statahead.c

lustre 1.6.5 source code

                        rc = ll_statahead_one(parent, de);
                        if (rc < 0) {
                                ext2_put_page(page);
                                GOTO(out, rc);
                        }
                }
                ext2_put_page(page);
                index++;
        }
        EXIT;
out:
        spin_lock(&lli->lli_lock);
        thread->t_flags = SVC_STOPPED;
        spin_unlock(&lli->lli_lock);
        cfs_waitq_signal(&sai->sai_waitq);
        cfs_waitq_signal(&thread->t_ctl_waitq);
        ll_sai_put(sai);
        dput(parent);
        CDEBUG(D_READA, "statahead thread stopped, pid %d\n",
               cfs_curproc_pid());
        return rc;
}

/* called in ll_file_release() */
void ll_stop_statahead(struct inode *inode, void *key)
{
        struct ll_inode_info *lli = ll_i2info(inode);
        struct ptlrpc_thread *thread;

        spin_lock(&lli->lli_lock);
        if (lli->lli_opendir_pid == 0 ||
            unlikely(lli->lli_opendir_key != key)) {
                spin_unlock(&lli->lli_lock);
                return;
        }

        lli->lli_opendir_key = NULL;
        lli->lli_opendir_pid = 0;

        if (lli->lli_sai) {
                struct l_wait_info lwi = { 0 };

                thread = &lli->lli_sai->sai_thread;
                if (!(thread->t_flags & SVC_STOPPED)) {
                        thread->t_flags = SVC_STOPPING;
                        spin_unlock(&lli->lli_lock);
                        cfs_waitq_signal(&thread->t_ctl_waitq);

                        CDEBUG(D_READA, "stopping statahead thread, pid %d\n",
                               cfs_curproc_pid());
                        l_wait_event(thread->t_ctl_waitq,
                                     thread->t_flags & SVC_STOPPED,
                                     &lwi);
                } else {
                        spin_unlock(&lli->lli_lock);
                }

                /*
                 * Put the ref which was held when first statahead_enter.
                 * It maybe not the last ref for some statahead requests
                 * maybe inflight.
                 */
                ll_sai_put(lli->lli_sai);
                return;
        }
        spin_unlock(&lli->lli_lock);
}

enum {
        LS_NONE_FIRST_DE = 0,   /* not first dirent, or is "." */
        LS_FIRST_DE,            /* the first non-hidden dirent */
        LS_FIRST_DOT_DE         /* the first hidden dirent, that is ".xxx" */
};

static int is_first_dirent(struct inode *dir, struct dentry *dentry)
{
        struct qstr   *d_name = &dentry->d_name;
        unsigned long  npages, index = 0;
        struct page   *page;
        ext2_dirent   *de;
        char          *kaddr, *limit;
        int            rc = LS_NONE_FIRST_DE, dot_de;
        ENTRY;

        while (1) {
                npages = dir_pages(dir);
                /* reach the end of dir */
                if (index >= npages) {
                        CDEBUG(D_READA, "reach end, index/npages %lu/%lu\n",
                               index, npages);
                        break;
                }

                page = ll_get_dir_page(dir, index);
                if (IS_ERR(page)) {
                        rc = PTR_ERR(page);
                        CERROR("error reading dir %lu/%u page %lu: rc %d\n",
                               dir->i_ino, dir->i_generation, index, rc);
                        break;
                }

                kaddr = page_address(page);
                limit = kaddr + CFS_PAGE_SIZE - EXT2_DIR_REC_LEN(1);
                de = (ext2_dirent *)kaddr;
                if (!index) {
                        de = ext2_next_entry(de); /* skip "." */
                        de = ext2_next_entry(de); /* skip ".." */
                }

                for (; (char*)de <= limit; de = ext2_next_entry(de)) {
                        if (!de->inode)
                                continue;

                        if (de->name[0] == '.')
                                dot_de = 1;
                        else
                                dot_de = 0;

                        if (dot_de && d_name->name[0] != '.') {
                                CDEBUG(D_READA, "%.*s skip hidden file %.*s\n",
                                       d_name->len, d_name->name,
                                       de->name_len, de->name);
                                continue;
                        }

                        if (d_name->len == de->name_len &&
                            !strncmp(d_name->name, de->name, d_name->len))
                                rc = LS_FIRST_DE + dot_de;
                        else
                                rc = LS_NONE_FIRST_DE;
                        ext2_put_page(page);
                        RETURN(rc);
                }
                ext2_put_page(page);
                index++;
        }
        RETURN(rc);
}

/* Start statahead thread if this is the first dir entry.
 * Otherwise if a thread is started already, wait it until it is ahead of me.
 * Return value: 
 *  0       -- miss
 *  1       -- hit
 *  -EEXIST -- stat ahead thread started, and this is the first dentry
 *  -EBADFD -- statahead thread exit and not dentry available
 *  others  -- error
 */
int do_statahead_enter(struct inode *dir, struct dentry **dentryp, int lookup)
{
        struct ll_sb_info        *sbi = ll_i2sbi(dir);
        struct ll_inode_info     *lli = ll_i2info(dir);
        struct ll_statahead_info *sai = lli->lli_sai;
        struct ll_sa_thread_args  sta;
        struct l_wait_info        lwi = { 0 };
        int                       rc;
        ENTRY;

        LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());

        if (sai) {
                if (unlikely(sai->sai_thread.t_flags & SVC_STOPPED &&
                             list_empty(&sai->sai_entries)))
                        RETURN(-EBADFD);

                if ((*dentryp)->d_name.name[0] == '.') {
                        if (likely(sai->sai_ls_all ||
                            sai->sai_miss_hidden >= sai->sai_skip_hidden)) {
                                /*
                                 * Hidden dentry is the first one, or statahead
                                 * thread does not skip so many hidden dentries
                                 * before "sai_ls_all" enabled as below.
                                 */
                        } else {
                                if (!sai->sai_ls_all)
                                        /*
                                         * It maybe because hidden dentry is not
                                         * the first one, "sai_ls_all" was not
                                         * set, then "ls -al" missed. Enable
                                         * "sai_ls_all" for such case.
                                         */
                                        sai->sai_ls_all = 1;

                                /*
                                 * Such "getattr" has been skipped before
                                 * "sai_ls_all" enabled as above.
                                 */
                                sai->sai_miss_hidden++;
                                RETURN(-ENOENT);
                        }
                }

                if (ll_sai_entry_stated(sai)) {
                        sbi->ll_sa_cached++;
                } else {
                        sbi->ll_sa_blocked++;
                        /* thread started already, avoid double-stat */
                        l_wait_event(sai->sai_waitq,
                                     ll_sai_entry_stated(sai) ||
                                     sai->sai_thread.t_flags & SVC_STOPPED,
                                     &lwi);
                }

                if (lookup) {
                        struct dentry *result;

                        result = d_lookup((*dentryp)->d_parent,
                                          &(*dentryp)->d_name);
                        if (result) {
                                LASSERT(result != *dentryp);
                                dput(*dentryp);
                                *dentryp = result;
                                RETURN(1);
                        }
                }
                /* do nothing for revalidate */
                RETURN(0);
        }

         /* I am the "lli_opendir_pid" owner, only me can set "lli_sai". */ 
        LASSERT(lli->lli_sai == NULL);

        rc = is_first_dirent(dir, *dentryp);
        if (rc == LS_NONE_FIRST_DE) {
                /* It is not "ls -{a}l" operation, no need statahead for it */
                spin_lock(&lli->lli_lock);
                lli->lli_opendir_key = NULL;
                lli->lli_opendir_pid = 0;
                spin_unlock(&lli->lli_lock);
                RETURN(-EBADF);
        }

        sai = ll_sai_alloc();
        if (sai == NULL)
                RETURN(-ENOMEM);
        
        sai->sai_inode  = igrab(dir);
        sai->sai_ls_all = (rc == LS_FIRST_DOT_DE);

        sta.sta_parent = (*dentryp)->d_parent;
        sta.sta_pid    = cfs_curproc_pid();

        lli->lli_sai = sai;
        rc = cfs_kernel_thread(ll_statahead_thread, &sta, 0);
        if (rc < 0) {
                CERROR("can't start ll_sa thread, rc: %d\n", rc);
                sai->sai_thread.t_flags = SVC_STOPPED;
                ll_sai_put(sai);
                LASSERT(lli->lli_sai == NULL);
                RETURN(rc);
        }

        l_wait_event(sai->sai_thread.t_ctl_waitq, 
                     sai->sai_thread.t_flags & (SVC_RUNNING | SVC_STOPPED),
                     &lwi);

        /*
         * We don't stat-ahead for the first dirent since we are already in
         * lookup, and -EEXIST also indicates that this is the first dirent.
         */
        RETURN(-EEXIST);
}

/* update hit/miss count */
void ll_statahead_exit(struct dentry *dentry, int result)
{
        struct dentry         *parent = dentry->d_parent;
        struct ll_inode_info  *lli = ll_i2info(parent->d_inode);
        struct ll_sb_info     *sbi = ll_i2sbi(parent->d_inode);
        struct ll_dentry_data *ldd = ll_d2d(dentry);

        if (lli->lli_opendir_pid != cfs_curproc_pid())
                return;

        if (lli->lli_sai) {
                struct ll_statahead_info *sai = lli->lli_sai;

                if (result == 1) {
                        sbi->ll_sa_hit++;
                        sai->sai_hit++;
                        sai->sai_consecutive_miss = 0;
                        sai->sai_max = min(2 * sai->sai_max, sbi->ll_sa_max);
                } else {
                        sbi->ll_sa_miss++;
                        sai->sai_miss++;
                        sai->sai_consecutive_miss++;
                        if (sa_low_hit(sai) &&
                            sai->sai_thread.t_flags & SVC_RUNNING) {
                                sbi->ll_sa_wrong++;
                                CDEBUG(D_READA, "statahead for dir %.*s hit "
                                       "ratio too low: hit/miss %u/%u, "
                                       "sent/replied %u/%u. stopping statahead "
                                       "thread: pid %d\n",
                                       parent->d_name.len, parent->d_name.name,
                                       sai->sai_hit, sai->sai_miss,
                                       sai->sai_sent, sai->sai_replied,
                                       cfs_curproc_pid());
                                spin_lock(&lli->lli_lock);
                                if (!(sai->sai_thread.t_flags & SVC_STOPPED))
                                        sai->sai_thread.t_flags = SVC_STOPPING;
                                spin_unlock(&lli->lli_lock);
                        }
                }

                cfs_waitq_signal(&sai->sai_thread.t_ctl_waitq);
                ll_sai_entry_put(sai);

                if (likely(ldd != NULL))
                        ldd->lld_sa_generation = sai->sai_generation;
        }
}