* Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
union lov_layout_state *state);
int (*llo_print)(const struct lu_env *env, void *cookie,
lu_printer_t p, const struct lu_object *o);
- struct cl_page *(*llo_page_init)(const struct lu_env *env,
- struct cl_object *obj,
- struct cl_page *page,
- cfs_page_t *vmpage);
+ int (*llo_page_init)(const struct lu_env *env, struct cl_object *obj,
+ struct cl_page *page, cfs_page_t *vmpage);
int (*llo_lock_init)(const struct lu_env *env,
struct cl_object *obj, struct cl_lock *lock,
const struct cl_io *io);
struct lov_object *lov,
union lov_layout_state *state)
{
- lov->u = *state;
}
static struct cl_object *lov_sub_find(const struct lu_env *env,
struct lov_oinfo *oinfo;
int result;
+ if (OBD_FAIL_CHECK(OBD_FAIL_LOV_INIT)) {
+ /* For sanity:test_206.
+ * Do not leave the object in cache to avoid accessing
+ * freed memory. This is because osc_object is referring to
+ * lov_oinfo of lsm_stripe_data which will be freed due to
+ * this failure. */
+ cl_object_kill(env, stripe);
+ cl_object_put(env, stripe);
+ return -EIO;
+ }
+
hdr = cl_object_header(lov2cl(lov));
subhdr = cl_object_header(stripe);
parent = subhdr->coh_parent;
if (r0->lo_sub != NULL) {
result = 0;
subconf->coc_inode = conf->coc_inode;
- cfs_spin_lock_init(&r0->lo_sub_lock);
+ spin_lock_init(&r0->lo_sub_lock);
/*
* Create stripe cl_objects.
*/
subdev = lovsub2cl_dev(dev->ld_target[ost_idx]);
subconf->u.coc_oinfo = oinfo;
LASSERTF(subdev != NULL, "not init ost %d\n", ost_idx);
+ /* In the function below, .hs_keycmp resolves to
+ * lu_obj_hop_keycmp() */
+ /* coverity[overrun-buffer-val] */
stripe = lov_sub_find(env, subdev, ofid, subconf);
if (!IS_ERR(stripe))
result = lov_init_sub(env, lov, stripe, r0, i);
union lov_layout_state *state)
{
LASSERT(lov->lo_type == LLT_EMPTY);
+ if (cfs_atomic_read(&lov->lo_active_ios) > 0)
+ RETURN(-EBUSY);
+
+ cl_object_prune(env, &lov->lo_cl);
return 0;
}
/* this wait-queue is signaled at the end of
* lu_object_free(). */
cfs_set_current_state(CFS_TASK_UNINT);
- cfs_spin_lock(&r0->lo_sub_lock);
- if (r0->lo_sub[idx] == los) {
- cfs_spin_unlock(&r0->lo_sub_lock);
- cfs_waitq_wait(waiter, CFS_TASK_UNINT);
- } else {
- cfs_spin_unlock(&r0->lo_sub_lock);
+ spin_lock(&r0->lo_sub_lock);
+ if (r0->lo_sub[idx] == los) {
+ spin_unlock(&r0->lo_sub_lock);
+ cfs_waitq_wait(waiter, CFS_TASK_UNINT);
+ } else {
+ spin_unlock(&r0->lo_sub_lock);
cfs_set_current_state(CFS_TASK_RUNNING);
break;
}
ENTRY;
dump_lsm(D_INODE, lsm);
- if (cfs_atomic_read(&lsm->lsm_refc) > 1)
+ if (cfs_atomic_read(&lov->lo_active_ios) > 0)
RETURN(-EBUSY);
if (r0->lo_sub != NULL) {
for (i = 0; i < r0->lo_nr; ++i) {
struct lovsub_object *los = r0->lo_sub[i];
- if (los != NULL)
+ if (los != NULL) {
+ cl_locks_prune(env, &los->lso_cl, 1);
/*
* If top-level object is to be evicted from
* the cache, so are its sub-objects.
*/
lov_subobject_kill(env, lov, los, i);
+ }
}
}
+ cl_object_prune(env, &lov->lo_cl);
RETURN(0);
}
static void lov_fini_raid0(const struct lu_env *env, struct lov_object *lov,
union lov_layout_state *state)
{
- struct lov_layout_raid0 *r0 = &state->raid0;
-
- ENTRY;
+ struct lov_layout_raid0 *r0 = &state->raid0;
+ ENTRY;
- if (r0->lo_sub != NULL) {
- OBD_FREE_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
- r0->lo_sub = NULL;
- }
+ if (r0->lo_sub != NULL) {
+ OBD_FREE_LARGE(r0->lo_sub, r0->lo_nr * sizeof r0->lo_sub[0]);
+ r0->lo_sub = NULL;
+ }
- LASSERTF(cfs_atomic_read(&lov->lo_lsm->lsm_refc) == 1,
- "actual %d proc %p.\n",
- cfs_atomic_read(&lov->lo_lsm->lsm_refc), cfs_current());
+ dump_lsm(D_INODE, lov->lo_lsm);
lov_free_memmd(&lov->lo_lsm);
- lov->lo_lsm = NULL;
EXIT;
}
static int lov_attr_get_raid0(const struct lu_env *env, struct cl_object *obj,
struct cl_attr *attr)
{
- struct lov_object *lov = cl2lov(obj);
+ struct lov_object *lov = cl2lov(obj);
struct lov_layout_raid0 *r0 = lov_r0(lov);
- struct lov_stripe_md *lsm = lov->lo_lsm;
- struct ost_lvb *lvb = &lov_env_info(env)->lti_lvb;
- __u64 kms;
- int result = 0;
+ struct cl_attr *lov_attr = &r0->lo_attr;
+ int result = 0;
ENTRY;
/* this is called w/o holding type guard mutex, so it must be inside
- * an on going IO otherwise lsm may be replaced. */
- LASSERT(cfs_atomic_read(&lsm->lsm_refc) > 1);
-
- if (!r0->lo_attr_valid) {
- /*
- * Fill LVB with attributes already initialized by the upper
- * layer.
- */
- cl_attr2lvb(lvb, attr);
- kms = attr->cat_kms;
-
- /*
- * XXX that should be replaced with a loop over sub-objects,
- * doing cl_object_attr_get() on them. But for now, let's
- * reuse old lov code.
- */
-
- /*
- * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
- * happy. It's not needed, because new code uses
- * ->coh_attr_guard spin-lock to protect consistency of
- * sub-object attributes.
- */
- lov_stripe_lock(lsm);
- result = lov_merge_lvb_kms(lsm, lvb, &kms);
- lov_stripe_unlock(lsm);
- if (result == 0) {
- cl_lvb2attr(attr, lvb);
- attr->cat_kms = kms;
- r0->lo_attr_valid = 1;
- r0->lo_attr = *attr;
- }
- } else
- *attr = r0->lo_attr;
- RETURN(result);
+ * an on going IO otherwise lsm may be replaced.
+ * LU-2117: it turns out there exists one exception. For mmaped files,
+ * the lock of those files may be requested in the other file's IO
+ * context, and this function is called in ccc_lock_state(), it will
+ * hit this assertion.
+ * Anyway, it's still okay to call attr_get w/o type guard as layout
+ * can't go if locks exist. */
+ /* LASSERT(cfs_atomic_read(&lsm->lsm_refc) > 1); */
+
+ if (!r0->lo_attr_valid) {
+ struct lov_stripe_md *lsm = lov->lo_lsm;
+ struct ost_lvb *lvb = &lov_env_info(env)->lti_lvb;
+ __u64 kms = 0;
+
+ memset(lvb, 0, sizeof(*lvb));
+ /* XXX: timestamps can be negative by sanity:test_39m,
+ * how can it be? */
+ lvb->lvb_atime = LLONG_MIN;
+ lvb->lvb_ctime = LLONG_MIN;
+ lvb->lvb_mtime = LLONG_MIN;
+
+ /*
+ * XXX that should be replaced with a loop over sub-objects,
+ * doing cl_object_attr_get() on them. But for now, let's
+ * reuse old lov code.
+ */
+
+ /*
+ * XXX take lsm spin-lock to keep lov_merge_lvb_kms()
+ * happy. It's not needed, because new code uses
+ * ->coh_attr_guard spin-lock to protect consistency of
+ * sub-object attributes.
+ */
+ lov_stripe_lock(lsm);
+ result = lov_merge_lvb_kms(lsm, lvb, &kms);
+ lov_stripe_unlock(lsm);
+ if (result == 0) {
+ cl_lvb2attr(lov_attr, lvb);
+ lov_attr->cat_kms = kms;
+ r0->lo_attr_valid = 1;
+ }
+ }
+ if (result == 0) { /* merge results */
+ attr->cat_blocks = lov_attr->cat_blocks;
+ attr->cat_size = lov_attr->cat_size;
+ attr->cat_kms = lov_attr->cat_kms;
+ if (attr->cat_atime < lov_attr->cat_atime)
+ attr->cat_atime = lov_attr->cat_atime;
+ if (attr->cat_ctime < lov_attr->cat_ctime)
+ attr->cat_ctime = lov_attr->cat_ctime;
+ if (attr->cat_mtime < lov_attr->cat_mtime)
+ attr->cat_mtime = lov_attr->cat_mtime;
+ }
+ RETURN(result);
}
const static struct lov_layout_operations lov_dispatch[] = {
.llo_install = lov_install_empty,
.llo_print = lov_print_empty,
.llo_page_init = lov_page_init_empty,
- .llo_lock_init = NULL,
+ .llo_lock_init = lov_lock_init_empty,
.llo_io_init = lov_io_init_empty,
.llo_getattr = lov_attr_get_empty
},
static inline void lov_conf_freeze(struct lov_object *lov)
{
if (lov->lo_owner != cfs_current())
- cfs_down_read(&lov->lo_type_guard);
+ down_read(&lov->lo_type_guard);
}
static inline void lov_conf_thaw(struct lov_object *lov)
{
if (lov->lo_owner != cfs_current())
- cfs_up_read(&lov->lo_type_guard);
+ up_read(&lov->lo_type_guard);
}
#define LOV_2DISPATCH_MAYLOCK(obj, op, lock, ...) \
lov_conf_thaw(__obj); \
} while (0)
+static void lov_conf_lock(struct lov_object *lov)
+{
+ LASSERT(lov->lo_owner != cfs_current());
+ down_write(&lov->lo_type_guard);
+ LASSERT(lov->lo_owner == NULL);
+ lov->lo_owner = cfs_current();
+}
+
+static void lov_conf_unlock(struct lov_object *lov)
+{
+ lov->lo_owner = NULL;
+ up_write(&lov->lo_type_guard);
+}
+
static int lov_layout_wait(const struct lu_env *env, struct lov_object *lov)
{
struct l_wait_info lwi = { 0 };
- struct lov_stripe_md *lsm = lov->lo_lsm;
ENTRY;
- if (!lov->lo_lsm_invalid || lsm == NULL)
+ if (!lov->lo_layout_invalid)
RETURN(0);
- l_wait_event(lov->lo_waitq, cfs_atomic_read(&lsm->lsm_refc) == 1, &lwi);
+ while (cfs_atomic_read(&lov->lo_active_ios) > 0) {
+ lov_conf_unlock(lov);
+
+ CDEBUG(D_INODE, "file:"DFID" wait for active IO, now: %d.\n",
+ PFID(lu_object_fid(lov2lu(lov))),
+ cfs_atomic_read(&lov->lo_active_ios));
+
+ l_wait_event(lov->lo_waitq,
+ cfs_atomic_read(&lov->lo_active_ios) == 0, &lwi);
+ lov_conf_lock(lov);
+ }
RETURN(0);
}
-static int lov_layout_change(const struct lu_env *env,
+static int lov_layout_change(const struct lu_env *unused,
struct lov_object *lov, enum lov_layout_type llt,
const struct cl_object_conf *conf)
{
int result;
- union lov_layout_state *state = &lov_env_info(env)->lti_state;
+ union lov_layout_state *state = &lov->u;
const struct lov_layout_operations *old_ops;
const struct lov_layout_operations *new_ops;
struct cl_object_header *hdr = cl_object_header(&lov->lo_cl);
void *cookie;
- struct lu_env *nested;
+ struct lu_env *env;
int refcheck;
LASSERT(0 <= lov->lo_type && lov->lo_type < ARRAY_SIZE(lov_dispatch));
ENTRY;
cookie = cl_env_reenter();
- nested = cl_env_get(&refcheck);
- if (!IS_ERR(nested))
- cl_object_prune(nested, &lov->lo_cl);
- else
- result = PTR_ERR(nested);
- cl_env_put(nested, &refcheck);
- cl_env_reexit(cookie);
+ env = cl_env_get(&refcheck);
+ if (IS_ERR(env)) {
+ cl_env_reexit(cookie);
+ RETURN(PTR_ERR(env));
+ }
old_ops = &lov_dispatch[lov->lo_type];
new_ops = &lov_dispatch[llt];
result = old_ops->llo_delete(env, lov, &lov->u);
if (result == 0) {
old_ops->llo_fini(env, lov, &lov->u);
+
+ LASSERT(cfs_atomic_read(&lov->lo_active_ios) == 0);
LASSERT(cfs_list_empty(&hdr->coh_locks));
LASSERT(hdr->coh_tree.rnode == NULL);
LASSERT(hdr->coh_pages == 0);
+ lov->lo_type = LLT_EMPTY;
result = new_ops->llo_init(env,
lu2lov_dev(lov->lo_cl.co_lu.lo_dev),
lov, conf, state);
if (result == 0) {
new_ops->llo_install(env, lov, state);
lov->lo_type = llt;
+ } else {
+ new_ops->llo_delete(env, lov, state);
+ new_ops->llo_fini(env, lov, state);
+ /* this file becomes an EMPTY file. */
}
}
+
+ cl_env_put(env, &refcheck);
+ cl_env_reexit(cookie);
RETURN(result);
}
struct lov_device *dev = lu2lov_dev(obj->lo_dev);
struct lov_object *lov = lu2lov(obj);
const struct cl_object_conf *cconf = lu2cl_conf(conf);
- union lov_layout_state *set = &lov_env_info(env)->lti_state;
+ union lov_layout_state *set = &lov->u;
const struct lov_layout_operations *ops;
int result;
ENTRY;
- cfs_init_rwsem(&lov->lo_type_guard);
+ init_rwsem(&lov->lo_type_guard);
+ cfs_atomic_set(&lov->lo_active_ios, 0);
cfs_waitq_init(&lov->lo_waitq);
+ cl_object_page_init(lu2cl(obj), sizeof(struct lov_page));
+
/* no locking is necessary, as object is being created */
lov->lo_type = cconf->u.coc_md->lsm != NULL ? LLT_RAID0 : LLT_EMPTY;
ops = &lov_dispatch[lov->lo_type];
result = ops->llo_init(env, dev, lov, cconf, set);
if (result == 0)
ops->llo_install(env, lov, set);
- else
- ops->llo_fini(env, lov, set);
RETURN(result);
}
static int lov_conf_set(const struct lu_env *env, struct cl_object *obj,
const struct cl_object_conf *conf)
{
- struct lov_stripe_md *lsm = conf->u.coc_md->lsm;
+ struct lov_stripe_md *lsm = NULL;
struct lov_object *lov = cl2lov(obj);
int result = 0;
ENTRY;
- /*
- * Only LLT_EMPTY <-> LLT_RAID0 transitions are supported.
- */
- LASSERT(lov->lo_owner != cfs_current());
- cfs_down_write(&lov->lo_type_guard);
- LASSERT(lov->lo_owner == NULL);
- lov->lo_owner = cfs_current();
-
- if (conf->coc_invalidate) {
- lov->lo_lsm_invalid = 1;
+ lov_conf_lock(lov);
+ if (conf->coc_opc == OBJECT_CONF_INVALIDATE) {
+ lov->lo_layout_invalid = true;
GOTO(out, result = 0);
}
- if (conf->coc_validate_only) {
- if (!lov->lo_lsm_invalid)
- GOTO(out, result = 0);
+ if (conf->coc_opc == OBJECT_CONF_WAIT) {
+ result = lov_layout_wait(env, lov);
+ GOTO(out, result);
+ }
+
+ LASSERT(conf->coc_opc == OBJECT_CONF_SET);
- lov_layout_wait(env, lov);
- /* fall through to set up new layout */
+ if (conf->u.coc_md != NULL)
+ lsm = conf->u.coc_md->lsm;
+ if ((lsm == NULL && lov->lo_lsm == NULL) ||
+ (lsm != NULL && lov->lo_lsm != NULL &&
+ lov->lo_lsm->lsm_layout_gen == lsm->lsm_layout_gen)) {
+ /* same version of layout */
+ lov->lo_layout_invalid = false;
+ GOTO(out, result = 0);
+ }
+
+ /* will change layout - check if there still exists active IO. */
+ if (cfs_atomic_read(&lov->lo_active_ios) > 1) {
+ lov->lo_layout_invalid = true;
+ GOTO(out, result = -EBUSY);
}
+ /*
+ * Only LLT_EMPTY <-> LLT_RAID0 transitions are supported.
+ */
switch (lov->lo_type) {
case LLT_EMPTY:
if (lsm != NULL)
default:
LBUG();
}
- lov->lo_lsm_invalid = result != 0;
+ lov->lo_layout_invalid = result != 0;
EXIT;
out:
- lov->lo_owner = NULL;
- cfs_up_write(&lov->lo_type_guard);
+ lov_conf_unlock(lov);
RETURN(result);
}
return LOV_2DISPATCH(lu2lov(o), llo_print, env, cookie, p, o);
}
-struct cl_page *lov_page_init(const struct lu_env *env, struct cl_object *obj,
- struct cl_page *page, cfs_page_t *vmpage)
+int lov_page_init(const struct lu_env *env, struct cl_object *obj,
+ struct cl_page *page, cfs_page_t *vmpage)
{
- return LOV_2DISPATCH(cl2lov(obj),
- llo_page_init, env, obj, page, vmpage);
+ return LOV_2DISPATCH_NOLOCK(cl2lov(obj),
+ llo_page_init, env, obj, page, vmpage);
}
/**
* layer. Dispatches to the appropriate layout io initialization method.
*/
int lov_io_init(const struct lu_env *env, struct cl_object *obj,
- struct cl_io *io)
+ struct cl_io *io)
{
- struct lov_io *lio = lov_env_io(env);
-
- CL_IO_SLICE_CLEAN(lov_env_io(env), lis_cl);
-
- /* hold lsm before initializing because io relies on it */
- lio->lis_lsm = lov_lsm_addref(cl2lov(obj));
-
- /*
- * Do not take lock in case of CIT_MISC io, because
- *
- * - if this is an io for a glimpse, then we don't care;
- *
- * - if this not a glimpse (writepage or lock cancellation), then
- * layout change cannot happen because a page or a lock
- * already exist; and
- *
- * - lock ordering (lock mutex nests within layout rw-semaphore)
- * is obeyed in case of lock cancellation.
- */
- return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
- io->ci_type != CIT_MISC, env, obj, io);
+ CL_IO_SLICE_CLEAN(lov_env_io(env), lis_cl);
+ return LOV_2DISPATCH_MAYLOCK(cl2lov(obj), llo_io_init,
+ !io->ci_ignore_layout, env, obj, io);
}
/**
}
int lov_lock_init(const struct lu_env *env, struct cl_object *obj,
- struct cl_lock *lock, const struct cl_io *io)
+ struct cl_lock *lock, const struct cl_io *io)
{
- return LOV_2DISPATCH(cl2lov(obj), llo_lock_init, env, obj, lock, io);
+ /* No need to lock because we've taken one refcount of layout. */
+ return LOV_2DISPATCH_NOLOCK(cl2lov(obj), llo_lock_init, env, obj, lock,
+ io);
}
static const struct cl_object_operations lov_ops = {
struct lov_stripe_md *lsm = NULL;
lov_conf_freeze(lov);
- if (!lov->lo_lsm_invalid && lov->lo_lsm != NULL) {
+ if (lov->lo_lsm != NULL) {
lsm = lsm_addref(lov->lo_lsm);
- CDEBUG(D_INODE, "lsm %p addref %d by %p.\n",
- lsm, cfs_atomic_read(&lsm->lsm_refc), cfs_current());
+ CDEBUG(D_INODE, "lsm %p addref %d/%d by %p.\n",
+ lsm, cfs_atomic_read(&lsm->lsm_refc),
+ lov->lo_layout_invalid, cfs_current());
}
lov_conf_thaw(lov);
return lsm;
CDEBUG(D_INODE, "lsm %p decref %d by %p.\n",
lsm, cfs_atomic_read(&lsm->lsm_refc), cfs_current());
- if (lov_free_memmd(&lsm) <= 1 && lov->lo_lsm_invalid)
- cfs_waitq_signal(&lov->lo_waitq);
+ lov_free_memmd(&lsm);
}
struct lov_stripe_md *lov_lsm_get(struct cl_object *clobj)
}
EXPORT_SYMBOL(lov_lsm_get);
-void lov_lsm_put(struct cl_object *clobj, struct lov_stripe_md *lsm)
+void lov_lsm_put(struct cl_object *unused, struct lov_stripe_md *lsm)
{
- struct lu_object *luobj;
-
- if (clobj == NULL || lsm == NULL)
- return;
-
- luobj = lu_object_locate(&cl_object_header(clobj)->coh_lu,
- &lov_device_type);
- LASSERT(luobj != NULL);
-
- lov_lsm_decref(lu2lov(luobj), lsm);
+ if (lsm != NULL)
+ lov_free_memmd(&lsm);
}
EXPORT_SYMBOL(lov_lsm_put);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff