/* link into sai hash table locally */
cfs_list_t se_hash;
/* entry reference count */
- cfs_atomic_t se_refcount;
+ atomic_t se_refcount;
/* entry index in the sai */
__u64 se_index;
/* low layer ldlm lock handle */
static inline int sa_sent_full(struct ll_statahead_info *sai)
{
- return cfs_atomic_read(&sai->sai_cache_count) >= sai->sai_max;
+ return atomic_read(&sai->sai_cache_count) >= sai->sai_max;
}
static inline int sa_received_empty(struct ll_statahead_info *sai)
entry->se_index = index;
- /*
- * Statahead entry reference rules:
- *
- * 1) When statahead entry is initialized, its reference is set as 2.
- * One reference is used by the directory scanner. When the scanner
- * searches the statahead cache for the given name, it can perform
- * lockless hash lookup (only the scanner can remove entry from hash
- * list), and once found, it needn't to call "atomic_inc()" for the
- * entry reference. So the performance is improved. After using the
- * statahead entry, the scanner will call "atomic_dec()" to drop the
- * reference held when initialization. If it is the last reference,
- * the statahead entry will be freed.
- *
- * 2) All other threads, including statahead thread and ptlrpcd thread,
- * when they process the statahead entry, the reference for target
- * should be held to guarantee the entry will not be released by the
- * directory scanner. After processing the entry, these threads will
- * drop the entry reference. If it is the last reference, the entry
- * will be freed.
- *
- * The second reference when initializes the statahead entry is used
- * by the statahead thread, following the rule 2).
- */
- cfs_atomic_set(&entry->se_refcount, 2);
- entry->se_stat = SA_ENTRY_INIT;
- entry->se_size = entry_size;
- dname = (char *)entry + sizeof(struct ll_sa_entry);
- memcpy(dname, name, len);
- dname[len] = 0;
- entry->se_qstr.hash = full_name_hash(name, len);
- entry->se_qstr.len = len;
- entry->se_qstr.name = dname;
-
- lli = ll_i2info(sai->sai_inode);
+ /*
+ * Statahead entry reference rules:
+ *
+ * 1) When statahead entry is initialized, its reference is set as 2.
+ * One reference is used by the directory scanner. When the scanner
+ * searches the statahead cache for the given name, it can perform
+ * lockless hash lookup (only the scanner can remove entry from hash
+ * list), and once found, it needn't to call "atomic_inc()" for the
+ * entry reference. So the performance is improved. After using the
+ * statahead entry, the scanner will call "atomic_dec()" to drop the
+ * reference held when initialization. If it is the last reference,
+ * the statahead entry will be freed.
+ *
+ * 2) All other threads, including statahead thread and ptlrpcd thread,
+ * when they process the statahead entry, the reference for target
+ * should be held to guarantee the entry will not be released by the
+ * directory scanner. After processing the entry, these threads will
+ * drop the entry reference. If it is the last reference, the entry
+ * will be freed.
+ *
+ * The second reference when initializes the statahead entry is used
+ * by the statahead thread, following the rule 2).
+ */
+ atomic_set(&entry->se_refcount, 2);
+ entry->se_stat = SA_ENTRY_INIT;
+ entry->se_size = entry_size;
+ dname = (char *)entry + sizeof(struct ll_sa_entry);
+ memcpy(dname, name, len);
+ dname[len] = 0;
+ entry->se_qstr.hash = full_name_hash(name, len);
+ entry->se_qstr.len = len;
+ entry->se_qstr.name = dname;
+
+ lli = ll_i2info(sai->sai_inode);
spin_lock(&lli->lli_sa_lock);
cfs_list_add_tail(&entry->se_link, &sai->sai_entries);
CFS_INIT_LIST_HEAD(&entry->se_list);
ll_sa_entry_enhash(sai, entry);
spin_unlock(&lli->lli_sa_lock);
- cfs_atomic_inc(&sai->sai_cache_count);
+ atomic_inc(&sai->sai_cache_count);
RETURN(entry);
}
cfs_list_for_each_entry(entry, &sai->sai_entries, se_link) {
if (entry->se_index == index) {
LASSERT(atomic_read(&entry->se_refcount) > 0);
- cfs_atomic_inc(&entry->se_refcount);
+ atomic_inc(&entry->se_refcount);
return entry;
}
if (entry->se_index > index)
static void ll_sa_entry_put(struct ll_statahead_info *sai,
struct ll_sa_entry *entry)
{
- if (cfs_atomic_dec_and_test(&entry->se_refcount)) {
+ if (atomic_dec_and_test(&entry->se_refcount)) {
CDEBUG(D_READA, "free sa entry %.*s(%p) index "LPU64"\n",
entry->se_qstr.len, entry->se_qstr.name, entry,
entry->se_index);
iput(entry->se_inode);
OBD_FREE(entry, entry->se_size);
- cfs_atomic_dec(&sai->sai_cache_count);
+ atomic_dec(&sai->sai_cache_count);
}
}
}
if (added > 0)
- cfs_waitq_signal(&sai->sai_agl_thread.t_ctl_waitq);
+ wake_up(&sai->sai_agl_thread.t_ctl_waitq);
}
static struct ll_statahead_info *ll_sai_alloc(void)
{
- struct ll_statahead_info *sai;
- int i;
- ENTRY;
+ struct ll_statahead_info *sai;
+ int i;
+ ENTRY;
- OBD_ALLOC_PTR(sai);
- if (!sai)
- RETURN(NULL);
+ OBD_ALLOC_PTR(sai);
+ if (!sai)
+ RETURN(NULL);
- cfs_atomic_set(&sai->sai_refcount, 1);
+ atomic_set(&sai->sai_refcount, 1);
spin_lock(&sai_generation_lock);
sai->sai_generation = ++sai_generation;
sai->sai_generation = ++sai_generation;
spin_unlock(&sai_generation_lock);
- sai->sai_max = LL_SA_RPC_MIN;
- sai->sai_index = 1;
- cfs_waitq_init(&sai->sai_waitq);
- cfs_waitq_init(&sai->sai_thread.t_ctl_waitq);
- cfs_waitq_init(&sai->sai_agl_thread.t_ctl_waitq);
+ sai->sai_max = LL_SA_RPC_MIN;
+ sai->sai_index = 1;
+ init_waitqueue_head(&sai->sai_waitq);
+ init_waitqueue_head(&sai->sai_thread.t_ctl_waitq);
+ init_waitqueue_head(&sai->sai_agl_thread.t_ctl_waitq);
CFS_INIT_LIST_HEAD(&sai->sai_entries);
- CFS_INIT_LIST_HEAD(&sai->sai_entries_received);
- CFS_INIT_LIST_HEAD(&sai->sai_entries_stated);
- CFS_INIT_LIST_HEAD(&sai->sai_entries_agl);
+ CFS_INIT_LIST_HEAD(&sai->sai_entries_received);
+ CFS_INIT_LIST_HEAD(&sai->sai_entries_stated);
+ CFS_INIT_LIST_HEAD(&sai->sai_entries_agl);
- for (i = 0; i < LL_SA_CACHE_SIZE; i++) {
- CFS_INIT_LIST_HEAD(&sai->sai_cache[i]);
+ for (i = 0; i < LL_SA_CACHE_SIZE; i++) {
+ CFS_INIT_LIST_HEAD(&sai->sai_cache[i]);
spin_lock_init(&sai->sai_cache_lock[i]);
- }
- cfs_atomic_set(&sai->sai_cache_count, 0);
+ }
+ atomic_set(&sai->sai_cache_count, 0);
- RETURN(sai);
+ RETURN(sai);
}
static inline struct ll_statahead_info *
ll_sai_get(struct ll_statahead_info *sai)
{
- cfs_atomic_inc(&sai->sai_refcount);
- return sai;
+ atomic_inc(&sai->sai_refcount);
+ return sai;
}
static void ll_sai_put(struct ll_statahead_info *sai)
{
- struct inode *inode = sai->sai_inode;
- struct ll_inode_info *lli = ll_i2info(inode);
- ENTRY;
+ struct inode *inode = sai->sai_inode;
+ struct ll_inode_info *lli = ll_i2info(inode);
+ ENTRY;
- if (cfs_atomic_dec_and_lock(&sai->sai_refcount, &lli->lli_sa_lock)) {
- struct ll_sa_entry *entry, *next;
+ if (atomic_dec_and_lock(&sai->sai_refcount, &lli->lli_sa_lock)) {
+ struct ll_sa_entry *entry, *next;
- if (unlikely(cfs_atomic_read(&sai->sai_refcount) > 0)) {
- /* It is race case, the interpret callback just hold
- * a reference count */
+ if (unlikely(atomic_read(&sai->sai_refcount) > 0)) {
+ /* It is race case, the interpret callback just hold
+ * a reference count */
spin_unlock(&lli->lli_sa_lock);
RETURN_EXIT;
}
LASSERT(sa_received_empty(sai));
LASSERT(list_empty(&sai->sai_entries_stated));
- LASSERT(cfs_atomic_read(&sai->sai_cache_count) == 0);
- LASSERT(agl_list_empty(sai));
+ LASSERT(atomic_read(&sai->sai_cache_count) == 0);
+ LASSERT(agl_list_empty(sai));
- iput(inode);
- OBD_FREE_PTR(sai);
- }
+ iput(inode);
+ OBD_FREE_PTR(sai);
+ }
- EXIT;
+ EXIT;
}
/* Do NOT forget to drop inode refcount when into sai_entries_agl. */
RETURN_EXIT;
}
entry = sa_first_received_entry(sai);
- cfs_atomic_inc(&entry->se_refcount);
+ atomic_inc(&entry->se_refcount);
cfs_list_del_init(&entry->se_list);
spin_unlock(&lli->lli_sa_lock);
if (rc)
GOTO(out, rc);
- CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
- child, child->i_ino, child->i_generation);
+ CDEBUG(D_DLMTRACE, "%s: setting l_data to inode "DFID"(%p)\n",
+ ll_get_fsname(child->i_sb, NULL, 0),
+ PFID(ll_inode2fid(child)), child);
ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, child, it, NULL);
entry->se_inode = child;
EXIT;
out:
- /* The "ll_sa_entry_to_stated()" will drop related ldlm ibits lock
- * reference count by calling "ll_intent_drop_lock()" in spite of the
- * above operations failed or not. Do not worry about calling
- * "ll_intent_drop_lock()" more than once. */
+ /* The "ll_sa_entry_to_stated()" will drop related ldlm ibits lock
+ * reference count by calling "ll_intent_drop_lock()" in spite of the
+ * above operations failed or not. Do not worry about calling
+ * "ll_intent_drop_lock()" more than once. */
rc = ll_sa_entry_to_stated(sai, entry,
rc < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC);
if (rc == 0 && entry->se_index == sai->sai_index_wait)
- cfs_waitq_signal(&sai->sai_waitq);
- ll_sa_entry_put(sai, entry);
+ wake_up(&sai->sai_waitq);
+ ll_sa_entry_put(sai, entry);
}
static int ll_statahead_interpret(struct ptlrpc_request *req,
ll_sa_entry_put(sai, entry);
if (wakeup)
- cfs_waitq_signal(&sai->sai_thread.t_ctl_waitq);
+ wake_up(&sai->sai_thread.t_ctl_waitq);
}
EXIT;
if (dentry != NULL)
dput(dentry);
- if (rc) {
- rc1 = ll_sa_entry_to_stated(sai, entry,
- rc < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC);
- if (rc1 == 0 && entry->se_index == sai->sai_index_wait)
- cfs_waitq_signal(&sai->sai_waitq);
- } else {
- sai->sai_sent++;
- }
+ if (rc) {
+ rc1 = ll_sa_entry_to_stated(sai, entry,
+ rc < 0 ? SA_ENTRY_INVA : SA_ENTRY_SUCC);
+ if (rc1 == 0 && entry->se_index == sai->sai_index_wait)
+ wake_up(&sai->sai_waitq);
+ } else {
+ sai->sai_sent++;
+ }
sai->sai_index++;
/* drop one refcount on entry by ll_sa_entry_alloc */
struct l_wait_info lwi = { 0 };
ENTRY;
- CDEBUG(D_READA, "agl thread started: [pid %d] [parent %.*s]\n",
- cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
+ thread->t_pid = current_pid();
+ CDEBUG(D_READA, "agl thread started: sai %p, parent %.*s\n",
+ sai, parent->d_name.len, parent->d_name.name);
- atomic_inc(&sbi->ll_agl_total);
+ atomic_inc(&sbi->ll_agl_total);
spin_lock(&plli->lli_agl_lock);
sai->sai_agl_valid = 1;
- thread_set_flags(thread, SVC_RUNNING);
+ if (thread_is_init(thread))
+ /* If someone else has changed the thread state
+ * (e.g. already changed to SVC_STOPPING), we can't just
+ * blindly overwrite that setting. */
+ thread_set_flags(thread, SVC_RUNNING);
spin_unlock(&plli->lli_agl_lock);
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&thread->t_ctl_waitq);
while (1) {
l_wait_event(thread->t_ctl_waitq,
}
thread_set_flags(thread, SVC_STOPPED);
spin_unlock(&plli->lli_agl_lock);
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&thread->t_ctl_waitq);
ll_sai_put(sai);
- CDEBUG(D_READA, "agl thread stopped: [pid %d] [parent %.*s]\n",
- cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
+ CDEBUG(D_READA, "agl thread stopped: sai %p, parent %.*s\n",
+ sai, parent->d_name.len, parent->d_name.name);
RETURN(0);
}
struct ptlrpc_thread *thread = &sai->sai_agl_thread;
struct l_wait_info lwi = { 0 };
struct ll_inode_info *plli;
- cfs_task_t *task;
+ struct task_struct *task;
ENTRY;
- CDEBUG(D_READA, "start agl thread: [pid %d] [parent %.*s]\n",
- cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
+ CDEBUG(D_READA, "start agl thread: sai %p, parent %.*s\n",
+ sai, parent->d_name.len, parent->d_name.name);
plli = ll_i2info(parent->d_inode);
task = kthread_run(ll_agl_thread, parent,
static int ll_statahead_thread(void *arg)
{
- struct dentry *parent = (struct dentry *)arg;
- struct inode *dir = parent->d_inode;
- struct ll_inode_info *plli = ll_i2info(dir);
- struct ll_inode_info *clli;
- struct ll_sb_info *sbi = ll_i2sbi(dir);
- struct ll_statahead_info *sai = ll_sai_get(plli->lli_sai);
- struct ptlrpc_thread *thread = &sai->sai_thread;
- struct ptlrpc_thread *agl_thread = &sai->sai_agl_thread;
- struct page *page;
- __u64 pos = 0;
- int first = 0;
- int rc = 0;
- struct ll_dir_chain chain;
- struct l_wait_info lwi = { 0 };
- ENTRY;
+ struct dentry *parent = (struct dentry *)arg;
+ struct inode *dir = parent->d_inode;
+ struct ll_inode_info *plli = ll_i2info(dir);
+ struct ll_inode_info *clli;
+ struct ll_sb_info *sbi = ll_i2sbi(dir);
+ struct ll_statahead_info *sai = ll_sai_get(plli->lli_sai);
+ struct ptlrpc_thread *thread = &sai->sai_thread;
+ struct ptlrpc_thread *agl_thread = &sai->sai_agl_thread;
+ int first = 0;
+ int rc = 0;
+ struct md_op_data *op_data;
+ struct ll_dir_chain chain;
+ struct l_wait_info lwi = { 0 };
+ struct lu_dirent *ent;
+ struct page *page = NULL;
+ ENTRY;
- CDEBUG(D_READA, "statahead thread started: [pid %d] [parent %.*s]\n",
- cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
+ thread->t_pid = current_pid();
+ CDEBUG(D_READA, "statahead thread starting: sai %p, parent %.*s\n",
+ sai, parent->d_name.len, parent->d_name.name);
- if (sbi->ll_flags & LL_SBI_AGL_ENABLED)
- ll_start_agl(parent, sai);
+ op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
+ LUSTRE_OPC_ANY, dir);
+ if (IS_ERR(op_data))
+ RETURN(PTR_ERR(op_data));
- atomic_inc(&sbi->ll_sa_total);
+ op_data->op_hash_offset = 0;
+ op_data->op_max_pages =
+ ll_i2sbi(dir)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
+
+ if (sbi->ll_flags & LL_SBI_AGL_ENABLED)
+ ll_start_agl(parent, sai);
+
+ atomic_inc(&sbi->ll_sa_total);
spin_lock(&plli->lli_sa_lock);
- thread_set_flags(thread, SVC_RUNNING);
+ if (thread_is_init(thread))
+ /* If someone else has changed the thread state
+ * (e.g. already changed to SVC_STOPPING), we can't just
+ * blindly overwrite that setting. */
+ thread_set_flags(thread, SVC_RUNNING);
spin_unlock(&plli->lli_sa_lock);
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&thread->t_ctl_waitq);
ll_dir_chain_init(&chain);
- page = ll_get_dir_page(dir, pos, &chain);
-
- while (1) {
- struct lu_dirpage *dp;
- struct lu_dirent *ent;
-
- if (IS_ERR(page)) {
- rc = PTR_ERR(page);
- CDEBUG(D_READA, "error reading dir "DFID" at "LPU64
- "/"LPU64": [rc %d] [parent %u]\n",
- PFID(ll_inode2fid(dir)), pos, sai->sai_index,
- rc, plli->lli_opendir_pid);
- GOTO(out, rc);
- }
-
- dp = page_address(page);
- for (ent = lu_dirent_start(dp); ent != NULL;
- ent = lu_dirent_next(ent)) {
- __u64 hash;
- int namelen;
- char *name;
-
- hash = le64_to_cpu(ent->lde_hash);
- if (unlikely(hash < pos))
- /*
- * Skip until we find target hash value.
- */
- continue;
-
- namelen = le16_to_cpu(ent->lde_namelen);
- if (unlikely(namelen == 0))
- /*
- * Skip dummy record.
- */
- continue;
-
- name = ent->lde_name;
- if (name[0] == '.') {
- if (namelen == 1) {
- /*
- * skip "."
- */
- continue;
- } else if (name[1] == '.' && namelen == 2) {
- /*
- * skip ".."
- */
- continue;
- } else if (!sai->sai_ls_all) {
- /*
- * skip hidden files.
- */
- sai->sai_skip_hidden++;
- continue;
- }
- }
+ for (ent = ll_dir_entry_start(dir, op_data, &page);
+ ent != NULL && !IS_ERR(ent);
+ ent = ll_dir_entry_next(dir, op_data, ent, &page)) {
+ __u64 hash;
+ int namelen;
+ char *name;
+
+ hash = le64_to_cpu(ent->lde_hash);
+ if (unlikely(hash < op_data->op_hash_offset))
+ /*
+ * Skip until we find target hash value.
+ */
+ continue;
+
+ namelen = le16_to_cpu(ent->lde_namelen);
+ if (unlikely(namelen == 0))
+ /*
+ * Skip dummy record.
+ */
+ continue;
+
+ name = ent->lde_name;
+ if (name[0] == '.') {
+ if (namelen == 1) {
+ /*
+ * skip "."
+ */
+ continue;
+ } else if (name[1] == '.' && namelen == 2) {
+ /*
+ * skip ".."
+ */
+ continue;
+ } else if (!sai->sai_ls_all) {
+ /*
+ * skip hidden files.
+ */
+ sai->sai_skip_hidden++;
+ continue;
+ }
+ }
- /*
- * don't stat-ahead first entry.
- */
- if (unlikely(++first == 1))
- continue;
+ /*
+ * don't stat-ahead first entry.
+ */
+ if (unlikely(++first == 1))
+ continue;
keep_it:
- l_wait_event(thread->t_ctl_waitq,
- !sa_sent_full(sai) ||
- !sa_received_empty(sai) ||
- !agl_list_empty(sai) ||
- !thread_is_running(thread),
- &lwi);
+ l_wait_event(thread->t_ctl_waitq,
+ !sa_sent_full(sai) ||
+ !sa_received_empty(sai) ||
+ !agl_list_empty(sai) ||
+ !thread_is_running(thread),
+ &lwi);
interpret_it:
- while (!sa_received_empty(sai))
- ll_post_statahead(sai);
-
- if (unlikely(!thread_is_running(thread))) {
- ll_release_page(page, 0);
- GOTO(out, rc = 0);
- }
-
- /* If no window for metadata statahead, but there are
- * some AGL entries to be triggered, then try to help
- * to process the AGL entries. */
- if (sa_sent_full(sai)) {
- spin_lock(&plli->lli_agl_lock);
- while (!agl_list_empty(sai)) {
- clli = agl_first_entry(sai);
- cfs_list_del_init(&clli->lli_agl_list);
- spin_unlock(&plli->lli_agl_lock);
- ll_agl_trigger(&clli->lli_vfs_inode,
- sai);
-
- if (!sa_received_empty(sai))
- goto interpret_it;
-
- if (unlikely(
- !thread_is_running(thread))) {
- ll_release_page(page, 0);
- GOTO(out, rc = 0);
- }
-
- if (!sa_sent_full(sai))
- goto do_it;
-
- spin_lock(&plli->lli_agl_lock);
- }
- spin_unlock(&plli->lli_agl_lock);
-
- goto keep_it;
- }
+ while (!sa_received_empty(sai))
+ ll_post_statahead(sai);
-do_it:
- ll_statahead_one(parent, name, namelen);
- }
- pos = le64_to_cpu(dp->ldp_hash_end);
- if (pos == MDS_DIR_END_OFF) {
- /*
- * End of directory reached.
- */
- ll_release_page(page, 0);
- while (1) {
- l_wait_event(thread->t_ctl_waitq,
- !sa_received_empty(sai) ||
- sai->sai_sent == sai->sai_replied||
- !thread_is_running(thread),
- &lwi);
-
- while (!sa_received_empty(sai))
- ll_post_statahead(sai);
-
- if (unlikely(!thread_is_running(thread)))
- GOTO(out, rc = 0);
-
- if (sai->sai_sent == sai->sai_replied &&
- sa_received_empty(sai))
- break;
- }
+ if (unlikely(!thread_is_running(thread)))
+ GOTO(out, rc = 0);
+ /* If no window for metadata statahead, but there are
+ * some AGL entries to be triggered, then try to help
+ * to process the AGL entries. */
+ if (sa_sent_full(sai)) {
spin_lock(&plli->lli_agl_lock);
- while (!agl_list_empty(sai) &&
- thread_is_running(thread)) {
+ while (!agl_list_empty(sai)) {
clli = agl_first_entry(sai);
cfs_list_del_init(&clli->lli_agl_list);
spin_unlock(&plli->lli_agl_lock);
- ll_agl_trigger(&clli->lli_vfs_inode, sai);
+ ll_agl_trigger(&clli->lli_vfs_inode,
+ sai);
+
+ if (!sa_received_empty(sai))
+ goto interpret_it;
+
+ if (unlikely(
+ !thread_is_running(thread)))
+ GOTO(out, rc = 0);
+
+ if (!sa_sent_full(sai))
+ goto do_it;
+
spin_lock(&plli->lli_agl_lock);
}
spin_unlock(&plli->lli_agl_lock);
- GOTO(out, rc = 0);
- } else if (1) {
- /*
- * chain is exhausted.
- * Normal case: continue to the next page.
- */
- ll_release_page(page, le32_to_cpu(dp->ldp_flags) &
- LDF_COLLIDE);
- sai->sai_in_readpage = 1;
- page = ll_get_dir_page(dir, pos, &chain);
- sai->sai_in_readpage = 0;
- } else {
- LASSERT(le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
- ll_release_page(page, 1);
- /*
- * go into overflow page.
- */
- }
- }
- EXIT;
+ goto keep_it;
+ }
+
+do_it:
+ ll_statahead_one(parent, name, namelen);
+ }
+
+ if (page != NULL) {
+ kunmap(page);
+ page_cache_release(page);
+ }
+ /*
+ * End of directory reached.
+ */
+ while (1) {
+ l_wait_event(thread->t_ctl_waitq,
+ !sa_received_empty(sai) ||
+ sai->sai_sent == sai->sai_replied ||
+ !thread_is_running(thread),
+ &lwi);
+
+ while (!sa_received_empty(sai))
+ ll_post_statahead(sai);
+
+ if (unlikely(!thread_is_running(thread)))
+ GOTO(out, rc = 0);
+
+ if (sai->sai_sent == sai->sai_replied &&
+ sa_received_empty(sai))
+ break;
+ }
+
+ spin_lock(&plli->lli_agl_lock);
+ while (!agl_list_empty(sai) &&
+ thread_is_running(thread)) {
+ clli = agl_first_entry(sai);
+ cfs_list_del_init(&clli->lli_agl_list);
+ spin_unlock(&plli->lli_agl_lock);
+ ll_agl_trigger(&clli->lli_vfs_inode, sai);
+ spin_lock(&plli->lli_agl_lock);
+ }
+ spin_unlock(&plli->lli_agl_lock);
out:
+ EXIT;
+ ll_finish_md_op_data(op_data);
if (sai->sai_agl_valid) {
spin_lock(&plli->lli_agl_lock);
thread_set_flags(agl_thread, SVC_STOPPING);
spin_unlock(&plli->lli_agl_lock);
- cfs_waitq_signal(&agl_thread->t_ctl_waitq);
+ wake_up(&agl_thread->t_ctl_waitq);
- CDEBUG(D_READA, "stop agl thread: [pid %d]\n",
- cfs_curproc_pid());
- l_wait_event(agl_thread->t_ctl_waitq,
- thread_is_stopped(agl_thread),
- &lwi);
+ CDEBUG(D_READA, "stop agl thread: sai %p pid %u\n",
+ sai, (unsigned int)agl_thread->t_pid);
+ l_wait_event(agl_thread->t_ctl_waitq,
+ thread_is_stopped(agl_thread),
+ &lwi);
} else {
/* Set agl_thread flags anyway. */
thread_set_flags(&sai->sai_agl_thread, SVC_STOPPED);
}
thread_set_flags(thread, SVC_STOPPED);
spin_unlock(&plli->lli_sa_lock);
- cfs_waitq_signal(&sai->sai_waitq);
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&sai->sai_waitq);
+ wake_up(&thread->t_ctl_waitq);
ll_sai_put(sai);
dput(parent);
- CDEBUG(D_READA, "statahead thread stopped: [pid %d] [parent %.*s]\n",
- cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
- return rc;
+ CDEBUG(D_READA, "statahead thread stopped: sai %p, parent %.*s\n",
+ sai, parent->d_name.len, parent->d_name.name);
+ return rc;
}
/**
if (!thread_is_stopped(thread)) {
thread_set_flags(thread, SVC_STOPPING);
spin_unlock(&lli->lli_sa_lock);
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&thread->t_ctl_waitq);
- CDEBUG(D_READA, "stop statahead thread: [pid %d]\n",
- cfs_curproc_pid());
+ CDEBUG(D_READA, "stop statahead thread: sai %p pid %u\n",
+ lli->lli_sai, (unsigned int)thread->t_pid);
l_wait_event(thread->t_ctl_waitq,
thread_is_stopped(thread),
&lwi);
static int is_first_dirent(struct inode *dir, struct dentry *dentry)
{
- struct ll_dir_chain chain;
- struct qstr *target = &dentry->d_name;
- struct page *page;
- __u64 pos = 0;
- int dot_de;
- int rc = LS_NONE_FIRST_DE;
- ENTRY;
-
- ll_dir_chain_init(&chain);
- page = ll_get_dir_page(dir, pos, &chain);
-
- while (1) {
- struct lu_dirpage *dp;
- struct lu_dirent *ent;
-
- if (IS_ERR(page)) {
- struct ll_inode_info *lli = ll_i2info(dir);
-
- rc = PTR_ERR(page);
- CERROR("error reading dir "DFID" at "LPU64": "
- "[rc %d] [parent %u]\n",
- PFID(ll_inode2fid(dir)), pos,
- rc, lli->lli_opendir_pid);
- break;
- }
+ struct ll_dir_chain chain;
+ struct qstr *target = &dentry->d_name;
+ struct md_op_data *op_data;
+ int dot_de;
+ struct lu_dirent *ent;
+ struct page *page = NULL;
+ int rc = LS_NONE_FIRST_DE;
+ ENTRY;
- dp = page_address(page);
- for (ent = lu_dirent_start(dp); ent != NULL;
- ent = lu_dirent_next(ent)) {
- __u64 hash;
- int namelen;
- char *name;
-
- hash = le64_to_cpu(ent->lde_hash);
- /* The ll_get_dir_page() can return any page containing
- * the given hash which may be not the start hash. */
- if (unlikely(hash < pos))
- continue;
-
- namelen = le16_to_cpu(ent->lde_namelen);
- if (unlikely(namelen == 0))
- /*
- * skip dummy record.
- */
- continue;
+ ll_dir_chain_init(&chain);
- name = ent->lde_name;
- if (name[0] == '.') {
- if (namelen == 1)
- /*
- * skip "."
- */
- continue;
- else if (name[1] == '.' && namelen == 2)
- /*
- * skip ".."
- */
- continue;
- else
- dot_de = 1;
- } else {
- dot_de = 0;
- }
+ op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
+ LUSTRE_OPC_ANY, dir);
+ if (IS_ERR(op_data))
+ GOTO(out, rc = PTR_ERR(op_data));
+ /**
+ *FIXME choose the start offset of the readdir
+ */
+ op_data->op_stripe_offset = 0;
+ op_data->op_hash_offset = 0;
+ op_data->op_max_pages =
+ ll_i2sbi(dir)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
+
+ for (ent = ll_dir_entry_start(dir, op_data, &page);
+ ent != NULL && !IS_ERR(ent);
+ ent = ll_dir_entry_next(dir, op_data, ent, &page)) {
+ __u64 hash;
+ int namelen;
+ char *name;
+
+ hash = le64_to_cpu(ent->lde_hash);
+ /* The ll_get_dir_page() can return any page containing
+ * the given hash which may be not the start hash. */
+ if (unlikely(hash < op_data->op_hash_offset))
+ continue;
+
+ namelen = le16_to_cpu(ent->lde_namelen);
+ if (unlikely(namelen == 0))
+ /*
+ * skip dummy record.
+ */
+ continue;
+
+ name = ent->lde_name;
+ if (name[0] == '.') {
+ if (namelen == 1)
+ /*
+ * skip "."
+ */
+ continue;
+ else if (name[1] == '.' && namelen == 2)
+ /*
+ * skip ".."
+ */
+ continue;
+ else
+ dot_de = 1;
+ } else {
+ dot_de = 0;
+ }
- if (dot_de && target->name[0] != '.') {
- CDEBUG(D_READA, "%.*s skip hidden file %.*s\n",
- target->len, target->name,
- namelen, name);
- continue;
- }
+ if (dot_de && target->name[0] != '.') {
+ CDEBUG(D_READA, "%.*s skip hidden file %.*s\n",
+ target->len, target->name,
+ namelen, name);
+ continue;
+ }
- if (target->len != namelen ||
- memcmp(target->name, name, namelen) != 0)
- rc = LS_NONE_FIRST_DE;
- else if (!dot_de)
- rc = LS_FIRST_DE;
- else
- rc = LS_FIRST_DOT_DE;
+ if (target->len != namelen ||
+ memcmp(target->name, name, namelen) != 0)
+ rc = LS_NONE_FIRST_DE;
+ else if (!dot_de)
+ rc = LS_FIRST_DE;
+ else
+ rc = LS_FIRST_DOT_DE;
- ll_release_page(page, 0);
- GOTO(out, rc);
- }
- pos = le64_to_cpu(dp->ldp_hash_end);
- if (pos == MDS_DIR_END_OFF) {
- /*
- * End of directory reached.
- */
- ll_release_page(page, 0);
- break;
- } else if (1) {
- /*
- * chain is exhausted
- * Normal case: continue to the next page.
- */
- ll_release_page(page, le32_to_cpu(dp->ldp_flags) &
- LDF_COLLIDE);
- page = ll_get_dir_page(dir, pos, &chain);
- } else {
- /*
- * go into overflow page.
- */
- LASSERT(le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
- ll_release_page(page, 1);
- }
- }
+ break;
+ }
EXIT;
+ if (page != NULL) {
+ kunmap(page);
+ page_cache_release(page);
+ }
+ ll_finish_md_op_data(op_data);
out:
- ll_dir_chain_fini(&chain);
+ ll_dir_chain_fini(&chain);
return rc;
}
} else {
struct ll_inode_info *lli = ll_i2info(sai->sai_inode);
- sai->sai_miss++;
- sai->sai_consecutive_miss++;
- if (sa_low_hit(sai) && thread_is_running(thread)) {
- atomic_inc(&sbi->ll_sa_wrong);
- CDEBUG(D_READA, "Statahead for dir "DFID" hit "
- "ratio too low: hit/miss "LPU64"/"LPU64
- ", sent/replied "LPU64"/"LPU64", stopping "
- "statahead thread: pid %d\n",
- PFID(&lli->lli_fid), sai->sai_hit,
- sai->sai_miss, sai->sai_sent,
- sai->sai_replied, cfs_curproc_pid());
+ sai->sai_miss++;
+ sai->sai_consecutive_miss++;
+ if (sa_low_hit(sai) && thread_is_running(thread)) {
+ atomic_inc(&sbi->ll_sa_wrong);
+ CDEBUG(D_READA, "Statahead for dir "DFID" hit "
+ "ratio too low: hit/miss "LPU64"/"LPU64
+ ", sent/replied "LPU64"/"LPU64", stopping "
+ "statahead thread\n",
+ PFID(&lli->lli_fid), sai->sai_hit,
+ sai->sai_miss, sai->sai_sent,
+ sai->sai_replied);
spin_lock(&lli->lli_sa_lock);
if (!thread_is_stopped(thread))
thread_set_flags(thread, SVC_STOPPING);
}
if (!thread_is_stopped(thread))
- cfs_waitq_signal(&thread->t_ctl_waitq);
+ wake_up(&thread->t_ctl_waitq);
EXIT;
}
int do_statahead_enter(struct inode *dir, struct dentry **dentryp,
int only_unplug)
{
- struct ll_inode_info *lli = ll_i2info(dir);
- struct ll_statahead_info *sai = lli->lli_sai;
- struct dentry *parent;
- struct ll_sa_entry *entry;
- struct ptlrpc_thread *thread;
- struct l_wait_info lwi = { 0 };
- int rc = 0;
- struct ll_inode_info *plli;
- ENTRY;
+ struct ll_inode_info *lli = ll_i2info(dir);
+ struct ll_statahead_info *sai = lli->lli_sai;
+ struct dentry *parent;
+ struct ll_sa_entry *entry;
+ struct ptlrpc_thread *thread;
+ struct l_wait_info lwi = { 0 };
+ struct task_struct *task;
+ int rc = 0;
+ struct ll_inode_info *plli;
+ ENTRY;
- LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
+ LASSERT(lli->lli_opendir_pid == current_pid());
if (sai) {
thread = &sai->sai_thread;
RETURN(entry ? 1 : -EAGAIN);
}
- /* if statahead is busy in readdir, help it do post-work */
- while (!ll_sa_entry_stated(entry) &&
- sai->sai_in_readpage &&
- !sa_received_empty(sai))
- ll_post_statahead(sai);
-
if (!ll_sa_entry_stated(entry)) {
sai->sai_index_wait = entry->se_index;
lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(30), NULL,
ll_inode2fid(inode), &bits);
if (rc == 1) {
if ((*dentryp)->d_inode == NULL) {
- *dentryp = ll_splice_alias(inode,
- *dentryp);
- if (IS_ERR(*dentryp)) {
+ struct dentry *alias;
+
+ alias = ll_splice_alias(inode,
+ *dentryp);
+ if (IS_ERR(alias)) {
ll_sai_unplug(sai, entry);
- RETURN(PTR_ERR(*dentryp));
+ RETURN(PTR_ERR(alias));
}
+ *dentryp = alias;
} else if ((*dentryp)->d_inode != inode) {
/* revalidate, but inode is recreated */
CDEBUG(D_READA,
- "stale dentry %.*s inode %lu/%u, "
- "statahead inode %lu/%u\n",
- (*dentryp)->d_name.len,
- (*dentryp)->d_name.name,
- (*dentryp)->d_inode->i_ino,
- (*dentryp)->d_inode->i_generation,
- inode->i_ino,
- inode->i_generation);
- ll_sai_unplug(sai, entry);
+ "%s: stale dentry %.*s inode "
+ DFID", statahead inode "DFID
+ "\n",
+ ll_get_fsname((*dentryp)->d_inode->i_sb, NULL, 0),
+ (*dentryp)->d_name.len,
+ (*dentryp)->d_name.name,
+ PFID(ll_inode2fid((*dentryp)->d_inode)),
+ PFID(ll_inode2fid(inode)));
+ ll_intent_release(&it);
+ ll_sai_unplug(sai, entry);
RETURN(-ESTALE);
} else {
iput(inode);
GOTO(out, rc = -EAGAIN);
}
- CDEBUG(D_READA, "start statahead thread: [pid %d] [parent %.*s]\n",
- cfs_curproc_pid(), parent->d_name.len, parent->d_name.name);
+ CDEBUG(D_READA, "start statahead thread: sai %p, parent %.*s\n",
+ sai, parent->d_name.len, parent->d_name.name);
+ /* The sai buffer already has one reference taken at allocation time,
+ * but as soon as we expose the sai by attaching it to the lli that
+ * default reference can be dropped by another thread calling
+ * ll_stop_statahead. We need to take a local reference to protect
+ * the sai buffer while we intend to access it. */
+ ll_sai_get(sai);
lli->lli_sai = sai;
plli = ll_i2info(parent->d_inode);
- rc = PTR_ERR(kthread_run(ll_statahead_thread, parent,
- "ll_sa_%u", plli->lli_opendir_pid));
+ task = kthread_run(ll_statahead_thread, parent, "ll_sa_%u",
+ plli->lli_opendir_pid);
thread = &sai->sai_thread;
- if (IS_ERR_VALUE(rc)) {
- CERROR("can't start ll_sa thread, rc: %d\n", rc);
+ if (IS_ERR(task)) {
+ rc = PTR_ERR(task);
+ CERROR("cannot start ll_sa thread: rc = %d\n", rc);
dput(parent);
lli->lli_opendir_key = NULL;
thread_set_flags(thread, SVC_STOPPED);
thread_set_flags(&sai->sai_agl_thread, SVC_STOPPED);
+ /* Drop both our own local reference and the default
+ * reference from allocation time. */
ll_sai_put(sai);
+ ll_sai_put(sai);
LASSERT(lli->lli_sai == NULL);
RETURN(-EAGAIN);
}
l_wait_event(thread->t_ctl_waitq,
thread_is_running(thread) || thread_is_stopped(thread),
&lwi);
+ ll_sai_put(sai);
/*
* We don't stat-ahead for the first dirent since we are already in