4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * cl code shared between vvp and liblustre (and other Lustre clients in the
39 * Author: Nikita Danilov <nikita.danilov@sun.com>
42 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <libcfs/libcfs.h>
46 #include <linux/sched.h>
48 #include <linux/quotaops.h>
49 #include <linux/highmem.h>
50 #include <linux/pagemap.h>
51 #include <linux/rbtree.h>
54 #include <obd_support.h>
55 #include <lustre_fid.h>
56 #include <lustre_dlm.h>
57 #include <lustre_ver.h>
58 #include <lustre_mdc.h>
59 #include <cl_object.h>
61 #include "llite_internal.h"
63 static const struct cl_req_operations ccc_req_ops;
66 * ccc_ prefix stands for "Common Client Code".
69 static struct kmem_cache *ccc_lock_kmem;
70 static struct kmem_cache *ccc_object_kmem;
71 static struct kmem_cache *ccc_thread_kmem;
72 static struct kmem_cache *ccc_session_kmem;
73 static struct kmem_cache *ccc_req_kmem;
75 static struct lu_kmem_descr ccc_caches[] = {
77 .ckd_cache = &ccc_lock_kmem,
78 .ckd_name = "ccc_lock_kmem",
79 .ckd_size = sizeof (struct ccc_lock)
82 .ckd_cache = &ccc_object_kmem,
83 .ckd_name = "ccc_object_kmem",
84 .ckd_size = sizeof (struct ccc_object)
87 .ckd_cache = &ccc_thread_kmem,
88 .ckd_name = "ccc_thread_kmem",
89 .ckd_size = sizeof (struct ccc_thread_info),
92 .ckd_cache = &ccc_session_kmem,
93 .ckd_name = "ccc_session_kmem",
94 .ckd_size = sizeof (struct ccc_session)
97 .ckd_cache = &ccc_req_kmem,
98 .ckd_name = "ccc_req_kmem",
99 .ckd_size = sizeof (struct ccc_req)
106 /*****************************************************************************
108 * Vvp device and device type functions.
112 void *ccc_key_init(const struct lu_context *ctx, struct lu_context_key *key)
114 struct ccc_thread_info *info;
116 OBD_SLAB_ALLOC_PTR_GFP(info, ccc_thread_kmem, GFP_NOFS);
118 info = ERR_PTR(-ENOMEM);
122 void ccc_key_fini(const struct lu_context *ctx,
123 struct lu_context_key *key, void *data)
125 struct ccc_thread_info *info = data;
126 OBD_SLAB_FREE_PTR(info, ccc_thread_kmem);
129 void *ccc_session_key_init(const struct lu_context *ctx,
130 struct lu_context_key *key)
132 struct ccc_session *session;
134 OBD_SLAB_ALLOC_PTR_GFP(session, ccc_session_kmem, GFP_NOFS);
136 session = ERR_PTR(-ENOMEM);
140 void ccc_session_key_fini(const struct lu_context *ctx,
141 struct lu_context_key *key, void *data)
143 struct ccc_session *session = data;
144 OBD_SLAB_FREE_PTR(session, ccc_session_kmem);
147 struct lu_context_key ccc_key = {
148 .lct_tags = LCT_CL_THREAD,
149 .lct_init = ccc_key_init,
150 .lct_fini = ccc_key_fini
153 struct lu_context_key ccc_session_key = {
154 .lct_tags = LCT_SESSION,
155 .lct_init = ccc_session_key_init,
156 .lct_fini = ccc_session_key_fini
160 /* type constructor/destructor: ccc_type_{init,fini,start,stop}(). */
161 // LU_TYPE_INIT_FINI(ccc, &ccc_key, &ccc_session_key);
163 int ccc_device_init(const struct lu_env *env, struct lu_device *d,
164 const char *name, struct lu_device *next)
166 struct ccc_device *vdv;
171 vdv->cdv_next = lu2cl_dev(next);
173 LASSERT(d->ld_site != NULL && next->ld_type != NULL);
174 next->ld_site = d->ld_site;
175 rc = next->ld_type->ldt_ops->ldto_device_init(
176 env, next, next->ld_type->ldt_name, NULL);
179 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
184 struct lu_device *ccc_device_fini(const struct lu_env *env,
187 return cl2lu_dev(lu2ccc_dev(d)->cdv_next);
190 struct lu_device *ccc_device_alloc(const struct lu_env *env,
191 struct lu_device_type *t,
192 struct lustre_cfg *cfg,
193 const struct lu_device_operations *luops,
194 const struct cl_device_operations *clops)
196 struct ccc_device *vdv;
197 struct lu_device *lud;
198 struct cl_site *site;
204 RETURN(ERR_PTR(-ENOMEM));
206 lud = &vdv->cdv_cl.cd_lu_dev;
207 cl_device_init(&vdv->cdv_cl, t);
208 ccc2lu_dev(vdv)->ld_ops = luops;
209 vdv->cdv_cl.cd_ops = clops;
213 rc = cl_site_init(site, &vdv->cdv_cl);
215 rc = lu_site_init_finish(&site->cs_lu);
217 LASSERT(lud->ld_site == NULL);
218 CERROR("Cannot init lu_site, rc %d.\n", rc);
224 ccc_device_free(env, lud);
230 struct lu_device *ccc_device_free(const struct lu_env *env,
233 struct ccc_device *vdv = lu2ccc_dev(d);
234 struct cl_site *site = lu2cl_site(d->ld_site);
235 struct lu_device *next = cl2lu_dev(vdv->cdv_next);
237 if (d->ld_site != NULL) {
241 cl_device_fini(lu2cl_dev(d));
246 int ccc_req_init(const struct lu_env *env, struct cl_device *dev,
252 OBD_SLAB_ALLOC_PTR_GFP(vrq, ccc_req_kmem, GFP_NOFS);
254 cl_req_slice_add(req, &vrq->crq_cl, dev, &ccc_req_ops);
262 * An `emergency' environment used by ccc_inode_fini() when cl_env_get()
263 * fails. Access to this environment is serialized by ccc_inode_fini_guard
266 static struct lu_env *ccc_inode_fini_env = NULL;
269 * A mutex serializing calls to slp_inode_fini() under extreme memory
270 * pressure, when environments cannot be allocated.
272 static DEFINE_MUTEX(ccc_inode_fini_guard);
273 static int dummy_refcheck;
275 int ccc_global_init(struct lu_device_type *device_type)
279 result = lu_kmem_init(ccc_caches);
283 result = lu_device_type_init(device_type);
287 ccc_inode_fini_env = cl_env_alloc(&dummy_refcheck,
288 LCT_REMEMBER|LCT_NOREF);
289 if (IS_ERR(ccc_inode_fini_env)) {
290 result = PTR_ERR(ccc_inode_fini_env);
294 ccc_inode_fini_env->le_ctx.lc_cookie = 0x4;
297 lu_device_type_fini(device_type);
299 lu_kmem_fini(ccc_caches);
303 void ccc_global_fini(struct lu_device_type *device_type)
305 if (ccc_inode_fini_env != NULL) {
306 cl_env_put(ccc_inode_fini_env, &dummy_refcheck);
307 ccc_inode_fini_env = NULL;
309 lu_device_type_fini(device_type);
310 lu_kmem_fini(ccc_caches);
313 /*****************************************************************************
319 struct lu_object *ccc_object_alloc(const struct lu_env *env,
320 const struct lu_object_header *unused,
321 struct lu_device *dev,
322 const struct cl_object_operations *clops,
323 const struct lu_object_operations *luops)
325 struct ccc_object *vob;
326 struct lu_object *obj;
328 OBD_SLAB_ALLOC_PTR_GFP(vob, ccc_object_kmem, GFP_NOFS);
330 struct cl_object_header *hdr;
333 hdr = &vob->cob_header;
334 cl_object_header_init(hdr);
335 hdr->coh_page_bufsize = cfs_size_round(sizeof(struct cl_page));
337 lu_object_init(obj, &hdr->coh_lu, dev);
338 lu_object_add_top(&hdr->coh_lu, obj);
340 vob->cob_cl.co_ops = clops;
347 int ccc_object_init0(const struct lu_env *env,
348 struct ccc_object *vob,
349 const struct cl_object_conf *conf)
351 vob->cob_inode = conf->coc_inode;
352 atomic_set(&vob->cob_transient_pages, 0);
353 cl_object_page_init(&vob->cob_cl, sizeof(struct ccc_page));
357 int ccc_object_init(const struct lu_env *env, struct lu_object *obj,
358 const struct lu_object_conf *conf)
360 struct ccc_device *dev = lu2ccc_dev(obj->lo_dev);
361 struct ccc_object *vob = lu2ccc(obj);
362 struct lu_object *below;
363 struct lu_device *under;
366 under = &dev->cdv_next->cd_lu_dev;
367 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header, under);
369 const struct cl_object_conf *cconf;
371 cconf = lu2cl_conf(conf);
372 INIT_LIST_HEAD(&vob->cob_pending_list);
373 lu_object_add(obj, below);
374 result = ccc_object_init0(env, vob, cconf);
380 void ccc_object_free(const struct lu_env *env, struct lu_object *obj)
382 struct ccc_object *vob = lu2ccc(obj);
385 lu_object_header_fini(obj->lo_header);
386 OBD_SLAB_FREE_PTR(vob, ccc_object_kmem);
389 int ccc_lock_init(const struct lu_env *env,
390 struct cl_object *obj, struct cl_lock *lock,
391 const struct cl_io *unused,
392 const struct cl_lock_operations *lkops)
394 struct ccc_lock *clk;
397 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
399 OBD_SLAB_ALLOC_PTR_GFP(clk, ccc_lock_kmem, GFP_NOFS);
401 cl_lock_slice_add(lock, &clk->clk_cl, obj, lkops);
408 int ccc_attr_set(const struct lu_env *env, struct cl_object *obj,
409 const struct cl_attr *attr, unsigned valid)
414 int ccc_object_glimpse(const struct lu_env *env,
415 const struct cl_object *obj, struct ost_lvb *lvb)
417 struct inode *inode = ccc_object_inode(obj);
420 lvb->lvb_mtime = LTIME_S(inode->i_mtime);
421 lvb->lvb_atime = LTIME_S(inode->i_atime);
422 lvb->lvb_ctime = LTIME_S(inode->i_ctime);
424 * LU-417: Add dirty pages block count lest i_blocks reports 0, some
425 * "cp" or "tar" on remote node may think it's a completely sparse file
428 if (lvb->lvb_size > 0 && lvb->lvb_blocks == 0)
429 lvb->lvb_blocks = dirty_cnt(inode);
435 int ccc_conf_set(const struct lu_env *env, struct cl_object *obj,
436 const struct cl_object_conf *conf)
438 /* TODO: destroy all pages attached to this object. */
442 static void ccc_object_size_lock(struct cl_object *obj)
444 struct inode *inode = ccc_object_inode(obj);
446 ll_inode_size_lock(inode);
447 cl_object_attr_lock(obj);
450 static void ccc_object_size_unlock(struct cl_object *obj)
452 struct inode *inode = ccc_object_inode(obj);
454 cl_object_attr_unlock(obj);
455 ll_inode_size_unlock(inode);
458 /*****************************************************************************
464 int ccc_fail(const struct lu_env *env, const struct cl_page_slice *slice)
473 void ccc_transient_page_verify(const struct cl_page *page)
477 int ccc_transient_page_own(const struct lu_env *env,
478 const struct cl_page_slice *slice,
479 struct cl_io *unused,
482 ccc_transient_page_verify(slice->cpl_page);
486 void ccc_transient_page_assume(const struct lu_env *env,
487 const struct cl_page_slice *slice,
488 struct cl_io *unused)
490 ccc_transient_page_verify(slice->cpl_page);
493 void ccc_transient_page_unassume(const struct lu_env *env,
494 const struct cl_page_slice *slice,
495 struct cl_io *unused)
497 ccc_transient_page_verify(slice->cpl_page);
500 void ccc_transient_page_disown(const struct lu_env *env,
501 const struct cl_page_slice *slice,
502 struct cl_io *unused)
504 ccc_transient_page_verify(slice->cpl_page);
507 void ccc_transient_page_discard(const struct lu_env *env,
508 const struct cl_page_slice *slice,
509 struct cl_io *unused)
511 struct cl_page *page = slice->cpl_page;
513 ccc_transient_page_verify(slice->cpl_page);
516 * For transient pages, remove it from the radix tree.
518 cl_page_delete(env, page);
521 int ccc_transient_page_prep(const struct lu_env *env,
522 const struct cl_page_slice *slice,
523 struct cl_io *unused)
526 /* transient page should always be sent. */
530 /*****************************************************************************
536 void ccc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice)
538 struct ccc_lock *clk = cl2ccc_lock(slice);
539 OBD_SLAB_FREE_PTR(clk, ccc_lock_kmem);
542 int ccc_lock_enqueue(const struct lu_env *env,
543 const struct cl_lock_slice *slice,
544 struct cl_io *unused, struct cl_sync_io *anchor)
546 CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
550 /*****************************************************************************
556 void ccc_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
558 struct cl_io *io = ios->cis_io;
560 CLOBINVRNT(env, io->ci_obj, ccc_object_invariant(io->ci_obj));
563 int ccc_io_one_lock_index(const struct lu_env *env, struct cl_io *io,
564 __u32 enqflags, enum cl_lock_mode mode,
565 pgoff_t start, pgoff_t end)
567 struct ccc_io *cio = ccc_env_io(env);
568 struct cl_lock_descr *descr = &cio->cui_link.cill_descr;
569 struct cl_object *obj = io->ci_obj;
571 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
574 CDEBUG(D_VFSTRACE, "lock: %d [%lu, %lu]\n", mode, start, end);
576 memset(&cio->cui_link, 0, sizeof cio->cui_link);
578 if (cio->cui_fd && (cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
579 descr->cld_mode = CLM_GROUP;
580 descr->cld_gid = cio->cui_fd->fd_grouplock.cg_gid;
582 descr->cld_mode = mode;
584 descr->cld_obj = obj;
585 descr->cld_start = start;
586 descr->cld_end = end;
587 descr->cld_enq_flags = enqflags;
589 cl_io_lock_add(env, io, &cio->cui_link);
593 void ccc_io_update_iov(const struct lu_env *env,
594 struct ccc_io *cio, struct cl_io *io)
597 size_t size = io->u.ci_rw.crw_count;
599 cio->cui_iov_olen = 0;
600 if (!cl_is_normalio(env, io) || cio->cui_tot_nrsegs == 0)
603 for (i = 0; i < cio->cui_tot_nrsegs; i++) {
604 struct iovec *iv = &cio->cui_iov[i];
606 if (iv->iov_len < size)
609 if (iv->iov_len > size) {
610 cio->cui_iov_olen = iv->iov_len;
617 cio->cui_nrsegs = i + 1;
618 LASSERTF(cio->cui_tot_nrsegs >= cio->cui_nrsegs,
619 "tot_nrsegs: %lu, nrsegs: %lu\n",
620 cio->cui_tot_nrsegs, cio->cui_nrsegs);
623 int ccc_io_one_lock(const struct lu_env *env, struct cl_io *io,
624 __u32 enqflags, enum cl_lock_mode mode,
625 loff_t start, loff_t end)
627 struct cl_object *obj = io->ci_obj;
628 return ccc_io_one_lock_index(env, io, enqflags, mode,
629 cl_index(obj, start), cl_index(obj, end));
632 void ccc_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
634 CLOBINVRNT(env, ios->cis_io->ci_obj,
635 ccc_object_invariant(ios->cis_io->ci_obj));
638 void ccc_io_advance(const struct lu_env *env,
639 const struct cl_io_slice *ios,
642 struct ccc_io *cio = cl2ccc_io(env, ios);
643 struct cl_io *io = ios->cis_io;
644 struct cl_object *obj = ios->cis_io->ci_obj;
646 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
648 if (!cl_is_normalio(env, io))
651 LASSERT(cio->cui_tot_nrsegs >= cio->cui_nrsegs);
652 LASSERT(cio->cui_tot_count >= nob);
654 cio->cui_iov += cio->cui_nrsegs;
655 cio->cui_tot_nrsegs -= cio->cui_nrsegs;
656 cio->cui_tot_count -= nob;
659 if (cio->cui_iov_olen > 0) {
663 cio->cui_tot_nrsegs++;
664 iv = &cio->cui_iov[0];
665 if (io->ci_continue) {
666 iv->iov_base += iv->iov_len;
667 LASSERT(cio->cui_iov_olen > iv->iov_len);
668 iv->iov_len = cio->cui_iov_olen - iv->iov_len;
670 /* restore the iov_len, in case of restart io. */
671 iv->iov_len = cio->cui_iov_olen;
673 cio->cui_iov_olen = 0;
678 * Helper function that if necessary adjusts file size (inode->i_size), when
679 * position at the offset \a pos is accessed. File size can be arbitrary stale
680 * on a Lustre client, but client at least knows KMS. If accessed area is
681 * inside [0, KMS], set file size to KMS, otherwise glimpse file size.
683 * Locking: cl_isize_lock is used to serialize changes to inode size and to
684 * protect consistency between inode size and cl_object
685 * attributes. cl_object_size_lock() protects consistency between cl_attr's of
686 * top-object and sub-objects.
688 int ccc_prep_size(const struct lu_env *env, struct cl_object *obj,
689 struct cl_io *io, loff_t start, size_t count, int *exceed)
691 struct cl_attr *attr = ccc_env_thread_attr(env);
692 struct inode *inode = ccc_object_inode(obj);
693 loff_t pos = start + count - 1;
698 * Consistency guarantees: following possibilities exist for the
699 * relation between region being accessed and real file size at this
702 * (A): the region is completely inside of the file;
704 * (B-x): x bytes of region are inside of the file, the rest is
707 * (C): the region is completely outside of the file.
709 * This classification is stable under DLM lock already acquired by
710 * the caller, because to change the class, other client has to take
711 * DLM lock conflicting with our lock. Also, any updates to ->i_size
712 * by other threads on this client are serialized by
713 * ll_inode_size_lock(). This guarantees that short reads are handled
714 * correctly in the face of concurrent writes and truncates.
716 ccc_object_size_lock(obj);
717 result = cl_object_attr_get(env, obj, attr);
722 * A glimpse is necessary to determine whether we
723 * return a short read (B) or some zeroes at the end
726 ccc_object_size_unlock(obj);
727 result = cl_glimpse_lock(env, io, inode, obj, 0);
728 if (result == 0 && exceed != NULL) {
729 /* If objective page index exceed end-of-file
730 * page index, return directly. Do not expect
731 * kernel will check such case correctly.
732 * linux-2.6.18-128.1.1 miss to do that.
734 loff_t size = i_size_read(inode);
735 unsigned long cur_index = start >>
738 if ((size == 0 && cur_index != 0) ||
739 (((size - 1) >> PAGE_CACHE_SHIFT) <
746 * region is within kms and, hence, within real file
747 * size (A). We need to increase i_size to cover the
748 * read region so that generic_file_read() will do its
749 * job, but that doesn't mean the kms size is
750 * _correct_, it is only the _minimum_ size. If
751 * someone does a stat they will get the correct size
752 * which will always be >= the kms value here.
755 if (i_size_read(inode) < kms) {
756 i_size_write(inode, kms);
758 DFID" updating i_size "LPU64"\n",
759 PFID(lu_object_fid(&obj->co_lu)),
760 (__u64)i_size_read(inode));
765 ccc_object_size_unlock(obj);
769 /*****************************************************************************
771 * Transfer operations.
775 void ccc_req_completion(const struct lu_env *env,
776 const struct cl_req_slice *slice, int ioret)
781 cl_stats_tally(slice->crs_dev, slice->crs_req->crq_type, ioret);
783 vrq = cl2ccc_req(slice);
784 OBD_SLAB_FREE_PTR(vrq, ccc_req_kmem);
788 * Implementation of struct cl_req_operations::cro_attr_set() for ccc
789 * layer. ccc is responsible for
807 void ccc_req_attr_set(const struct lu_env *env,
808 const struct cl_req_slice *slice,
809 const struct cl_object *obj,
810 struct cl_req_attr *attr, obd_valid flags)
814 obd_flag valid_flags;
817 inode = ccc_object_inode(obj);
818 valid_flags = OBD_MD_FLTYPE;
820 if ((flags & OBD_MD_FLOSSCAPA) != 0) {
821 LASSERT(attr->cra_capa == NULL);
822 attr->cra_capa = cl_capa_lookup(inode,
823 slice->crs_req->crq_type);
826 if (slice->crs_req->crq_type == CRT_WRITE) {
827 if (flags & OBD_MD_FLEPOCH) {
828 oa->o_valid |= OBD_MD_FLEPOCH;
829 oa->o_ioepoch = ll_i2info(inode)->lli_ioepoch;
830 valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
831 OBD_MD_FLUID | OBD_MD_FLGID;
834 obdo_from_inode(oa, inode, valid_flags & flags);
835 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
836 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_INVALID_PFID))
838 memcpy(attr->cra_jobid, ll_i2info(inode)->lli_jobid,
842 static const struct cl_req_operations ccc_req_ops = {
843 .cro_attr_set = ccc_req_attr_set,
844 .cro_completion = ccc_req_completion
847 int cl_setattr_ost(struct inode *inode, const struct iattr *attr,
848 struct obd_capa *capa)
857 env = cl_env_get(&refcheck);
859 RETURN(PTR_ERR(env));
861 io = ccc_env_thread_io(env);
862 io->ci_obj = ll_i2info(inode)->lli_clob;
864 io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime);
865 io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime);
866 io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime);
867 io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size;
868 io->u.ci_setattr.sa_valid = attr->ia_valid;
869 io->u.ci_setattr.sa_capa = capa;
872 if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0) {
873 struct ccc_io *cio = ccc_env_io(env);
875 if (attr->ia_valid & ATTR_FILE)
876 /* populate the file descriptor for ftruncate to honor
877 * group lock - see LU-787 */
878 cio->cui_fd = LUSTRE_FPRIVATE(attr->ia_file);
880 result = cl_io_loop(env, io);
882 result = io->ci_result;
885 if (unlikely(io->ci_need_restart))
887 /* HSM import case: file is released, cannot be restored
888 * no need to fail except if restore registration failed
890 if (result == -ENODATA && io->ci_restore_needed &&
891 io->ci_result != -ENODATA)
893 cl_env_put(env, &refcheck);
897 /*****************************************************************************
903 struct lu_device *ccc2lu_dev(struct ccc_device *vdv)
905 return &vdv->cdv_cl.cd_lu_dev;
908 struct ccc_device *lu2ccc_dev(const struct lu_device *d)
910 return container_of0(d, struct ccc_device, cdv_cl.cd_lu_dev);
913 struct ccc_device *cl2ccc_dev(const struct cl_device *d)
915 return container_of0(d, struct ccc_device, cdv_cl);
918 struct lu_object *ccc2lu(struct ccc_object *vob)
920 return &vob->cob_cl.co_lu;
923 struct ccc_object *lu2ccc(const struct lu_object *obj)
925 return container_of0(obj, struct ccc_object, cob_cl.co_lu);
928 struct ccc_object *cl2ccc(const struct cl_object *obj)
930 return container_of0(obj, struct ccc_object, cob_cl);
933 struct ccc_lock *cl2ccc_lock(const struct cl_lock_slice *slice)
935 return container_of(slice, struct ccc_lock, clk_cl);
938 struct ccc_io *cl2ccc_io(const struct lu_env *env,
939 const struct cl_io_slice *slice)
943 cio = container_of(slice, struct ccc_io, cui_cl);
944 LASSERT(cio == ccc_env_io(env));
948 struct ccc_req *cl2ccc_req(const struct cl_req_slice *slice)
950 return container_of0(slice, struct ccc_req, crq_cl);
953 struct page *cl2vm_page(const struct cl_page_slice *slice)
955 return cl2ccc_page(slice)->cpg_page;
958 /*****************************************************************************
963 int ccc_object_invariant(const struct cl_object *obj)
965 struct inode *inode = ccc_object_inode(obj);
966 struct ll_inode_info *lli = ll_i2info(inode);
968 return (S_ISREG(inode->i_mode) || inode->i_mode == 0) &&
969 lli->lli_clob == obj;
972 struct inode *ccc_object_inode(const struct cl_object *obj)
974 return cl2ccc(obj)->cob_inode;
978 * Returns a pointer to cl_page associated with \a vmpage, without acquiring
979 * additional reference to the resulting page. This is an unsafe version of
980 * cl_vmpage_page() that can only be used under vmpage lock.
982 struct cl_page *ccc_vmpage_page_transient(struct page *vmpage)
984 KLASSERT(PageLocked(vmpage));
985 return (struct cl_page *)vmpage->private;
989 * Initialize or update CLIO structures for regular files when new
990 * meta-data arrives from the server.
992 * \param inode regular file inode
993 * \param md new file metadata from MDS
994 * - allocates cl_object if necessary,
995 * - updated layout, if object was already here.
997 int cl_file_inode_init(struct inode *inode, struct lustre_md *md)
1000 struct ll_inode_info *lli;
1001 struct cl_object *clob;
1002 struct lu_site *site;
1004 struct cl_object_conf conf = {
1013 LASSERT(md->body->mbo_valid & OBD_MD_FLID);
1014 LASSERT(S_ISREG(inode->i_mode));
1016 env = cl_env_get(&refcheck);
1018 return PTR_ERR(env);
1020 site = ll_i2sbi(inode)->ll_site;
1021 lli = ll_i2info(inode);
1022 fid = &lli->lli_fid;
1023 LASSERT(fid_is_sane(fid));
1025 if (lli->lli_clob == NULL) {
1026 /* clob is slave of inode, empty lli_clob means for new inode,
1027 * there is no clob in cache with the given fid, so it is
1028 * unnecessary to perform lookup-alloc-lookup-insert, just
1029 * alloc and insert directly. */
1030 LASSERT(inode->i_state & I_NEW);
1031 conf.coc_lu.loc_flags = LOC_F_NEW;
1032 clob = cl_object_find(env, lu2cl_dev(site->ls_top_dev),
1034 if (!IS_ERR(clob)) {
1036 * No locking is necessary, as new inode is
1037 * locked by I_NEW bit.
1039 lli->lli_clob = clob;
1040 lli->lli_has_smd = lsm_has_objects(md->lsm);
1041 lu_object_ref_add(&clob->co_lu, "inode", inode);
1043 result = PTR_ERR(clob);
1045 result = cl_conf_set(env, lli->lli_clob, &conf);
1048 cl_env_put(env, &refcheck);
1051 CERROR("Failure to initialize cl object "DFID": %d\n",
1057 * Wait for others drop their references of the object at first, then we drop
1058 * the last one, which will lead to the object be destroyed immediately.
1059 * Must be called after cl_object_kill() against this object.
1061 * The reason we want to do this is: destroying top object will wait for sub
1062 * objects being destroyed first, so we can't let bottom layer (e.g. from ASTs)
1063 * to initiate top object destroying which may deadlock. See bz22520.
1065 static void cl_object_put_last(struct lu_env *env, struct cl_object *obj)
1067 struct lu_object_header *header = obj->co_lu.lo_header;
1068 wait_queue_t waiter;
1070 if (unlikely(atomic_read(&header->loh_ref) != 1)) {
1071 struct lu_site *site = obj->co_lu.lo_dev->ld_site;
1072 struct lu_site_bkt_data *bkt;
1074 bkt = lu_site_bkt_from_fid(site, &header->loh_fid);
1076 init_waitqueue_entry_current(&waiter);
1077 add_wait_queue(&bkt->lsb_marche_funebre, &waiter);
1080 set_current_state(TASK_UNINTERRUPTIBLE);
1081 if (atomic_read(&header->loh_ref) == 1)
1083 waitq_wait(&waiter, TASK_UNINTERRUPTIBLE);
1086 set_current_state(TASK_RUNNING);
1087 remove_wait_queue(&bkt->lsb_marche_funebre, &waiter);
1090 cl_object_put(env, obj);
1093 void cl_inode_fini(struct inode *inode)
1096 struct ll_inode_info *lli = ll_i2info(inode);
1097 struct cl_object *clob = lli->lli_clob;
1104 cookie = cl_env_reenter();
1105 env = cl_env_get(&refcheck);
1106 emergency = IS_ERR(env);
1108 mutex_lock(&ccc_inode_fini_guard);
1109 LASSERT(ccc_inode_fini_env != NULL);
1110 cl_env_implant(ccc_inode_fini_env, &refcheck);
1111 env = ccc_inode_fini_env;
1114 * cl_object cache is a slave to inode cache (which, in turn
1115 * is a slave to dentry cache), don't keep cl_object in memory
1116 * when its master is evicted.
1118 cl_object_kill(env, clob);
1119 lu_object_ref_del(&clob->co_lu, "inode", inode);
1120 cl_object_put_last(env, clob);
1121 lli->lli_clob = NULL;
1123 cl_env_unplant(ccc_inode_fini_env, &refcheck);
1124 mutex_unlock(&ccc_inode_fini_guard);
1126 cl_env_put(env, &refcheck);
1127 cl_env_reexit(cookie);
1132 * return IF_* type for given lu_dirent entry.
1133 * IF_* flag shld be converted to particular OS file type in
1134 * platform llite module.
1136 __u16 ll_dirent_type_get(struct lu_dirent *ent)
1139 struct luda_type *lt;
1142 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
1143 const unsigned align = sizeof(struct luda_type) - 1;
1145 len = le16_to_cpu(ent->lde_namelen);
1146 len = (len + align) & ~align;
1147 lt = (void *)ent->lde_name + len;
1148 type = IFTODT(le16_to_cpu(lt->lt_type));
1154 * build inode number from passed @fid */
1155 __u64 cl_fid_build_ino(const struct lu_fid *fid, int api32)
1157 if (BITS_PER_LONG == 32 || api32)
1158 RETURN(fid_flatten32(fid));
1160 RETURN(fid_flatten(fid));
1164 * build inode generation from passed @fid. If our FID overflows the 32-bit
1165 * inode number then return a non-zero generation to distinguish them. */
1166 __u32 cl_fid_build_gen(const struct lu_fid *fid)
1171 if (fid_is_igif(fid)) {
1172 gen = lu_igif_gen(fid);
1176 gen = (fid_flatten(fid) >> 32);
1180 /* lsm is unreliable after hsm implementation as layout can be changed at
1181 * any time. This is only to support old, non-clio-ized interfaces. It will
1182 * cause deadlock if clio operations are called with this extra layout refcount
1183 * because in case the layout changed during the IO, ll_layout_refresh() will
1184 * have to wait for the refcount to become zero to destroy the older layout.
1186 * Notice that the lsm returned by this function may not be valid unless called
1187 * inside layout lock - MDS_INODELOCK_LAYOUT. */
1188 struct lov_stripe_md *ccc_inode_lsm_get(struct inode *inode)
1190 return lov_lsm_get(ll_i2info(inode)->lli_clob);
1193 void inline ccc_inode_lsm_put(struct inode *inode, struct lov_stripe_md *lsm)
1195 lov_lsm_put(ll_i2info(inode)->lli_clob, lsm);