file.c

lustre 1.6.5 source code

        lock_page(page);

        /* Already truncated by somebody */
        if (!page->mapping)
                GOTO(out, rc = 0);

        mapping = page->mapping;

        ll_teardown_mmaps(mapping,
                          (__u64)page->index << PAGE_CACHE_SHIFT,
                          ((__u64)page->index<<PAGE_CACHE_SHIFT)|
                                                              ~PAGE_CACHE_MASK);        
        LL_CDEBUG_PAGE(D_PAGE, page, "removing page\n");

        if (!discard && clear_page_dirty_for_io(page)) {
                LASSERT(page->mapping);
                rc = ll_call_writepage(page->mapping->host, page);
                /* either waiting for io to complete or reacquiring
                 * the lock that the failed writepage released */
                lock_page(page);
                wait_on_page_writeback(page);
                if (rc != 0) {
                        CERROR("writepage inode %lu(%p) of page %p "
                               "failed: %d\n", mapping->host->i_ino,
                               mapping->host, page, rc);
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
                        if (rc == -ENOSPC)
                                set_bit(AS_ENOSPC, &mapping->flags);
                        else
                                set_bit(AS_EIO, &mapping->flags);
#else
                        mapping->gfp_mask |= AS_EIO_MASK;
#endif
                }
        }
        if (page->mapping != NULL) {
                struct ll_async_page *llap = llap_cast_private(page);
                // checking again to account for writeback's lock_page()
                LL_CDEBUG_PAGE(D_PAGE, page, "truncating\n");
                if (llap)
                        ll_ra_accounting(llap, page->mapping);
                ll_truncate_complete_page(page);
        }
        EXIT;
out:
        LASSERT(!PageWriteback(page));
        unlock_page(page);
        page_cache_release(page);

        return 0;
}

int ll_extent_lock_cancel_cb(struct ldlm_lock *lock, struct ldlm_lock_desc *new,
                             void *data, int flag)
{
        struct inode *inode;
        struct ll_inode_info *lli;
        struct lov_stripe_md *lsm;
        int stripe;
        __u64 kms;

        ENTRY;

        if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) {
                LDLM_ERROR(lock, "cancelling lock with bad data %p", data);
                LBUG();
        }

        inode = ll_inode_from_lock(lock);
        if (inode == NULL)
                RETURN(0);
        lli = ll_i2info(inode);
        if (lli == NULL)
                GOTO(iput, 0);
        if (lli->lli_smd == NULL)
                GOTO(iput, 0);
        lsm = lli->lli_smd;

        stripe = ll_lock_to_stripe_offset(inode, lock);
        if (stripe < 0)
                GOTO(iput, 0);

        lov_stripe_lock(lsm);
        lock_res_and_lock(lock);
        kms = ldlm_extent_shift_kms(lock,
                                    lsm->lsm_oinfo[stripe]->loi_kms);

        if (lsm->lsm_oinfo[stripe]->loi_kms != kms)
                LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
                           lsm->lsm_oinfo[stripe]->loi_kms, kms);
        lsm->lsm_oinfo[stripe]->loi_kms = kms;
        unlock_res_and_lock(lock);
        lov_stripe_unlock(lsm);
        ll_try_done_writing(inode);
        EXIT;
iput:
        iput(inode);

        return 0;
}

#if 0
int ll_async_completion_ast(struct ldlm_lock *lock, int flags, void *data)
{
        /* XXX ALLOCATE - 160 bytes */
        struct inode *inode = ll_inode_from_lock(lock);
        struct ll_inode_info *lli = ll_i2info(inode);
        struct lustre_handle lockh = { 0 };
        struct ost_lvb *lvb;
        int stripe;
        ENTRY;

        if (flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
                     LDLM_FL_BLOCK_CONV)) {
                LBUG(); /* not expecting any blocked async locks yet */
                LDLM_DEBUG(lock, "client-side async enqueue returned a blocked "
                           "lock, returning");
                ldlm_lock_dump(D_OTHER, lock, 0);
                ldlm_reprocess_all(lock->l_resource);
                RETURN(0);
        }

        LDLM_DEBUG(lock, "client-side async enqueue: granted/glimpsed");

        stripe = ll_lock_to_stripe_offset(inode, lock);
        if (stripe < 0)
                goto iput;

        if (lock->l_lvb_len) {
                struct lov_stripe_md *lsm = lli->lli_smd;
                __u64 kms;
                lvb = lock->l_lvb_data;
                lsm->lsm_oinfo[stripe].loi_rss = lvb->lvb_size;

                lock_res_and_lock(lock);
                ll_inode_size_lock(inode, 1);
                kms = MAX(lsm->lsm_oinfo[stripe].loi_kms, lvb->lvb_size);
                kms = ldlm_extent_shift_kms(NULL, kms);
                if (lsm->lsm_oinfo[stripe].loi_kms != kms)
                        LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
                                   lsm->lsm_oinfo[stripe].loi_kms, kms);
                lsm->lsm_oinfo[stripe].loi_kms = kms;
                ll_inode_size_unlock(inode, 1);
                unlock_res_and_lock(lock);
        }

iput:
        iput(inode);
        wake_up(&lock->l_waitq);

        ldlm_lock2handle(lock, &lockh);
        ldlm_lock_decref(&lockh, LCK_PR);
        RETURN(0);
}
#endif

static int ll_glimpse_callback(struct ldlm_lock *lock, void *reqp)
{
        struct ptlrpc_request *req = reqp;
        struct inode *inode = ll_inode_from_lock(lock);
        struct ll_inode_info *lli;
        struct lov_stripe_md *lsm;
        struct ost_lvb *lvb;
        int rc, stripe;
        int size[2] = { sizeof(struct ptlrpc_body), sizeof(*lvb) };
        ENTRY;

        if (inode == NULL)
                GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
        lli = ll_i2info(inode);
        if (lli == NULL)
                GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
        lsm = lli->lli_smd;
        if (lsm == NULL)
                GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);

        /* First, find out which stripe index this lock corresponds to. */
        stripe = ll_lock_to_stripe_offset(inode, lock);
        if (stripe < 0)
                GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);

        rc = lustre_pack_reply(req, 2, size, NULL);
        if (rc)
                GOTO(iput, rc);

        lvb = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*lvb));
        lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms;
        lvb->lvb_mtime = LTIME_S(inode->i_mtime);
        lvb->lvb_atime = LTIME_S(inode->i_atime);
        lvb->lvb_ctime = LTIME_S(inode->i_ctime);

        LDLM_DEBUG(lock, "i_size: %llu -> stripe number %u -> kms "LPU64
                   " atime "LPU64", mtime "LPU64", ctime "LPU64,
                   i_size_read(inode), stripe, lvb->lvb_size, lvb->lvb_mtime,
                   lvb->lvb_atime, lvb->lvb_ctime);
 iput:
        iput(inode);

 out:
        /* These errors are normal races, so we don't want to fill the console
         * with messages by calling ptlrpc_error() */
        if (rc == -ELDLM_NO_LOCK_DATA)
                lustre_pack_reply(req, 1, NULL, NULL);

        req->rq_status = rc;
        return rc;
}

int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
                     lstat_t *st)
{
        struct lustre_handle lockh = { 0 };
        struct ldlm_enqueue_info einfo = { 0 };
        struct obd_info oinfo = { { { 0 } } };
        struct ost_lvb lvb;
        int rc;
        
        ENTRY;
        
        einfo.ei_type = LDLM_EXTENT;
        einfo.ei_mode = LCK_PR;
        einfo.ei_cb_bl = osc_extent_blocking_cb;
        einfo.ei_cb_cp = ldlm_completion_ast;
        einfo.ei_cb_gl = ll_glimpse_callback;
        einfo.ei_cbdata = NULL;

        oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF;
        oinfo.oi_lockh = &lockh;
        oinfo.oi_md = lsm;
        oinfo.oi_flags = LDLM_FL_HAS_INTENT;

        rc = obd_enqueue_rqset(sbi->ll_osc_exp, &oinfo, &einfo);
        if (rc == -ENOENT)
                RETURN(rc);
        if (rc != 0) {
                CERROR("obd_enqueue returned rc %d, "
                       "returning -EIO\n", rc);
                RETURN(rc > 0 ? -EIO : rc);
        }
        
        lov_stripe_lock(lsm);
        memset(&lvb, 0, sizeof(lvb));
        obd_merge_lvb(sbi->ll_osc_exp, lsm, &lvb, 0);
        st->st_size = lvb.lvb_size;
        st->st_blocks = lvb.lvb_blocks;
        st->st_mtime = lvb.lvb_mtime;
        st->st_atime = lvb.lvb_atime;
        st->st_ctime = lvb.lvb_ctime;
        lov_stripe_unlock(lsm);
        
        RETURN(rc);
}

/* NB: obd_merge_lvb will prefer locally cached writes if they extend the
 * file (because it prefers KMS over RSS when larger) */
int ll_glimpse_size(struct inode *inode, int ast_flags)
{
        struct ll_inode_info *lli = ll_i2info(inode);
        struct ll_sb_info *sbi = ll_i2sbi(inode);
        struct lustre_handle lockh = { 0 };
        struct ldlm_enqueue_info einfo = { 0 };
        struct obd_info oinfo = { { { 0 } } };
        struct ost_lvb lvb;
        int rc;
        ENTRY;

        CDEBUG(D_DLMTRACE, "Glimpsing inode %lu\n", inode->i_ino);

        if (!lli->lli_smd) {
                CDEBUG(D_DLMTRACE, "No objects for inode %lu\n", inode->i_ino);
                RETURN(0);
        }

        /* NOTE: this looks like DLM lock request, but it may not be one. Due
         *       to LDLM_FL_HAS_INTENT flag, this is glimpse request, that
         *       won't revoke any conflicting DLM locks held. Instead,
         *       ll_glimpse_callback() will be called on each client
         *       holding a DLM lock against this file, and resulting size
         *       will be returned for each stripe. DLM lock on [0, EOF] is
         *       acquired only if there were no conflicting locks. */
        einfo.ei_type = LDLM_EXTENT;
        einfo.ei_mode = LCK_PR;
        einfo.ei_cb_bl = osc_extent_blocking_cb;
        einfo.ei_cb_cp = ldlm_completion_ast;
        einfo.ei_cb_gl = ll_glimpse_callback;
        einfo.ei_cbdata = inode;

        oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF;
        oinfo.oi_lockh = &lockh;
        oinfo.oi_md = lli->lli_smd;
        oinfo.oi_flags = ast_flags | LDLM_FL_HAS_INTENT;

        rc = obd_enqueue_rqset(sbi->ll_osc_exp, &oinfo, &einfo);
        if (rc == -ENOENT)
                RETURN(rc);
        if (rc != 0) {
                CERROR("obd_enqueue returned rc %d, returning -EIO\n", rc);
                RETURN(rc > 0 ? -EIO : rc);
        }

        ll_inode_size_lock(inode, 1);
        inode_init_lvb(inode, &lvb);
        rc = obd_merge_lvb(sbi->ll_osc_exp, lli->lli_smd, &lvb, 0);
        i_size_write(inode, lvb.lvb_size);
        inode->i_blocks = lvb.lvb_blocks;
        LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
        LTIME_S(inode->i_atime) = lvb.lvb_atime;
        LTIME_S(inode->i_ctime) = lvb.lvb_ctime;
        ll_inode_size_unlock(inode, 1);

        CDEBUG(D_DLMTRACE, "glimpse: size: %llu, blocks: %llu\n",
               i_size_read(inode), (long long)inode->i_blocks);

        RETURN(rc);
}

int ll_extent_lock(struct ll_file_data *fd, struct inode *inode,
                   struct lov_stripe_md *lsm, int mode,
                   ldlm_policy_data_t *policy, struct lustre_handle *lockh,
                   int ast_flags)
{
        struct ll_sb_info *sbi = ll_i2sbi(inode);
        struct ost_lvb lvb;
        struct ldlm_enqueue_info einfo = { 0 };
        struct obd_info oinfo = { { { 0 } } };
        int rc;
        ENTRY;

        LASSERT(!lustre_handle_is_used(lockh));
        LASSERT(lsm != NULL);

        /* don't drop the mmapped file to LRU */
        if (mapping_mapped(inode->i_mapping))
                ast_flags |= LDLM_FL_NO_LRU;

        /* XXX phil: can we do this?  won't it screw the file size up? */
        if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
            (sbi->ll_flags & LL_SBI_NOLCK))
                RETURN(0);

        CDEBUG(D_DLMTRACE, "Locking inode %lu, start "LPU64" end "LPU64"\n",
               inode->i_ino, policy->l_extent.start, policy->l_extent.end);

        einfo.ei_type = LDLM_EXTENT;
        einfo.ei_mode = mode;
        einfo.ei_cb_bl = osc_extent_blocking_cb;
        einfo.ei_cb_cp = ldlm_completion_ast;
        einfo.ei_cb_gl = ll_glimpse_callback;
        einfo.ei_cbdata = inode;

        oinfo.oi_policy = *policy;
        oinfo.oi_lockh = lockh;