-static DEFINE_MUTEX(ccc_inode_fini_guard);
-static int dummy_refcheck;
-
-int ccc_global_init(struct lu_device_type *device_type)
-{
- int result;
-
- result = lu_kmem_init(ccc_caches);
- if (result)
- return result;
-
- result = lu_device_type_init(device_type);
- if (result)
- goto out_kmem;
-
- ccc_inode_fini_env = cl_env_alloc(&dummy_refcheck,
- LCT_REMEMBER|LCT_NOREF);
- if (IS_ERR(ccc_inode_fini_env)) {
- result = PTR_ERR(ccc_inode_fini_env);
- goto out_device;
- }
-
- ccc_inode_fini_env->le_ctx.lc_cookie = 0x4;
- return 0;
-out_device:
- lu_device_type_fini(device_type);
-out_kmem:
- lu_kmem_fini(ccc_caches);
- return result;
-}
-
-void ccc_global_fini(struct lu_device_type *device_type)
-{
- if (ccc_inode_fini_env != NULL) {
- cl_env_put(ccc_inode_fini_env, &dummy_refcheck);
- ccc_inode_fini_env = NULL;
- }
- lu_device_type_fini(device_type);
- lu_kmem_fini(ccc_caches);
-}
-
-/*****************************************************************************
- *
- * Object operations.
- *
- */
-
-struct lu_object *ccc_object_alloc(const struct lu_env *env,
- const struct lu_object_header *unused,
- struct lu_device *dev,
- const struct cl_object_operations *clops,
- const struct lu_object_operations *luops)
-{
- struct ccc_object *vob;
- struct lu_object *obj;
-
- OBD_SLAB_ALLOC_PTR_GFP(vob, ccc_object_kmem, GFP_NOFS);
- if (vob != NULL) {
- struct cl_object_header *hdr;
-
- obj = ccc2lu(vob);
- hdr = &vob->cob_header;
- cl_object_header_init(hdr);
- hdr->coh_page_bufsize = cfs_size_round(sizeof(struct cl_page));
-
- lu_object_init(obj, &hdr->coh_lu, dev);
- lu_object_add_top(&hdr->coh_lu, obj);
-
- vob->cob_cl.co_ops = clops;
- obj->lo_ops = luops;
- } else
- obj = NULL;
- return obj;
-}
-
-int ccc_object_init0(const struct lu_env *env,
- struct ccc_object *vob,
- const struct cl_object_conf *conf)
-{
- vob->cob_inode = conf->coc_inode;
- vob->cob_transient_pages = 0;
- cl_object_page_init(&vob->cob_cl, sizeof(struct ccc_page));
- return 0;
-}
-
-int ccc_object_init(const struct lu_env *env, struct lu_object *obj,
- const struct lu_object_conf *conf)
-{
- struct ccc_device *dev = lu2ccc_dev(obj->lo_dev);
- struct ccc_object *vob = lu2ccc(obj);
- struct lu_object *below;
- struct lu_device *under;
- int result;
-
- under = &dev->cdv_next->cd_lu_dev;
- below = under->ld_ops->ldo_object_alloc(env, obj->lo_header, under);
- if (below != NULL) {
- const struct cl_object_conf *cconf;
-
- cconf = lu2cl_conf(conf);
- INIT_LIST_HEAD(&vob->cob_pending_list);
- lu_object_add(obj, below);
- result = ccc_object_init0(env, vob, cconf);
- } else
- result = -ENOMEM;
- return result;
-}
-
-void ccc_object_free(const struct lu_env *env, struct lu_object *obj)
-{
- struct ccc_object *vob = lu2ccc(obj);
-
- lu_object_fini(obj);
- lu_object_header_fini(obj->lo_header);
- OBD_SLAB_FREE_PTR(vob, ccc_object_kmem);
-}
-
-int ccc_lock_init(const struct lu_env *env,
- struct cl_object *obj, struct cl_lock *lock,
- const struct cl_io *unused,
- const struct cl_lock_operations *lkops)
-{
- struct ccc_lock *clk;
- int result;
-
- CLOBINVRNT(env, obj, ccc_object_invariant(obj));
-
- OBD_SLAB_ALLOC_PTR_GFP(clk, ccc_lock_kmem, GFP_NOFS);
- if (clk != NULL) {
- cl_lock_slice_add(lock, &clk->clk_cl, obj, lkops);
- result = 0;
- } else
- result = -ENOMEM;
- return result;
-}
-
-int ccc_attr_set(const struct lu_env *env, struct cl_object *obj,
- const struct cl_attr *attr, unsigned valid)
-{
- return 0;
-}
-
-int ccc_object_glimpse(const struct lu_env *env,
- const struct cl_object *obj, struct ost_lvb *lvb)
-{
- struct inode *inode = ccc_object_inode(obj);
-
- ENTRY;
- lvb->lvb_mtime = cl_inode_mtime(inode);
- lvb->lvb_atime = cl_inode_atime(inode);
- lvb->lvb_ctime = cl_inode_ctime(inode);
- /*
- * LU-417: Add dirty pages block count lest i_blocks reports 0, some
- * "cp" or "tar" on remote node may think it's a completely sparse file
- * and skip it.
- */
- if (lvb->lvb_size > 0 && lvb->lvb_blocks == 0)
- lvb->lvb_blocks = dirty_cnt(inode);
- RETURN(0);
-}
-
-
-
-int ccc_conf_set(const struct lu_env *env, struct cl_object *obj,
- const struct cl_object_conf *conf)
-{
- /* TODO: destroy all pages attached to this object. */
- return 0;
-}
-
-static void ccc_object_size_lock(struct cl_object *obj)
-{
- struct inode *inode = ccc_object_inode(obj);
-
- cl_isize_lock(inode);
- cl_object_attr_lock(obj);
-}
-
-static void ccc_object_size_unlock(struct cl_object *obj)
-{
- struct inode *inode = ccc_object_inode(obj);
-
- cl_object_attr_unlock(obj);
- cl_isize_unlock(inode);
-}
-
-/*****************************************************************************
- *
- * Page operations.
- *
- */
-
-int ccc_fail(const struct lu_env *env, const struct cl_page_slice *slice)
-{
- /*
- * Cached read?
- */
- LBUG();
- return 0;
-}
-
-void ccc_transient_page_verify(const struct cl_page *page)
-{
-}
-
-int ccc_transient_page_own(const struct lu_env *env,
- const struct cl_page_slice *slice,
- struct cl_io *unused,
- int nonblock)
-{
- ccc_transient_page_verify(slice->cpl_page);
- return 0;
-}
-
-void ccc_transient_page_assume(const struct lu_env *env,
- const struct cl_page_slice *slice,
- struct cl_io *unused)
-{
- ccc_transient_page_verify(slice->cpl_page);
-}
-
-void ccc_transient_page_unassume(const struct lu_env *env,
- const struct cl_page_slice *slice,
- struct cl_io *unused)
-{
- ccc_transient_page_verify(slice->cpl_page);
-}
-
-void ccc_transient_page_disown(const struct lu_env *env,
- const struct cl_page_slice *slice,
- struct cl_io *unused)
-{
- ccc_transient_page_verify(slice->cpl_page);
-}
-
-void ccc_transient_page_discard(const struct lu_env *env,
- const struct cl_page_slice *slice,
- struct cl_io *unused)
-{
- struct cl_page *page = slice->cpl_page;
-
- ccc_transient_page_verify(slice->cpl_page);
-
- /*
- * For transient pages, remove it from the radix tree.
- */
- cl_page_delete(env, page);
-}
-
-int ccc_transient_page_prep(const struct lu_env *env,
- const struct cl_page_slice *slice,
- struct cl_io *unused)
-{
- ENTRY;
- /* transient page should always be sent. */
- RETURN(0);
-}
-
-/*****************************************************************************
- *
- * Lock operations.
- *
- */
-
-void ccc_lock_delete(const struct lu_env *env,
- const struct cl_lock_slice *slice)
-{
- CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
-}
-
-void ccc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice)
-{
- struct ccc_lock *clk = cl2ccc_lock(slice);
- OBD_SLAB_FREE_PTR(clk, ccc_lock_kmem);
-}
-
-int ccc_lock_enqueue(const struct lu_env *env,
- const struct cl_lock_slice *slice,
- struct cl_io *unused, __u32 enqflags)
-{
- CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
- return 0;
-}
-
-int ccc_lock_use(const struct lu_env *env, const struct cl_lock_slice *slice)
-{
- CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
- return 0;
-}
-
-int ccc_lock_unuse(const struct lu_env *env, const struct cl_lock_slice *slice)
-{
- CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
- return 0;
-}
-
-int ccc_lock_wait(const struct lu_env *env, const struct cl_lock_slice *slice)
-{
- CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
- return 0;
-}
-
-/**
- * Implementation of cl_lock_operations::clo_fits_into() methods for ccc
- * layer. This function is executed every time io finds an existing lock in
- * the lock cache while creating new lock. This function has to decide whether
- * cached lock "fits" into io.
- *
- * \param slice lock to be checked
- * \param io IO that wants a lock.
- *
- * \see lov_lock_fits_into().
- */
-int ccc_lock_fits_into(const struct lu_env *env,
- const struct cl_lock_slice *slice,
- const struct cl_lock_descr *need,
- const struct cl_io *io)
-{
- const struct cl_lock *lock = slice->cls_lock;
- const struct cl_lock_descr *descr = &lock->cll_descr;
- const struct ccc_io *cio = ccc_env_io(env);
- int result;
-
- ENTRY;
- /*
- * Work around DLM peculiarity: it assumes that glimpse
- * (LDLM_FL_HAS_INTENT) lock is always LCK_PR, and returns reads lock
- * when asked for LCK_PW lock with LDLM_FL_HAS_INTENT flag set. Make
- * sure that glimpse doesn't get CLM_WRITE top-lock, so that it
- * doesn't enqueue CLM_WRITE sub-locks.
- */
- if (cio->cui_glimpse)
- result = descr->cld_mode != CLM_WRITE;
-
- /*
- * Also, don't match incomplete write locks for read, otherwise read
- * would enqueue missing sub-locks in the write mode.
- */
- else if (need->cld_mode != descr->cld_mode)
- result = lock->cll_state >= CLS_ENQUEUED;
- else
- result = 1;
- RETURN(result);
-}
-
-/**
- * Implements cl_lock_operations::clo_state() method for ccc layer, invoked
- * whenever lock state changes. Transfers object attributes, that might be
- * updated as a result of lock acquiring into inode.
- */
-void ccc_lock_state(const struct lu_env *env,
- const struct cl_lock_slice *slice,
- enum cl_lock_state state)
-{
- struct cl_lock *lock = slice->cls_lock;
- ENTRY;
-
- /*
- * Refresh inode attributes when the lock is moving into CLS_HELD
- * state, and only when this is a result of real enqueue, rather than
- * of finding lock in the cache.
- */
- if (state == CLS_HELD && lock->cll_state < CLS_HELD) {
- struct cl_object *obj;
- struct inode *inode;
-
- obj = slice->cls_obj;
- inode = ccc_object_inode(obj);
-
- /* vmtruncate() sets the i_size
- * under both a DLM lock and the
- * ll_inode_size_lock(). If we don't get the
- * ll_inode_size_lock() here we can match the DLM lock and
- * reset i_size. generic_file_write can then trust the
- * stale i_size when doing appending writes and effectively
- * cancel the result of the truncate. Getting the
- * ll_inode_size_lock() after the enqueue maintains the DLM
- * -> ll_inode_size_lock() acquiring order. */
- if (lock->cll_descr.cld_start == 0 &&
- lock->cll_descr.cld_end == CL_PAGE_EOF)
- cl_merge_lvb(env, inode);
- }
- EXIT;
-}
-
-/*****************************************************************************
- *
- * io operations.
- *
- */
-
-void ccc_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
-{
- struct cl_io *io = ios->cis_io;
-
- CLOBINVRNT(env, io->ci_obj, ccc_object_invariant(io->ci_obj));
-}
-
-int ccc_io_one_lock_index(const struct lu_env *env, struct cl_io *io,
- __u32 enqflags, enum cl_lock_mode mode,
- pgoff_t start, pgoff_t end)
-{
- struct ccc_io *cio = ccc_env_io(env);
- struct cl_lock_descr *descr = &cio->cui_link.cill_descr;
- struct cl_object *obj = io->ci_obj;
-
- CLOBINVRNT(env, obj, ccc_object_invariant(obj));
- ENTRY;
-
- CDEBUG(D_VFSTRACE, "lock: %d [%lu, %lu]\n", mode, start, end);
-
- memset(&cio->cui_link, 0, sizeof cio->cui_link);
-
- if (cio->cui_fd && (cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
- descr->cld_mode = CLM_GROUP;
- descr->cld_gid = cio->cui_fd->fd_grouplock.cg_gid;
- } else {
- descr->cld_mode = mode;
- }
- descr->cld_obj = obj;
- descr->cld_start = start;
- descr->cld_end = end;
- descr->cld_enq_flags = enqflags;
-
- cl_io_lock_add(env, io, &cio->cui_link);
- RETURN(0);
-}
-
-void ccc_io_update_iov(const struct lu_env *env,
- struct ccc_io *cio, struct cl_io *io)
-{
- int i;
- size_t size = io->u.ci_rw.crw_count;
-
- cio->cui_iov_olen = 0;
- if (!cl_is_normalio(env, io) || cio->cui_tot_nrsegs == 0)
- return;
-
- for (i = 0; i < cio->cui_tot_nrsegs; i++) {
- struct iovec *iv = &cio->cui_iov[i];
-
- if (iv->iov_len < size)
- size -= iv->iov_len;
- else {
- if (iv->iov_len > size) {
- cio->cui_iov_olen = iv->iov_len;
- iv->iov_len = size;
- }
- break;
- }
- }
-
- cio->cui_nrsegs = i + 1;
- LASSERTF(cio->cui_tot_nrsegs >= cio->cui_nrsegs,
- "tot_nrsegs: %lu, nrsegs: %lu\n",
- cio->cui_tot_nrsegs, cio->cui_nrsegs);
-}
-
-int ccc_io_one_lock(const struct lu_env *env, struct cl_io *io,
- __u32 enqflags, enum cl_lock_mode mode,
- loff_t start, loff_t end)
-{
- struct cl_object *obj = io->ci_obj;
- return ccc_io_one_lock_index(env, io, enqflags, mode,
- cl_index(obj, start), cl_index(obj, end));
-}
-
-void ccc_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
-{
- CLOBINVRNT(env, ios->cis_io->ci_obj,
- ccc_object_invariant(ios->cis_io->ci_obj));
-}
-
-void ccc_io_advance(const struct lu_env *env,
- const struct cl_io_slice *ios,
- size_t nob)
-{
- struct ccc_io *cio = cl2ccc_io(env, ios);
- struct cl_io *io = ios->cis_io;
- struct cl_object *obj = ios->cis_io->ci_obj;
-
- CLOBINVRNT(env, obj, ccc_object_invariant(obj));
-
- if (!cl_is_normalio(env, io))
- return;
-
- LASSERT(cio->cui_tot_nrsegs >= cio->cui_nrsegs);
- LASSERT(cio->cui_tot_count >= nob);
-
- cio->cui_iov += cio->cui_nrsegs;
- cio->cui_tot_nrsegs -= cio->cui_nrsegs;
- cio->cui_tot_count -= nob;
-
- /* update the iov */
- if (cio->cui_iov_olen > 0) {
- struct iovec *iv;
-
- cio->cui_iov--;
- cio->cui_tot_nrsegs++;
- iv = &cio->cui_iov[0];
- if (io->ci_continue) {
- iv->iov_base += iv->iov_len;
- LASSERT(cio->cui_iov_olen > iv->iov_len);
- iv->iov_len = cio->cui_iov_olen - iv->iov_len;
- } else {
- /* restore the iov_len, in case of restart io. */
- iv->iov_len = cio->cui_iov_olen;
- }
- cio->cui_iov_olen = 0;
- }
-}
-
-/**
- * Helper function that if necessary adjusts file size (inode->i_size), when
- * position at the offset \a pos is accessed. File size can be arbitrary stale
- * on a Lustre client, but client at least knows KMS. If accessed area is
- * inside [0, KMS], set file size to KMS, otherwise glimpse file size.
- *
- * Locking: cl_isize_lock is used to serialize changes to inode size and to
- * protect consistency between inode size and cl_object
- * attributes. cl_object_size_lock() protects consistency between cl_attr's of
- * top-object and sub-objects.
- */
-int ccc_prep_size(const struct lu_env *env, struct cl_object *obj,
- struct cl_io *io, loff_t start, size_t count, int *exceed)
-{
- struct cl_attr *attr = ccc_env_thread_attr(env);
- struct inode *inode = ccc_object_inode(obj);
- loff_t pos = start + count - 1;
- loff_t kms;
- int result;
-
- /*
- * Consistency guarantees: following possibilities exist for the
- * relation between region being accessed and real file size at this
- * moment:
- *
- * (A): the region is completely inside of the file;
- *
- * (B-x): x bytes of region are inside of the file, the rest is
- * outside;
- *
- * (C): the region is completely outside of the file.
- *
- * This classification is stable under DLM lock already acquired by
- * the caller, because to change the class, other client has to take
- * DLM lock conflicting with our lock. Also, any updates to ->i_size
- * by other threads on this client are serialized by
- * ll_inode_size_lock(). This guarantees that short reads are handled
- * correctly in the face of concurrent writes and truncates.
- */
- ccc_object_size_lock(obj);
- result = cl_object_attr_get(env, obj, attr);
- if (result == 0) {
- kms = attr->cat_kms;
- if (pos > kms) {
- /*
- * A glimpse is necessary to determine whether we
- * return a short read (B) or some zeroes at the end
- * of the buffer (C)
- */
- ccc_object_size_unlock(obj);
- result = cl_glimpse_lock(env, io, inode, obj, 0);
- if (result == 0 && exceed != NULL) {
- /* If objective page index exceed end-of-file
- * page index, return directly. Do not expect
- * kernel will check such case correctly.
- * linux-2.6.18-128.1.1 miss to do that.
- * --bug 17336 */
- loff_t size = cl_isize_read(inode);
- unsigned long cur_index = start >>
- PAGE_CACHE_SHIFT;
-
- if ((size == 0 && cur_index != 0) ||
- (((size - 1) >> PAGE_CACHE_SHIFT) <
- cur_index))
- *exceed = 1;
- }
- return result;
- } else {
- /*
- * region is within kms and, hence, within real file
- * size (A). We need to increase i_size to cover the
- * read region so that generic_file_read() will do its
- * job, but that doesn't mean the kms size is
- * _correct_, it is only the _minimum_ size. If
- * someone does a stat they will get the correct size
- * which will always be >= the kms value here.
- * b=11081
- */
- if (cl_isize_read(inode) < kms) {
- cl_isize_write_nolock(inode, kms);
- CDEBUG(D_VFSTRACE,
- DFID" updating i_size "LPU64"\n",
- PFID(lu_object_fid(&obj->co_lu)),
- (__u64)cl_isize_read(inode));
-
- }
- }
- }
- ccc_object_size_unlock(obj);
- return result;
-}
-
-/*****************************************************************************
- *
- * Transfer operations.
- *
- */