqUYa
 n9aEUHslwHUF+Deu/S9RmVzhTjuD0IRbURSeayimFFe71kHhKsHShOKQMUkhu6zQ
 AMsQRq9VrWy/3x3C+qpcbEJ3BIqyGLbiQByOBx96kg9Zk14io3GEmSlqZcxbsKTq
 /JXjanaEcUwtKKccOC6g+O+G7VlskO9gLi/Fj/x98R92UBEqpEtVZb8MLCdpiLY/
 SHJHbC7Fpw==
 =w0Sf
 -----END PGP SIGNATURE-----

Merge tag 'io_uring-5.12-2021-03-05' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:
 "A bit of a mix between fallout from the worker change, cleanups and
  reductions now possible from that change, and fixes in general. In
  detail:

   - Fully serialize manager and worker creation, fixing races due to
     that.

   - Clean up some naming that had gone stale.

   - SQPOLL fixes.

   - Fix race condition around task_work rework that went into this
     merge window.

   - Implement unshare. Used for when the original task does unshare(2)
     or setuid/seteuid and friends, drops the original workers and forks
     new ones.

   - Drop the only remaining piece of state shuffling we had left, which
     was cred. Move it into issue instead, and we can drop all of that
     code too.

   - Kill f_op->flush() usage. That was such a nasty hack that we had
     out of necessity, we no longer need it.

   - Following from ->flush() removal, we can also drop various bits of
     ctx state related to SQPOLL and cancelations.

   - Fix an issue with IOPOLL retry, which originally was fallout from a
     filemap change (removing iov_iter_revert()), but uncovered an issue
     with iovec re-import too late.

   - Fix an issue with system suspend.

   - Use xchg() for fallback work, instead of cmpxchg().

   - Properly destroy io-wq on exec.

   - Add create_io_thread() core helper, and use that in io-wq and
     io_uring. This allows us to remove various silly completion events
     related to thread setup.

   - A few error handling fixes.

  This should be the grunt of fixes necessary for the new workers, next
  week should be quieter. We've got a pending series from Pavel on
  cancelations, and how tasks and rings are indexed. Outside of that,
  should just be minor fixes. Even with these fixes, we're still killing
  a net ~80 lines"

* tag 'io_uring-5.12-2021-03-05' of git://git.kernel.dk/linux-block: (41 commits)
  io_uring: don't restrict issue_flags for io_openat
  io_uring: make SQPOLL thread parking saner
  io-wq: kill hashed waitqueue before manager exits
  io_uring: clear IOCB_WAITQ for non -EIOCBQUEUED return
  io_uring: don't keep looping for more events if we can't flush overflow
  io_uring: move to using create_io_thread()
  kernel: provide create_io_thread() helper
  io_uring: reliably cancel linked timeouts
  io_uring: cancel-match based on flags
  io-wq: ensure all pending work is canceled on exit
  io_uring: ensure that threads freeze on suspend
  io_uring: remove extra in_idle wake up
  io_uring: inline __io_queue_async_work()
  io_uring: inline io_req_clean_work()
  io_uring: choose right tctx->io_wq for try cancel
  io_uring: fix -EAGAIN retry with IOPOLL
  io-wq: fix error path leak of buffered write hash map
  io_uring: remove sqo_task
  io_uring: kill sqo_dead and sqo submission halting
  io_uring: ignore double poll add on the same waitqueue head
  ...
master
Linus Torvalds 2 months ago
parent
commit
f292e8730a
6 changed files with 361 additions and 439 deletions
  1. +121
    -140
      fs/io-wq.c
  2. +2
    -3
      fs/io-wq.h
  3. +205
    -295
      fs/io_uring.c
  4. +1
    -1
      include/linux/io_uring.h
  5. +2
    -0
      include/linux/sched/task.h
  6. +30
    -0
      kernel/fork.c

+ 121
- 140
fs/io-wq.c View File

@ -16,6 +16,7 @@
#include <linux/rculist_nulls.h>
#include <linux/cpu.h>
#include <linux/tracehook.h>
#include <linux/freezer.h>
#include "../kernel/sched/sched.h"
#include "io-wq.h"
@ -52,9 +53,6 @@ struct io_worker {
struct io_wq_work *cur_work;
spinlock_t lock;
const struct cred *cur_creds;
const struct cred *saved_creds;
struct completion ref_done;
struct rcu_head rcu;
@ -117,7 +115,10 @@ struct io_wq {
struct io_wq_hash *hash;
refcount_t refs;
struct completion done;
struct completion exited;
atomic_t worker_refs;
struct completion worker_done;
struct hlist_node cpuhp_node;
@ -126,6 +127,17 @@ struct io_wq {
static enum cpuhp_state io_wq_online;
struct io_cb_cancel_data {
work_cancel_fn *fn;
void *data;
int nr_running;
int nr_pending;
bool cancel_all;
};
static void io_wqe_cancel_pending_work(struct io_wqe *wqe,
struct io_cb_cancel_data *match);
static bool io_worker_get(struct io_worker *worker)
{
return refcount_inc_not_zero(&worker->ref);
@ -175,11 +187,6 @@ static void io_worker_exit(struct io_worker *worker)
worker->flags = 0;
preempt_enable();
if (worker->saved_creds) {
revert_creds(worker->saved_creds);
worker->cur_creds = worker->saved_creds = NULL;
}
raw_spin_lock_irq(&wqe->lock);
if (flags & IO_WORKER_F_FREE)
hlist_nulls_del_rcu(&worker->nulls_node);
@ -188,7 +195,9 @@ static void io_worker_exit(struct io_worker *worker)
raw_spin_unlock_irq(&wqe->lock);
kfree_rcu(worker, rcu);
io_wq_put(wqe->wq);
if (atomic_dec_and_test(&wqe->wq->worker_refs))
complete(&wqe->wq->worker_done);
do_exit(0);
}
static inline bool io_wqe_run_queue(struct io_wqe *wqe)
@ -263,12 +272,6 @@ static void io_wqe_dec_running(struct io_worker *worker)
io_wqe_wake_worker(wqe, acct);
}
static void io_worker_start(struct io_worker *worker)
{
worker->flags |= (IO_WORKER_F_UP | IO_WORKER_F_RUNNING);
io_wqe_inc_running(worker);
}
/*
* Worker will start processing some work. Move it to the busy list, if
* it's currently on the freelist
@ -319,10 +322,6 @@ static void __io_worker_idle(struct io_wqe *wqe, struct io_worker *worker)
worker->flags |= IO_WORKER_F_FREE;
hlist_nulls_add_head_rcu(&worker->nulls_node, &wqe->free_list);
}
if (worker->saved_creds) {
revert_creds(worker->saved_creds);
worker->cur_creds = worker->saved_creds = NULL;
}
}
static inline unsigned int io_get_work_hash(struct io_wq_work *work)
@ -397,18 +396,6 @@ static void io_flush_signals(void)
}
}
static void io_wq_switch_creds(struct io_worker *worker,
struct io_wq_work *work)
{
const struct cred *old_creds = override_creds(work->creds);
worker->cur_creds = work->creds;
if (worker->saved_creds)
put_cred(old_creds); /* creds set by previous switch */
else
worker->saved_creds = old_creds;
}
static void io_assign_current_work(struct io_worker *worker,
struct io_wq_work *work)
{
@ -458,8 +445,6 @@ get_next:
unsigned int hash = io_get_work_hash(work);
next_hashed = wq_next_work(work);
if (work->creds && worker->cur_creds != work->creds)
io_wq_switch_creds(worker, work);
wq->do_work(work);
io_assign_current_work(worker, NULL);
@ -495,8 +480,13 @@ static int io_wqe_worker(void *data)
struct io_worker *worker = data;
struct io_wqe *wqe = worker->wqe;
struct io_wq *wq = wqe->wq;
char buf[TASK_COMM_LEN];
worker->flags |= (IO_WORKER_F_UP | IO_WORKER_F_RUNNING);
io_wqe_inc_running(worker);
io_worker_start(worker);
sprintf(buf, "iou-wrk-%d", wq->task_pid);
set_task_comm(current, buf);
while (!test_bit(IO_WQ_BIT_EXIT, &wq->state)) {
set_current_state(TASK_INTERRUPTIBLE);
@ -571,67 +561,11 @@ void io_wq_worker_sleeping(struct task_struct *tsk)
raw_spin_unlock_irq(&worker->wqe->lock);
}
static int task_thread(void *data, int index)
{
struct io_worker *worker = data;
struct io_wqe *wqe = worker->wqe;
struct io_wqe_acct *acct = &wqe->acct[index];
struct io_wq *wq = wqe->wq;
char buf[TASK_COMM_LEN];
sprintf(buf, "iou-wrk-%d", wq->task_pid);
set_task_comm(current, buf);
current->pf_io_worker = worker;
worker->task = current;
set_cpus_allowed_ptr(current, cpumask_of_node(wqe->node));
current->flags |= PF_NO_SETAFFINITY;
raw_spin_lock_irq(&wqe->lock);
hlist_nulls_add_head_rcu(&worker->nulls_node, &wqe->free_list);
list_add_tail_rcu(&worker->all_list, &wqe->all_list);
worker->flags |= IO_WORKER_F_FREE;
if (index == IO_WQ_ACCT_BOUND)
worker->flags |= IO_WORKER_F_BOUND;
if (!acct->nr_workers && (worker->flags & IO_WORKER_F_BOUND))
worker->flags |= IO_WORKER_F_FIXED;
acct->nr_workers++;
raw_spin_unlock_irq(&wqe->lock);
io_wqe_worker(data);
do_exit(0);
}
static int task_thread_bound(void *data)
{
return task_thread(data, IO_WQ_ACCT_BOUND);
}
static int task_thread_unbound(void *data)
{
return task_thread(data, IO_WQ_ACCT_UNBOUND);
}
pid_t io_wq_fork_thread(int (*fn)(void *), void *arg)
{
unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
CLONE_IO|SIGCHLD;
struct kernel_clone_args args = {
.flags = ((lower_32_bits(flags) | CLONE_VM |
CLONE_UNTRACED) & ~CSIGNAL),
.exit_signal = (lower_32_bits(flags) & CSIGNAL),
.stack = (unsigned long)fn,
.stack_size = (unsigned long)arg,
};
return kernel_clone(&args);
}
static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index)
{
struct io_wqe_acct *acct = &wqe->acct[index];
struct io_worker *worker;
pid_t pid;
struct task_struct *tsk;
__set_current_state(TASK_RUNNING);
@ -645,17 +579,32 @@ static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index)
spin_lock_init(&worker->lock);
init_completion(&worker->ref_done);
refcount_inc(&wq->refs);
atomic_inc(&wq->worker_refs);
if (index == IO_WQ_ACCT_BOUND)
pid = io_wq_fork_thread(task_thread_bound, worker);
else
pid = io_wq_fork_thread(task_thread_unbound, worker);
if (pid < 0) {
io_wq_put(wq);
tsk = create_io_thread(io_wqe_worker, worker, wqe->node);
if (IS_ERR(tsk)) {
if (atomic_dec_and_test(&wq->worker_refs))
complete(&wq->worker_done);
kfree(worker);
return false;
}
tsk->pf_io_worker = worker;
worker->task = tsk;
set_cpus_allowed_ptr(tsk, cpumask_of_node(wqe->node));
tsk->flags |= PF_NOFREEZE | PF_NO_SETAFFINITY;
raw_spin_lock_irq(&wqe->lock);
hlist_nulls_add_head_rcu(&worker->nulls_node, &wqe->free_list);
list_add_tail_rcu(&worker->all_list, &wqe->all_list);
worker->flags |= IO_WORKER_F_FREE;
if (index == IO_WQ_ACCT_BOUND)
worker->flags |= IO_WORKER_F_BOUND;
if (!acct->nr_workers && (worker->flags & IO_WORKER_F_BOUND))
worker->flags |= IO_WORKER_F_FIXED;
acct->nr_workers++;
raw_spin_unlock_irq(&wqe->lock);
wake_up_new_task(tsk);
return true;
}
@ -664,6 +613,8 @@ static inline bool io_wqe_need_worker(struct io_wqe *wqe, int index)
{
struct io_wqe_acct *acct = &wqe->acct[index];
if (acct->nr_workers && test_bit(IO_WQ_BIT_EXIT, &wqe->wq->state))
return false;
/* if we have available workers or no work, no need */
if (!hlist_nulls_empty(&wqe->free_list) || !io_wqe_run_queue(wqe))
return false;
@ -697,6 +648,7 @@ static bool io_wq_for_each_worker(struct io_wqe *wqe,
static bool io_wq_worker_wake(struct io_worker *worker, void *data)
{
set_notify_signal(worker->task);
wake_up_process(worker->task);
return false;
}
@ -725,6 +677,23 @@ static void io_wq_check_workers(struct io_wq *wq)
}
}
static bool io_wq_work_match_all(struct io_wq_work *work, void *data)
{
return true;
}
static void io_wq_cancel_pending(struct io_wq *wq)
{
struct io_cb_cancel_data match = {
.fn = io_wq_work_match_all,
.cancel_all = true,
};
int node;
for_each_node(node)
io_wqe_cancel_pending_work(wq->wqes[node], &match);
}
/*
* Manager thread. Tasked with creating new workers, if we need them.
*/
@ -732,25 +701,38 @@ static int io_wq_manager(void *data)
{
struct io_wq *wq = data;
char buf[TASK_COMM_LEN];
int node;
sprintf(buf, "iou-mgr-%d", wq->task_pid);
set_task_comm(current, buf);
current->flags |= PF_IO_WORKER;
wq->manager = current;
complete(&wq->done);
do {
set_current_state(TASK_INTERRUPTIBLE);
io_wq_check_workers(wq);
schedule_timeout(HZ);
try_to_freeze();
if (fatal_signal_pending(current))
set_bit(IO_WQ_BIT_EXIT, &wq->state);
} while (!test_bit(IO_WQ_BIT_EXIT, &wq->state));
io_wq_check_workers(wq);
wq->manager = NULL;
io_wq_put(wq);
rcu_read_lock();
for_each_node(node)
io_wq_for_each_worker(wq->wqes[node], io_wq_worker_wake, NULL);
rcu_read_unlock();
/* we might not ever have created any workers */
if (atomic_read(&wq->worker_refs))
wait_for_completion(&wq->worker_done);
spin_lock_irq(&wq->hash->wait.lock);
for_each_node(node)
list_del_init(&wq->wqes[node]->wait.entry);
spin_unlock_irq(&wq->hash->wait.lock);
io_wq_cancel_pending(wq);
complete(&wq->exited);
do_exit(0);
}
@ -787,23 +769,20 @@ append:
static int io_wq_fork_manager(struct io_wq *wq)
{
int ret;
struct task_struct *tsk;
if (wq->manager)
return 0;
clear_bit(IO_WQ_BIT_EXIT, &wq->state);
refcount_inc(&wq->refs);
current->flags |= PF_IO_WORKER;
ret = io_wq_fork_thread(io_wq_manager, wq);
current->flags &= ~PF_IO_WORKER;
if (ret >= 0) {
wait_for_completion(&wq->done);
reinit_completion(&wq->worker_done);
tsk = create_io_thread(io_wq_manager, wq, NUMA_NO_NODE);
if (!IS_ERR(tsk)) {
wq->manager = get_task_struct(tsk);
wake_up_new_task(tsk);
return 0;
}
io_wq_put(wq);
return ret;
return PTR_ERR(tsk);
}
static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work)
@ -813,7 +792,8 @@ static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work)
unsigned long flags;
/* Can only happen if manager creation fails after exec */
if (unlikely(io_wq_fork_manager(wqe->wq))) {
if (io_wq_fork_manager(wqe->wq) ||
test_bit(IO_WQ_BIT_EXIT, &wqe->wq->state)) {
work->flags |= IO_WQ_WORK_CANCEL;
wqe->wq->do_work(work);
return;
@ -849,14 +829,6 @@ void io_wq_hash_work(struct io_wq_work *work, void *val)
work->flags |= (IO_WQ_WORK_HASHED | (bit << IO_WQ_HASH_SHIFT));
}
struct io_cb_cancel_data {
work_cancel_fn *fn;
void *data;
int nr_running;
int nr_pending;
bool cancel_all;
};
static bool io_wq_worker_cancel(struct io_worker *worker, void *data)
{
struct io_cb_cancel_data *match = data;
@ -1043,16 +1015,18 @@ struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data)
}
wq->task_pid = current->pid;
init_completion(&wq->done);
init_completion(&wq->exited);
refcount_set(&wq->refs, 1);
init_completion(&wq->worker_done);
atomic_set(&wq->worker_refs, 0);
ret = io_wq_fork_manager(wq);
if (!ret)
return wq;
io_wq_put(wq);
io_wq_put_hash(data->hash);
err:
io_wq_put_hash(data->hash);
cpuhp_state_remove_instance_nocalls(io_wq_online, &wq->cpuhp_node);
for_each_node(node)
kfree(wq->wqes[node]);
@ -1063,6 +1037,16 @@ err_wq:
return ERR_PTR(ret);
}
static void io_wq_destroy_manager(struct io_wq *wq)
{
if (wq->manager) {
wake_up_process(wq->manager);
wait_for_completion(&wq->exited);
put_task_struct(wq->manager);
wq->manager = NULL;
}
}
static void io_wq_destroy(struct io_wq *wq)
{
int node;
@ -1070,26 +1054,16 @@ static void io_wq_destroy(struct io_wq *wq)
cpuhp_state_remove_instance_nocalls(io_wq_online, &wq->cpuhp_node);
set_bit(IO_WQ_BIT_EXIT, &wq->state);
if (wq->manager)
wake_up_process(wq->manager);
rcu_read_lock();
for_each_node(node)
io_wq_for_each_worker(wq->wqes[node], io_wq_worker_wake, NULL);
rcu_read_unlock();
io_wq_destroy_manager(wq);
spin_lock_irq(&wq->hash->wait.lock);
for_each_node(node) {
struct io_wqe *wqe = wq->wqes[node];
list_del_init(&wqe->wait.entry);
WARN_ON_ONCE(!wq_list_empty(&wqe->work_list));
kfree(wqe);
}
spin_unlock_irq(&wq->hash->wait.lock);
io_wq_put_hash(wq->hash);
kfree(wq->wqes);
kfree(wq);
}
void io_wq_put(struct io_wq *wq)
@ -1098,6 +1072,13 @@ void io_wq_put(struct io_wq *wq)
io_wq_destroy(wq);
}
void io_wq_put_and_exit(struct io_wq *wq)
{
set_bit(IO_WQ_BIT_EXIT, &wq->state);
io_wq_destroy_manager(wq);
io_wq_put(wq);
}
static bool io_wq_worker_affinity(struct io_worker *worker, void *data)
{
struct task_struct *task = worker->task;


+ 2
- 3
fs/io-wq.h View File

@ -79,8 +79,8 @@ static inline void wq_list_del(struct io_wq_work_list *list,
struct io_wq_work {
struct io_wq_work_node list;
const struct cred *creds;
unsigned flags;
unsigned short personality;
};
static inline struct io_wq_work *wq_next_work(struct io_wq_work *work)
@ -114,12 +114,11 @@ struct io_wq_data {
struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data);
void io_wq_put(struct io_wq *wq);
void io_wq_put_and_exit(struct io_wq *wq);
void io_wq_enqueue(struct io_wq *wq, struct io_wq_work *work);
void io_wq_hash_work(struct io_wq_work *work, void *val);
pid_t io_wq_fork_thread(int (*fn)(void *), void *arg);
static inline bool io_wq_is_hashed(struct io_wq_work *work)
{
return work->flags & IO_WQ_WORK_HASHED;


+ 205
- 295
fs/io_uring.c
File diff suppressed because it is too large
View File


+ 1
- 1
include/linux/io_uring.h View File

@ -38,7 +38,7 @@ void __io_uring_free(struct task_struct *tsk);
static inline void io_uring_task_cancel(void)
{
if (current->io_uring && !xa_empty(&current->io_uring->xa))
if (current->io_uring)
__io_uring_task_cancel();
}
static inline void io_uring_files_cancel(struct files_struct *files)


+ 2
- 0
include/linux/sched/task.h View File

@ -31,6 +31,7 @@ struct kernel_clone_args {
/* Number of elements in *set_tid */
size_t set_tid_size;
int cgroup;
int io_thread;
struct cgroup *cgrp;
struct css_set *cset;
};
@ -82,6 +83,7 @@ extern void exit_files(struct task_struct *);
extern void exit_itimers(struct signal_struct *);
extern pid_t kernel_clone(struct kernel_clone_args *kargs);
struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node);
struct task_struct *fork_idle(int);
struct mm_struct *copy_init_mm(void);
extern pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);


+ 30
- 0
kernel/fork.c View File

@ -1940,6 +1940,8 @@ static __latent_entropy struct task_struct *copy_process(
p = dup_task_struct(current, node);
if (!p)
goto fork_out;
if (args->io_thread)
p->flags |= PF_IO_WORKER;
/*
* This _must_ happen before we call free_task(), i.e. before we jump
@ -2410,6 +2412,34 @@ struct mm_struct *copy_init_mm(void)
return dup_mm(NULL, &init_mm);
}
/*
* This is like kernel_clone(), but shaved down and tailored to just
* creating io_uring workers. It returns a created task, or an error pointer.
* The returned task is inactive, and the caller must fire it up through
* wake_up_new_task(p). All signals are blocked in the created task.
*/
struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
{
unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
CLONE_IO;
struct kernel_clone_args args = {
.flags = ((lower_32_bits(flags) | CLONE_VM |
CLONE_UNTRACED) & ~CSIGNAL),
.exit_signal = (lower_32_bits(flags) & CSIGNAL),
.stack = (unsigned long)fn,
.stack_size = (unsigned long)arg,
.io_thread = 1,
};
struct task_struct *tsk;
tsk = copy_process(NULL, 0, node, &args);
if (!IS_ERR(tsk)) {
sigfillset(&tsk->blocked);
sigdelsetmask(&tsk->blocked, sigmask(SIGKILL));
}
return tsk;
}
/*
* Ok, this is the main fork-routine.
*


Loading…
Cancel
Save