*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_CLASS
#include <linux/sched.h>
+#include <linux/list.h>
+#include <linux/list_sort.h>
#include <obd_class.h>
#include <obd_support.h>
#include <lustre_fid.h>
-#include <libcfs/list.h>
#include <cl_object.h>
#include "cl_internal.h"
*
*/
-#define cl_io_for_each(slice, io) \
- list_for_each_entry((slice), &io->ci_layers, cis_linkage)
-#define cl_io_for_each_reverse(slice, io) \
- list_for_each_entry_reverse((slice), &io->ci_layers, cis_linkage)
-
static inline int cl_io_type_is_valid(enum cl_io_type type)
{
return CIT_READ <= type && type < CIT_OP_NR;
}
/**
- * Returns true iff there is an IO ongoing in the given environment.
- */
-int cl_io_is_going(const struct lu_env *env)
-{
- return cl_env_info(env)->clt_current_io != NULL;
-}
-EXPORT_SYMBOL(cl_io_is_going);
-
-/**
* cl_io invariant that holds at all times when exported cl_io_*() functions
* are entered and left.
*/
void cl_io_fini(const struct lu_env *env, struct cl_io *io)
{
struct cl_io_slice *slice;
- struct cl_thread_info *info;
LINVRNT(cl_io_type_is_valid(io->ci_type));
LINVRNT(cl_io_invariant(io));
slice->cis_io = NULL;
}
io->ci_state = CIS_FINI;
- info = cl_env_info(env);
- if (info->clt_current_io == io)
- info->clt_current_io = NULL;
/* sanity check for layout change */
switch(io->ci_type) {
case CIT_READ:
case CIT_WRITE:
- break;
+ case CIT_DATA_VERSION:
case CIT_FAULT:
+ break;
case CIT_FSYNC:
LASSERT(!io->ci_need_restart);
break;
/* Check ignore layout change conf */
LASSERT(ergo(io->ci_ignore_layout || !io->ci_verify_layout,
!io->ci_need_restart));
+ case CIT_GLIMPSE:
+ break;
+ case CIT_LADVISE:
break;
default:
LBUG();
io->ci_type = iot;
INIT_LIST_HEAD(&io->ci_lockset.cls_todo);
- INIT_LIST_HEAD(&io->ci_lockset.cls_curr);
INIT_LIST_HEAD(&io->ci_lockset.cls_done);
INIT_LIST_HEAD(&io->ci_layers);
int cl_io_sub_init(const struct lu_env *env, struct cl_io *io,
enum cl_io_type iot, struct cl_object *obj)
{
- struct cl_thread_info *info = cl_env_info(env);
-
LASSERT(obj != cl_object_top(obj));
- if (info->clt_current_io == NULL)
- info->clt_current_io = io;
+
return cl_io_init0(env, io, iot, obj);
}
EXPORT_SYMBOL(cl_io_sub_init);
int cl_io_init(const struct lu_env *env, struct cl_io *io,
enum cl_io_type iot, struct cl_object *obj)
{
- struct cl_thread_info *info = cl_env_info(env);
+ LASSERT(obj == cl_object_top(obj));
- LASSERT(obj == cl_object_top(obj));
- LASSERT(info->clt_current_io == NULL);
+ /* clear I/O restart from previous instance */
+ io->ci_need_restart = 0;
- info->clt_current_io = io;
- return cl_io_init0(env, io, iot, obj);
+ return cl_io_init0(env, io, iot, obj);
}
EXPORT_SYMBOL(cl_io_init);
* \pre iot == CIT_READ || iot == CIT_WRITE
*/
int cl_io_rw_init(const struct lu_env *env, struct cl_io *io,
- enum cl_io_type iot, loff_t pos, size_t count)
+ enum cl_io_type iot, loff_t pos, size_t count)
{
- LINVRNT(iot == CIT_READ || iot == CIT_WRITE);
- LINVRNT(io->ci_obj != NULL);
- ENTRY;
+ LINVRNT(iot == CIT_READ || iot == CIT_WRITE);
+ LINVRNT(io->ci_obj != NULL);
+ ENTRY;
- LU_OBJECT_HEADER(D_VFSTRACE, env, &io->ci_obj->co_lu,
- "io range: %u ["LPU64", "LPU64") %u %u\n",
- iot, (__u64)pos, (__u64)pos + count,
- io->u.ci_rw.crw_nonblock, io->u.ci_wr.wr_append);
- io->u.ci_rw.crw_pos = pos;
- io->u.ci_rw.crw_count = count;
- RETURN(cl_io_init(env, io, iot, io->ci_obj));
+ LU_OBJECT_HEADER(D_VFSTRACE, env, &io->ci_obj->co_lu,
+ "io range: %u [%llu, %llu) %u %u\n",
+ iot, (__u64)pos, (__u64)pos + count,
+ io->u.ci_rw.crw_nonblock, io->u.ci_wr.wr_append);
+ io->u.ci_rw.crw_pos = pos;
+ io->u.ci_rw.crw_count = count;
+ RETURN(cl_io_init(env, io, iot, io->ci_obj));
}
EXPORT_SYMBOL(cl_io_rw_init);
-static inline const struct lu_fid *
-cl_lock_descr_fid(const struct cl_lock_descr *descr)
+static int cl_lock_descr_cmp(void *priv,
+ struct list_head *a, struct list_head *b)
{
- return lu_object_fid(&descr->cld_obj->co_lu);
-}
+ const struct cl_io_lock_link *l0 = list_entry(a, struct cl_io_lock_link,
+ cill_linkage);
+ const struct cl_io_lock_link *l1 = list_entry(b, struct cl_io_lock_link,
+ cill_linkage);
+ const struct cl_lock_descr *d0 = &l0->cill_descr;
+ const struct cl_lock_descr *d1 = &l1->cill_descr;
-static int cl_lock_descr_sort(const struct cl_lock_descr *d0,
- const struct cl_lock_descr *d1)
-{
- return lu_fid_cmp(cl_lock_descr_fid(d0), cl_lock_descr_fid(d1)) ?:
- __diff_normalize(d0->cld_start, d1->cld_start);
-}
-
-static int cl_lock_descr_cmp(const struct cl_lock_descr *d0,
- const struct cl_lock_descr *d1)
-{
- int ret;
-
- ret = lu_fid_cmp(cl_lock_descr_fid(d0), cl_lock_descr_fid(d1));
- if (ret)
- return ret;
- if (d0->cld_end < d1->cld_start)
- return -1;
- if (d0->cld_start > d0->cld_end)
- return 1;
- return 0;
+ return lu_fid_cmp(lu_object_fid(&d0->cld_obj->co_lu),
+ lu_object_fid(&d1->cld_obj->co_lu));
}
static void cl_lock_descr_merge(struct cl_lock_descr *d0,
- const struct cl_lock_descr *d1)
+ const struct cl_lock_descr *d1)
{
- d0->cld_start = min(d0->cld_start, d1->cld_start);
- d0->cld_end = max(d0->cld_end, d1->cld_end);
+ d0->cld_start = min(d0->cld_start, d1->cld_start);
+ d0->cld_end = max(d0->cld_end, d1->cld_end);
- if (d1->cld_mode == CLM_WRITE && d0->cld_mode != CLM_WRITE)
- d0->cld_mode = CLM_WRITE;
+ if (d1->cld_mode == CLM_WRITE && d0->cld_mode != CLM_WRITE)
+ d0->cld_mode = CLM_WRITE;
- if (d1->cld_mode == CLM_GROUP && d0->cld_mode != CLM_GROUP)
- d0->cld_mode = CLM_GROUP;
+ if (d1->cld_mode == CLM_GROUP && d0->cld_mode != CLM_GROUP)
+ d0->cld_mode = CLM_GROUP;
}
-/*
- * Sort locks in lexicographical order of their (fid, start-offset) pairs.
- */
-static void cl_io_locks_sort(struct cl_io *io)
-{
- int done = 0;
-
- ENTRY;
- /* hidden treasure: bubble sort for now. */
- do {
- struct cl_io_lock_link *curr;
- struct cl_io_lock_link *prev;
- struct cl_io_lock_link *temp;
-
- done = 1;
- prev = NULL;
-
- list_for_each_entry_safe(curr, temp, &io->ci_lockset.cls_todo,
- cill_linkage) {
- if (prev != NULL) {
- switch (cl_lock_descr_sort(&prev->cill_descr,
- &curr->cill_descr)) {
- case 0:
- /*
- * IMPOSSIBLE: Identical locks are
- * already removed at
- * this point.
- */
- default:
- LBUG();
- case +1:
- list_move_tail(&curr->cill_linkage,
- &prev->cill_linkage);
- done = 0;
- continue; /* don't change prev: it's
- * still "previous" */
- case -1: /* already in order */
- break;
- }
- }
- prev = curr;
- }
- } while (!done);
- EXIT;
-}
-
-/**
- * Check whether \a queue contains locks matching \a need.
- *
- * \retval +ve there is a matching lock in the \a queue
- * \retval 0 there are no matching locks in the \a queue
- */
-int cl_queue_match(const struct list_head *queue,
- const struct cl_lock_descr *need)
+static int cl_lockset_merge(const struct cl_lockset *set,
+ const struct cl_lock_descr *need)
{
struct cl_io_lock_link *scan;
- ENTRY;
- list_for_each_entry(scan, queue, cill_linkage) {
- if (cl_lock_descr_match(&scan->cill_descr, need))
- RETURN(+1);
- }
- RETURN(0);
-}
-EXPORT_SYMBOL(cl_queue_match);
-
-static int cl_queue_merge(const struct list_head *queue,
- const struct cl_lock_descr *need)
-{
- struct cl_io_lock_link *scan;
ENTRY;
-
- list_for_each_entry(scan, queue, cill_linkage) {
- if (cl_lock_descr_cmp(&scan->cill_descr, need))
+ list_for_each_entry(scan, &set->cls_todo, cill_linkage) {
+ if (!cl_object_same(scan->cill_descr.cld_obj, need->cld_obj))
continue;
+
+ /* Merge locks for the same object because ldlm lock server
+ * may expand the lock extent, otherwise there is a deadlock
+ * case if two conflicted locks are queueud for the same object
+ * and lock server expands one lock to overlap the another.
+ * The side effect is that it can generate a multi-stripe lock
+ * that may cause casacading problem */
cl_lock_descr_merge(&scan->cill_descr, need);
CDEBUG(D_VFSTRACE, "lock: %d: [%lu, %lu]\n",
scan->cill_descr.cld_mode, scan->cill_descr.cld_start,
RETURN(0);
}
-static int cl_lockset_match(const struct cl_lockset *set,
- const struct cl_lock_descr *need)
-{
- return cl_queue_match(&set->cls_curr, need) ||
- cl_queue_match(&set->cls_done, need);
-}
-
-static int cl_lockset_merge(const struct cl_lockset *set,
- const struct cl_lock_descr *need)
-{
- return cl_queue_merge(&set->cls_todo, need) ||
- cl_lockset_match(set, need);
-}
-
-static int cl_lockset_lock_one(const struct lu_env *env,
- struct cl_io *io, struct cl_lockset *set,
- struct cl_io_lock_link *link)
-{
- struct cl_lock *lock;
- int result;
-
- ENTRY;
-
- lock = cl_lock_request(env, io, &link->cill_descr, "io", io);
-
- if (!IS_ERR(lock)) {
- link->cill_lock = lock;
- list_move(&link->cill_linkage, &set->cls_curr);
- if (!(link->cill_descr.cld_enq_flags & CEF_ASYNC)) {
- result = cl_wait(env, lock);
- if (result == 0)
- list_move(&link->cill_linkage, &set->cls_done);
- } else
- result = 0;
- } else
- result = PTR_ERR(lock);
- RETURN(result);
-}
-
-static void cl_lock_link_fini(const struct lu_env *env, struct cl_io *io,
- struct cl_io_lock_link *link)
-{
- struct cl_lock *lock = link->cill_lock;
-
- ENTRY;
- list_del_init(&link->cill_linkage);
- if (lock != NULL) {
- cl_lock_release(env, lock, "io", io);
- link->cill_lock = NULL;
- }
- if (link->cill_fini != NULL)
- link->cill_fini(env, link);
- EXIT;
-}
-
static int cl_lockset_lock(const struct lu_env *env, struct cl_io *io,
- struct cl_lockset *set)
+ struct cl_lockset *set)
{
- struct cl_io_lock_link *link;
- struct cl_io_lock_link *temp;
- struct cl_lock *lock;
+ struct cl_io_lock_link *link;
+ struct cl_io_lock_link *temp;
int result;
ENTRY;
result = 0;
list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage) {
- if (!cl_lockset_match(set, &link->cill_descr)) {
- /* XXX some locking to guarantee that locks aren't
- * expanded in between. */
- result = cl_lockset_lock_one(env, io, set, link);
- if (result != 0)
- break;
- } else
- cl_lock_link_fini(env, io, link);
- }
- if (result == 0) {
- list_for_each_entry_safe(link, temp, &set->cls_curr,
- cill_linkage) {
- lock = link->cill_lock;
- result = cl_wait(env, lock);
- if (result == 0)
- list_move(&link->cill_linkage, &set->cls_done);
- else
- break;
- }
+ result = cl_lock_request(env, io, &link->cill_lock);
+ if (result < 0)
+ break;
+
+ list_move(&link->cill_linkage, &set->cls_done);
}
RETURN(result);
}
LINVRNT(cl_io_invariant(io));
ENTRY;
- cl_io_for_each(scan, io) {
- if (scan->cis_iop->op[io->ci_type].cio_lock == NULL)
- continue;
- result = scan->cis_iop->op[io->ci_type].cio_lock(env, scan);
- if (result != 0)
- break;
- }
+ list_for_each_entry(scan, &io->ci_layers, cis_linkage) {
+ if (scan->cis_iop->op[io->ci_type].cio_lock == NULL)
+ continue;
+ result = scan->cis_iop->op[io->ci_type].cio_lock(env, scan);
+ if (result != 0)
+ break;
+ }
if (result == 0) {
- cl_io_locks_sort(io);
+ /*
+ * Sort locks in lexicographical order of their (fid,
+ * start-offset) pairs to avoid deadlocks.
+ */
+ list_sort(NULL, &io->ci_lockset.cls_todo, cl_lock_descr_cmp);
result = cl_lockset_lock(env, io, &io->ci_lockset);
}
if (result != 0)
ENTRY;
set = &io->ci_lockset;
- list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage)
- cl_lock_link_fini(env, io, link);
-
- list_for_each_entry_safe(link, temp, &set->cls_curr, cill_linkage)
- cl_lock_link_fini(env, io, link);
+ list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage) {
+ list_del_init(&link->cill_linkage);
+ if (link->cill_fini != NULL)
+ link->cill_fini(env, link);
+ }
list_for_each_entry_safe(link, temp, &set->cls_done, cill_linkage) {
- cl_unuse(env, link->cill_lock);
- cl_lock_link_fini(env, io, link);
+ list_del_init(&link->cill_linkage);
+ cl_lock_release(env, &link->cill_lock);
+ if (link->cill_fini != NULL)
+ link->cill_fini(env, link);
}
- cl_io_for_each_reverse(scan, io) {
+
+ list_for_each_entry_reverse(scan, &io->ci_layers, cis_linkage) {
if (scan->cis_iop->op[io->ci_type].cio_unlock != NULL)
scan->cis_iop->op[io->ci_type].cio_unlock(env, scan);
}
io->ci_state = CIS_UNLOCKED;
- LASSERT(!cl_env_info(env)->clt_counters[CNL_TOP].ctc_nr_locks_acquired);
EXIT;
}
EXPORT_SYMBOL(cl_io_unlock);
ENTRY;
result = 0;
- cl_io_for_each(scan, io) {
- if (scan->cis_iop->op[io->ci_type].cio_iter_init == NULL)
- continue;
- result = scan->cis_iop->op[io->ci_type].cio_iter_init(env,
- scan);
- if (result != 0)
- break;
- }
+ list_for_each_entry(scan, &io->ci_layers, cis_linkage) {
+ if (scan->cis_iop->op[io->ci_type].cio_iter_init == NULL)
+ continue;
+ result = scan->cis_iop->op[io->ci_type].cio_iter_init(env,
+ scan);
+ if (result != 0)
+ break;
+ }
if (result == 0)
io->ci_state = CIS_IT_STARTED;
RETURN(result);
*/
void cl_io_iter_fini(const struct lu_env *env, struct cl_io *io)
{
- const struct cl_io_slice *scan;
+ const struct cl_io_slice *scan;
- LINVRNT(cl_io_is_loopable(io));
- LINVRNT(io->ci_state == CIS_UNLOCKED);
- LINVRNT(cl_io_invariant(io));
+ LINVRNT(cl_io_is_loopable(io));
+ LINVRNT(io->ci_state <= CIS_IT_STARTED ||
+ io->ci_state > CIS_IO_FINISHED);
+ LINVRNT(cl_io_invariant(io));
- ENTRY;
- cl_io_for_each_reverse(scan, io) {
- if (scan->cis_iop->op[io->ci_type].cio_iter_fini != NULL)
- scan->cis_iop->op[io->ci_type].cio_iter_fini(env, scan);
- }
- io->ci_state = CIS_IT_ENDED;
- EXIT;
+ ENTRY;
+ list_for_each_entry_reverse(scan, &io->ci_layers, cis_linkage) {
+ if (scan->cis_iop->op[io->ci_type].cio_iter_fini != NULL)
+ scan->cis_iop->op[io->ci_type].cio_iter_fini(env, scan);
+ }
+ io->ci_state = CIS_IT_ENDED;
+ EXIT;
}
EXPORT_SYMBOL(cl_io_iter_fini);
*/
void cl_io_rw_advance(const struct lu_env *env, struct cl_io *io, size_t nob)
{
- const struct cl_io_slice *scan;
+ const struct cl_io_slice *scan;
- LINVRNT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE ||
- nob == 0);
- LINVRNT(cl_io_is_loopable(io));
- LINVRNT(cl_io_invariant(io));
+ ENTRY;
- ENTRY;
+ LINVRNT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE ||
+ nob == 0);
+ LINVRNT(cl_io_is_loopable(io));
+ LINVRNT(cl_io_invariant(io));
- io->u.ci_rw.crw_pos += nob;
- io->u.ci_rw.crw_count -= nob;
+ io->u.ci_rw.crw_pos += nob;
+ io->u.ci_rw.crw_count -= nob;
- /* layers have to be notified. */
- cl_io_for_each_reverse(scan, io) {
- if (scan->cis_iop->op[io->ci_type].cio_advance != NULL)
- scan->cis_iop->op[io->ci_type].cio_advance(env, scan,
- nob);
- }
- EXIT;
+ /* layers have to be notified. */
+ list_for_each_entry_reverse(scan, &io->ci_layers, cis_linkage) {
+ if (scan->cis_iop->op[io->ci_type].cio_advance != NULL)
+ scan->cis_iop->op[io->ci_type].cio_advance(env, scan,
+ nob);
+ }
+ EXIT;
}
-EXPORT_SYMBOL(cl_io_rw_advance);
/**
* Adds a lock to a lockset.
int cl_io_lock_alloc_add(const struct lu_env *env, struct cl_io *io,
struct cl_lock_descr *descr)
{
- struct cl_io_lock_link *link;
- int result;
+ struct cl_io_lock_link *link;
+ int result;
- ENTRY;
- OBD_ALLOC_PTR(link);
- if (link != NULL) {
- link->cill_descr = *descr;
- link->cill_fini = cl_free_io_lock_link;
- result = cl_io_lock_add(env, io, link);
- if (result) /* lock match */
- link->cill_fini(env, link);
- } else
- result = -ENOMEM;
+ ENTRY;
+ OBD_ALLOC_PTR(link);
+ if (link != NULL) {
+ link->cill_descr = *descr;
+ link->cill_fini = cl_free_io_lock_link;
+ result = cl_io_lock_add(env, io, link);
+ if (result) /* lock match */
+ link->cill_fini(env, link);
+ } else
+ result = -ENOMEM;
- RETURN(result);
+ RETURN(result);
}
EXPORT_SYMBOL(cl_io_lock_alloc_add);
ENTRY;
io->ci_state = CIS_IO_GOING;
- cl_io_for_each(scan, io) {
- if (scan->cis_iop->op[io->ci_type].cio_start == NULL)
- continue;
- result = scan->cis_iop->op[io->ci_type].cio_start(env, scan);
- if (result != 0)
- break;
- }
+ list_for_each_entry(scan, &io->ci_layers, cis_linkage) {
+ if (scan->cis_iop->op[io->ci_type].cio_start == NULL)
+ continue;
+ result = scan->cis_iop->op[io->ci_type].cio_start(env, scan);
+ if (result != 0)
+ break;
+ }
if (result >= 0)
result = 0;
RETURN(result);
LINVRNT(cl_io_invariant(io));
ENTRY;
- cl_io_for_each_reverse(scan, io) {
- if (scan->cis_iop->op[io->ci_type].cio_end != NULL)
- scan->cis_iop->op[io->ci_type].cio_end(env, scan);
- /* TODO: error handling. */
- }
+ list_for_each_entry_reverse(scan, &io->ci_layers, cis_linkage) {
+ if (scan->cis_iop->op[io->ci_type].cio_end != NULL)
+ scan->cis_iop->op[io->ci_type].cio_end(env, scan);
+ /* TODO: error handling. */
+ }
io->ci_state = CIS_IO_FINISHED;
EXIT;
}
EXPORT_SYMBOL(cl_io_end);
-static const struct cl_page_slice *
-cl_io_slice_page(const struct cl_io_slice *ios, struct cl_page *page)
-{
- const struct cl_page_slice *slice;
-
- slice = cl_page_at(page, ios->cis_obj->co_lu.lo_dev->ld_type);
- LINVRNT(slice != NULL);
- return slice;
-}
-
/**
- * Called by read io, when page has to be read from the server.
+ * Called by read io, to decide the readahead extent
*
- * \see cl_io_operations::cio_read_page()
+ * \see cl_io_operations::cio_read_ahead()
*/
-int cl_io_read_page(const struct lu_env *env, struct cl_io *io,
- struct cl_page *page)
+int cl_io_read_ahead(const struct lu_env *env, struct cl_io *io,
+ pgoff_t start, struct cl_read_ahead *ra)
{
- const struct cl_io_slice *scan;
- struct cl_2queue *queue;
- int result = 0;
+ const struct cl_io_slice *scan;
+ int result = 0;
- LINVRNT(io->ci_type == CIT_READ || io->ci_type == CIT_FAULT);
- LINVRNT(cl_page_is_owned(page, io));
- LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED);
- LINVRNT(cl_io_invariant(io));
- ENTRY;
+ LINVRNT(io->ci_type == CIT_READ ||
+ io->ci_type == CIT_FAULT ||
+ io->ci_type == CIT_WRITE);
+ LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED);
+ LINVRNT(cl_io_invariant(io));
+ ENTRY;
- queue = &io->ci_queue;
+ list_for_each_entry(scan, &io->ci_layers, cis_linkage) {
+ if (scan->cis_iop->cio_read_ahead == NULL)
+ continue;
- cl_2queue_init(queue);
- /*
- * ->cio_read_page() methods called in the loop below are supposed to
- * never block waiting for network (the only subtle point is the
- * creation of new pages for read-ahead that might result in cache
- * shrinking, but currently only clean pages are shrunk and this
- * requires no network io).
- *
- * Should this ever starts blocking, retry loop would be needed for
- * "parallel io" (see CLO_REPEAT loops in cl_lock.c).
- */
- cl_io_for_each(scan, io) {
- if (scan->cis_iop->cio_read_page != NULL) {
- const struct cl_page_slice *slice;
-
- slice = cl_io_slice_page(scan, page);
- LINVRNT(slice != NULL);
- result = scan->cis_iop->cio_read_page(env, scan, slice);
- if (result != 0)
- break;
- }
- }
- if (result == 0 && queue->c2_qin.pl_nr > 0)
- result = cl_io_submit_rw(env, io, CRT_READ, queue);
- /*
- * Unlock unsent pages in case of error.
- */
- cl_page_list_disown(env, io, &queue->c2_qin);
- cl_2queue_fini(env, queue);
- RETURN(result);
+ result = scan->cis_iop->cio_read_ahead(env, scan, start, ra);
+ if (result != 0)
+ break;
+ }
+ RETURN(result > 0 ? 0 : result);
}
-EXPORT_SYMBOL(cl_io_read_page);
+EXPORT_SYMBOL(cl_io_read_ahead);
/**
* Commit a list of contiguous pages into writeback cache.
* \see cl_io_operations::cio_commit_async()
*/
int cl_io_commit_async(const struct lu_env *env, struct cl_io *io,
- struct cl_page_list *queue, int from, int to,
- cl_commit_cbt cb)
+ struct cl_page_list *queue, int from, int to,
+ cl_commit_cbt cb)
{
const struct cl_io_slice *scan;
int result = 0;
ENTRY;
- cl_io_for_each(scan, io) {
+ list_for_each_entry(scan, &io->ci_layers, cis_linkage) {
if (scan->cis_iop->cio_commit_async == NULL)
continue;
result = scan->cis_iop->cio_commit_async(env, scan, queue,
int result = 0;
ENTRY;
- cl_io_for_each(scan, io) {
+ list_for_each_entry(scan, &io->ci_layers, cis_linkage) {
if (scan->cis_iop->cio_submit == NULL)
continue;
result = scan->cis_iop->cio_submit(env, scan, crt, queue);
pg->cp_sync_io = anchor;
}
- cl_sync_io_init(anchor, queue->c2_qin.pl_nr, &cl_sync_io_end);
+ cl_sync_io_init(anchor, queue->c2_qin.pl_nr);
rc = cl_io_submit_rw(env, io, iot, queue);
if (rc == 0) {
/*
}
return result;
}
-EXPORT_SYMBOL(cl_io_cancel);
/**
* Main io loop.
*/
int cl_io_loop(const struct lu_env *env, struct cl_io *io)
{
- int result = 0;
+ int result = 0;
- LINVRNT(cl_io_is_loopable(io));
- ENTRY;
+ LINVRNT(cl_io_is_loopable(io));
+ ENTRY;
+
+ do {
+ size_t nob;
+
+ io->ci_continue = 0;
+ result = cl_io_iter_init(env, io);
+ if (result == 0) {
+ nob = io->ci_nob;
+ result = cl_io_lock(env, io);
+ if (result == 0) {
+ /*
+ * Notify layers that locks has been taken,
+ * and do actual i/o.
+ *
+ * - llite: kms, short read;
+ * - llite: generic_file_read();
+ */
+ result = cl_io_start(env, io);
+ /*
+ * Send any remaining pending
+ * io, etc.
+ *
+ ** - llite: ll_rw_stats_tally.
+ */
+ cl_io_end(env, io);
+ cl_io_unlock(env, io);
+ cl_io_rw_advance(env, io, io->ci_nob - nob);
+ }
+ }
+ cl_io_iter_fini(env, io);
+ } while (result == 0 && io->ci_continue);
+
+ if (result == -EWOULDBLOCK && io->ci_ndelay) {
+ io->ci_need_restart = 1;
+ result = 0;
+ }
- do {
- size_t nob;
-
- io->ci_continue = 0;
- result = cl_io_iter_init(env, io);
- if (result == 0) {
- nob = io->ci_nob;
- result = cl_io_lock(env, io);
- if (result == 0) {
- /*
- * Notify layers that locks has been taken,
- * and do actual i/o.
- *
- * - llite: kms, short read;
- * - llite: generic_file_read();
- */
- result = cl_io_start(env, io);
- /*
- * Send any remaining pending
- * io, etc.
- *
- * - llite: ll_rw_stats_tally.
- */
- cl_io_end(env, io);
- cl_io_unlock(env, io);
- cl_io_rw_advance(env, io, io->ci_nob - nob);
- }
- }
- cl_io_iter_fini(env, io);
- } while (result == 0 && io->ci_continue);
if (result == 0)
result = io->ci_result;
RETURN(result < 0 ? result : 0);
* \see cl_lock_slice_add(), cl_req_slice_add(), cl_page_slice_add()
*/
void cl_io_slice_add(struct cl_io *io, struct cl_io_slice *slice,
- struct cl_object *obj,
- const struct cl_io_operations *ops)
+ struct cl_object *obj,
+ const struct cl_io_operations *ops)
{
struct list_head *linkage = &slice->cis_linkage;
- LASSERT((linkage->prev == NULL && linkage->next == NULL) ||
+ LASSERT((linkage->prev == NULL && linkage->next == NULL) ||
list_empty(linkage));
- ENTRY;
+ ENTRY;
list_add_tail(linkage, &io->ci_layers);
- slice->cis_io = io;
- slice->cis_obj = obj;
- slice->cis_iop = ops;
- EXIT;
+ slice->cis_io = io;
+ slice->cis_obj = obj;
+ slice->cis_iop = ops;
+ EXIT;
}
EXPORT_SYMBOL(cl_io_slice_add);
}
EXPORT_SYMBOL(cl_page_list_splice);
-void cl_page_disown0(const struct lu_env *env,
- struct cl_io *io, struct cl_page *pg);
-
/**
* Disowns pages in a queue.
*/
EXPORT_SYMBOL(cl_page_list_fini);
/**
- * Owns all pages in a queue.
- */
-int cl_page_list_own(const struct lu_env *env,
- struct cl_io *io, struct cl_page_list *plist)
-{
- struct cl_page *page;
- struct cl_page *temp;
- int result;
-
- LINVRNT(plist->pl_owner == current);
-
- ENTRY;
- result = 0;
- cl_page_list_for_each_safe(page, temp, plist) {
- if (cl_page_own(env, io, page) == 0)
- result = result ?: page->cp_error;
- else
- cl_page_list_del(env, plist, page);
- }
- RETURN(result);
-}
-EXPORT_SYMBOL(cl_page_list_own);
-
-/**
* Assumes all pages in a queue.
*/
void cl_page_list_assume(const struct lu_env *env,
cl_page_list_for_each(page, plist)
cl_page_assume(env, io, page);
}
-EXPORT_SYMBOL(cl_page_list_assume);
/**
* Discards all pages in a queue.
cl_page_list_assume(env, io, &queue->c2_qin);
cl_page_list_assume(env, io, &queue->c2_qout);
}
-EXPORT_SYMBOL(cl_2queue_assume);
/**
* Finalize both page lists of a 2-queue.
}
/**
- * Adds request slice to the compound request.
- *
- * This is called by cl_device_operations::cdo_req_init() methods to add a
- * per-layer state to the request. New state is added at the end of
- * cl_req::crq_layers list, that is, it is at the bottom of the stack.
- *
- * \see cl_lock_slice_add(), cl_page_slice_add(), cl_io_slice_add()
- */
-void cl_req_slice_add(struct cl_req *req, struct cl_req_slice *slice,
- struct cl_device *dev,
- const struct cl_req_operations *ops)
-{
- ENTRY;
- list_add_tail(&slice->crs_linkage, &req->crq_layers);
- slice->crs_dev = dev;
- slice->crs_ops = ops;
- slice->crs_req = req;
- EXIT;
-}
-EXPORT_SYMBOL(cl_req_slice_add);
-
-static void cl_req_free(const struct lu_env *env, struct cl_req *req)
-{
- unsigned i;
-
- LASSERT(list_empty(&req->crq_pages));
- LASSERT(req->crq_nrpages == 0);
- LINVRNT(list_empty(&req->crq_layers));
- LINVRNT(equi(req->crq_nrobjs > 0, req->crq_o != NULL));
- ENTRY;
-
- if (req->crq_o != NULL) {
- for (i = 0; i < req->crq_nrobjs; ++i) {
- struct cl_object *obj = req->crq_o[i].ro_obj;
- if (obj != NULL) {
- lu_object_ref_del_at(&obj->co_lu,
- &req->crq_o[i].ro_obj_ref,
- "cl_req", req);
- cl_object_put(env, obj);
- }
- }
- OBD_FREE(req->crq_o, req->crq_nrobjs * sizeof req->crq_o[0]);
- }
- OBD_FREE_PTR(req);
- EXIT;
-}
-
-static int cl_req_init(const struct lu_env *env, struct cl_req *req,
- struct cl_page *page)
-{
- struct cl_device *dev;
- struct cl_page_slice *slice;
- int result;
-
- ENTRY;
- result = 0;
- list_for_each_entry(slice, &page->cp_layers, cpl_linkage) {
- dev = lu2cl_dev(slice->cpl_obj->co_lu.lo_dev);
- if (dev->cd_ops->cdo_req_init != NULL) {
- result = dev->cd_ops->cdo_req_init(env,
- dev, req);
- if (result != 0)
- break;
- }
- }
- RETURN(result);
-}
-
-/**
- * Invokes per-request transfer completion call-backs
- * (cl_req_operations::cro_completion()) bottom-to-top.
- */
-void cl_req_completion(const struct lu_env *env, struct cl_req *req, int rc)
-{
- struct cl_req_slice *slice;
-
- ENTRY;
- /*
- * for the lack of list_for_each_entry_reverse_safe()...
- */
- while (!list_empty(&req->crq_layers)) {
- slice = list_entry(req->crq_layers.prev,
- struct cl_req_slice, crs_linkage);
- list_del_init(&slice->crs_linkage);
- if (slice->crs_ops->cro_completion != NULL)
- slice->crs_ops->cro_completion(env, slice, rc);
- }
- cl_req_free(env, req);
- EXIT;
-}
-EXPORT_SYMBOL(cl_req_completion);
-
-/**
- * Allocates new transfer request.
- */
-struct cl_req *cl_req_alloc(const struct lu_env *env, struct cl_page *page,
- enum cl_req_type crt, int nr_objects)
-{
- struct cl_req *req;
-
- LINVRNT(nr_objects > 0);
- ENTRY;
-
- OBD_ALLOC_PTR(req);
- if (req != NULL) {
- int result;
-
- req->crq_type = crt;
- INIT_LIST_HEAD(&req->crq_pages);
- INIT_LIST_HEAD(&req->crq_layers);
-
- OBD_ALLOC(req->crq_o, nr_objects * sizeof req->crq_o[0]);
- if (req->crq_o != NULL) {
- req->crq_nrobjs = nr_objects;
- result = cl_req_init(env, req, page);
- } else
- result = -ENOMEM;
- if (result != 0) {
- cl_req_completion(env, req, result);
- req = ERR_PTR(result);
- }
- } else
- req = ERR_PTR(-ENOMEM);
- RETURN(req);
-}
-EXPORT_SYMBOL(cl_req_alloc);
-
-/**
- * Adds a page to a request.
- */
-void cl_req_page_add(const struct lu_env *env,
- struct cl_req *req, struct cl_page *page)
-{
- struct cl_object *obj;
- struct cl_req_obj *rqo;
- int i;
-
- ENTRY;
-
- LASSERT(list_empty(&page->cp_flight));
- LASSERT(page->cp_req == NULL);
-
- CL_PAGE_DEBUG(D_PAGE, env, page, "req %p, %d, %u\n",
- req, req->crq_type, req->crq_nrpages);
-
- list_add_tail(&page->cp_flight, &req->crq_pages);
- ++req->crq_nrpages;
- page->cp_req = req;
- obj = cl_object_top(page->cp_obj);
- for (i = 0, rqo = req->crq_o; obj != rqo->ro_obj; ++i, ++rqo) {
- if (rqo->ro_obj == NULL) {
- rqo->ro_obj = obj;
- cl_object_get(obj);
- lu_object_ref_add_at(&obj->co_lu, &rqo->ro_obj_ref,
- "cl_req", req);
- break;
- }
- }
- LASSERT(i < req->crq_nrobjs);
- EXIT;
-}
-EXPORT_SYMBOL(cl_req_page_add);
-
-/**
- * Removes a page from a request.
- */
-void cl_req_page_done(const struct lu_env *env, struct cl_page *page)
-{
- struct cl_req *req = page->cp_req;
-
- ENTRY;
-
- LASSERT(!list_empty(&page->cp_flight));
- LASSERT(req->crq_nrpages > 0);
-
- list_del_init(&page->cp_flight);
- --req->crq_nrpages;
- page->cp_req = NULL;
- EXIT;
-}
-EXPORT_SYMBOL(cl_req_page_done);
-
-/**
- * Notifies layers that request is about to depart by calling
- * cl_req_operations::cro_prep() top-to-bottom.
- */
-int cl_req_prep(const struct lu_env *env, struct cl_req *req)
-{
- int i;
- int result;
- const struct cl_req_slice *slice;
-
- ENTRY;
- /*
- * Check that the caller of cl_req_alloc() didn't lie about the number
- * of objects.
- */
- for (i = 0; i < req->crq_nrobjs; ++i)
- LASSERT(req->crq_o[i].ro_obj != NULL);
-
- result = 0;
- list_for_each_entry(slice, &req->crq_layers, crs_linkage) {
- if (slice->crs_ops->cro_prep != NULL) {
- result = slice->crs_ops->cro_prep(env, slice);
- if (result != 0)
- break;
- }
- }
- RETURN(result);
-}
-EXPORT_SYMBOL(cl_req_prep);
-
-/**
* Fills in attributes that are passed to server together with transfer. Only
* attributes from \a flags may be touched. This can be called multiple times
* for the same request.
*/
-void cl_req_attr_set(const struct lu_env *env, struct cl_req *req,
- struct cl_req_attr *attr, obd_valid flags)
+void cl_req_attr_set(const struct lu_env *env, struct cl_object *obj,
+ struct cl_req_attr *attr)
{
- const struct cl_req_slice *slice;
- struct cl_page *page;
- int i;
-
- LASSERT(!list_empty(&req->crq_pages));
- ENTRY;
+ struct cl_object *scan;
+ ENTRY;
- /* Take any page to use as a model. */
- page = list_entry(req->crq_pages.next, struct cl_page, cp_flight);
-
- for (i = 0; i < req->crq_nrobjs; ++i) {
- list_for_each_entry(slice, &req->crq_layers, crs_linkage) {
- const struct cl_page_slice *scan;
- const struct cl_object *obj;
-
- scan = cl_page_at(page,
- slice->crs_dev->cd_lu_dev.ld_type);
- LASSERT(scan != NULL);
- obj = scan->cpl_obj;
- if (slice->crs_ops->cro_attr_set != NULL)
- slice->crs_ops->cro_attr_set(env, slice, obj,
- attr + i, flags);
- }
- }
- EXIT;
+ cl_object_for_each(scan, obj) {
+ if (scan->co_ops->coo_req_attr_set != NULL)
+ scan->co_ops->coo_req_attr_set(env, scan, attr);
+ }
+ EXIT;
}
EXPORT_SYMBOL(cl_req_attr_set);
-/* cl_sync_io_callback assumes the caller must call cl_sync_io_wait() to
- * wait for the IO to finish. */
-void cl_sync_io_end(const struct lu_env *env, struct cl_sync_io *anchor)
-{
- wake_up_all(&anchor->csi_waitq);
-
- /* it's safe to nuke or reuse anchor now */
- atomic_set(&anchor->csi_barrier, 0);
-}
-EXPORT_SYMBOL(cl_sync_io_end);
-
/**
- * Initialize synchronous io wait anchor
+ * Initialize synchronous io wait \a anchor for \a nr pages with optional
+ * \a end handler.
+ * \param anchor owned by caller, initialzied here.
+ * \param nr number of pages initally pending in sync.
+ * \param end optional callback sync_io completion, can be used to
+ * trigger erasure coding, integrity, dedupe, or similar operation.
+ * \q end is called with a spinlock on anchor->csi_waitq.lock
*/
-void cl_sync_io_init(struct cl_sync_io *anchor, int nr,
- void (*end)(const struct lu_env *, struct cl_sync_io *))
+
+void cl_sync_io_init_notify(struct cl_sync_io *anchor, int nr,
+ cl_sync_io_end_t *end)
{
ENTRY;
+ memset(anchor, 0, sizeof(*anchor));
init_waitqueue_head(&anchor->csi_waitq);
atomic_set(&anchor->csi_sync_nr, nr);
- atomic_set(&anchor->csi_barrier, nr > 0);
anchor->csi_sync_rc = 0;
anchor->csi_end_io = end;
- LASSERT(end != NULL);
EXIT;
}
-EXPORT_SYMBOL(cl_sync_io_init);
+EXPORT_SYMBOL(cl_sync_io_init_notify);
/**
* Wait until all IO completes. Transfer completion routine has to call
int cl_sync_io_wait(const struct lu_env *env, struct cl_sync_io *anchor,
long timeout)
{
- struct l_wait_info lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(timeout),
- NULL, NULL, NULL);
- int rc;
+ int rc = 0;
ENTRY;
LASSERT(timeout >= 0);
- rc = l_wait_event(anchor->csi_waitq,
- atomic_read(&anchor->csi_sync_nr) == 0,
- &lwi);
- if (rc < 0) {
+ if (timeout > 0 &&
+ wait_event_idle_timeout(anchor->csi_waitq,
+ atomic_read(&anchor->csi_sync_nr) == 0,
+ cfs_time_seconds(timeout)) == 0) {
+ rc = -ETIMEDOUT;
CERROR("IO failed: %d, still wait for %d remaining entries\n",
rc, atomic_read(&anchor->csi_sync_nr));
+ }
- lwi = (struct l_wait_info) { 0 };
- (void)l_wait_event(anchor->csi_waitq,
- atomic_read(&anchor->csi_sync_nr) == 0,
- &lwi);
- } else {
+ wait_event_idle(anchor->csi_waitq,
+ atomic_read(&anchor->csi_sync_nr) == 0);
+ if (!rc)
rc = anchor->csi_sync_rc;
- }
+
+ /* We take the lock to ensure that cl_sync_io_note() has finished */
+ spin_lock(&anchor->csi_waitq.lock);
LASSERT(atomic_read(&anchor->csi_sync_nr) == 0);
+ spin_unlock(&anchor->csi_waitq.lock);
- /* wait until cl_sync_io_note() has done wakeup */
- while (unlikely(atomic_read(&anchor->csi_barrier) != 0)) {
- cpu_relax();
- }
RETURN(rc);
}
EXPORT_SYMBOL(cl_sync_io_wait);
* IO.
*/
LASSERT(atomic_read(&anchor->csi_sync_nr) > 0);
- if (atomic_dec_and_test(&anchor->csi_sync_nr)) {
- LASSERT(anchor->csi_end_io != NULL);
- anchor->csi_end_io(env, anchor);
+ if (atomic_dec_and_lock(&anchor->csi_sync_nr,
+ &anchor->csi_waitq.lock)) {
+ cl_sync_io_end_t *end_io = anchor->csi_end_io;
+
+ /*
+ * Holding the lock across both the decrement and
+ * the wakeup ensures cl_sync_io_wait() doesn't complete
+ * before the wakeup completes and the contents of
+ * of anchor become unsafe to access as the owner is free
+ * to immediately reclaim anchor when cl_sync_io_wait()
+ * completes.
+ */
+ wake_up_all_locked(&anchor->csi_waitq);
+ if (end_io)
+ end_io(env, anchor);
+ spin_unlock(&anchor->csi_waitq.lock);
+
/* Can't access anchor any more */
}
EXIT;