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
159 int ccc_req_init(const struct lu_env *env, struct cl_device *dev,
165 OBD_SLAB_ALLOC_PTR_GFP(vrq, ccc_req_kmem, GFP_NOFS);
167 cl_req_slice_add(req, &vrq->crq_cl, dev, &ccc_req_ops);
175 * An `emergency' environment used by ccc_inode_fini() when cl_env_get()
176 * fails. Access to this environment is serialized by ccc_inode_fini_guard
179 static struct lu_env *ccc_inode_fini_env = NULL;
182 * A mutex serializing calls to slp_inode_fini() under extreme memory
183 * pressure, when environments cannot be allocated.
185 static DEFINE_MUTEX(ccc_inode_fini_guard);
186 static int dummy_refcheck;
188 int ccc_global_init(struct lu_device_type *device_type)
192 result = lu_kmem_init(ccc_caches);
196 result = lu_device_type_init(device_type);
200 ccc_inode_fini_env = cl_env_alloc(&dummy_refcheck,
201 LCT_REMEMBER|LCT_NOREF);
202 if (IS_ERR(ccc_inode_fini_env)) {
203 result = PTR_ERR(ccc_inode_fini_env);
207 ccc_inode_fini_env->le_ctx.lc_cookie = 0x4;
210 lu_device_type_fini(device_type);
212 lu_kmem_fini(ccc_caches);
216 void ccc_global_fini(struct lu_device_type *device_type)
218 if (ccc_inode_fini_env != NULL) {
219 cl_env_put(ccc_inode_fini_env, &dummy_refcheck);
220 ccc_inode_fini_env = NULL;
222 lu_device_type_fini(device_type);
223 lu_kmem_fini(ccc_caches);
226 /*****************************************************************************
232 struct lu_object *ccc_object_alloc(const struct lu_env *env,
233 const struct lu_object_header *unused,
234 struct lu_device *dev,
235 const struct cl_object_operations *clops,
236 const struct lu_object_operations *luops)
238 struct ccc_object *vob;
239 struct lu_object *obj;
241 OBD_SLAB_ALLOC_PTR_GFP(vob, ccc_object_kmem, GFP_NOFS);
243 struct cl_object_header *hdr;
246 hdr = &vob->cob_header;
247 cl_object_header_init(hdr);
248 hdr->coh_page_bufsize = cfs_size_round(sizeof(struct cl_page));
250 lu_object_init(obj, &hdr->coh_lu, dev);
251 lu_object_add_top(&hdr->coh_lu, obj);
253 vob->cob_cl.co_ops = clops;
260 int ccc_object_init0(const struct lu_env *env,
261 struct ccc_object *vob,
262 const struct cl_object_conf *conf)
264 vob->cob_inode = conf->coc_inode;
265 atomic_set(&vob->cob_transient_pages, 0);
266 cl_object_page_init(&vob->cob_cl, sizeof(struct ccc_page));
270 int ccc_object_init(const struct lu_env *env, struct lu_object *obj,
271 const struct lu_object_conf *conf)
273 struct vvp_device *dev = lu2vvp_dev(obj->lo_dev);
274 struct ccc_object *vob = lu2ccc(obj);
275 struct lu_object *below;
276 struct lu_device *under;
279 under = &dev->vdv_next->cd_lu_dev;
280 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header, under);
282 const struct cl_object_conf *cconf;
284 cconf = lu2cl_conf(conf);
285 INIT_LIST_HEAD(&vob->cob_pending_list);
286 lu_object_add(obj, below);
287 result = ccc_object_init0(env, vob, cconf);
293 void ccc_object_free(const struct lu_env *env, struct lu_object *obj)
295 struct ccc_object *vob = lu2ccc(obj);
298 lu_object_header_fini(obj->lo_header);
299 OBD_SLAB_FREE_PTR(vob, ccc_object_kmem);
302 int ccc_lock_init(const struct lu_env *env,
303 struct cl_object *obj, struct cl_lock *lock,
304 const struct cl_io *unused,
305 const struct cl_lock_operations *lkops)
307 struct ccc_lock *clk;
310 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
312 OBD_SLAB_ALLOC_PTR_GFP(clk, ccc_lock_kmem, GFP_NOFS);
314 cl_lock_slice_add(lock, &clk->clk_cl, obj, lkops);
321 int ccc_attr_set(const struct lu_env *env, struct cl_object *obj,
322 const struct cl_attr *attr, unsigned valid)
327 int ccc_object_glimpse(const struct lu_env *env,
328 const struct cl_object *obj, struct ost_lvb *lvb)
330 struct inode *inode = ccc_object_inode(obj);
333 lvb->lvb_mtime = LTIME_S(inode->i_mtime);
334 lvb->lvb_atime = LTIME_S(inode->i_atime);
335 lvb->lvb_ctime = LTIME_S(inode->i_ctime);
337 * LU-417: Add dirty pages block count lest i_blocks reports 0, some
338 * "cp" or "tar" on remote node may think it's a completely sparse file
341 if (lvb->lvb_size > 0 && lvb->lvb_blocks == 0)
342 lvb->lvb_blocks = dirty_cnt(inode);
348 int ccc_conf_set(const struct lu_env *env, struct cl_object *obj,
349 const struct cl_object_conf *conf)
351 /* TODO: destroy all pages attached to this object. */
355 static void ccc_object_size_lock(struct cl_object *obj)
357 struct inode *inode = ccc_object_inode(obj);
359 ll_inode_size_lock(inode);
360 cl_object_attr_lock(obj);
363 static void ccc_object_size_unlock(struct cl_object *obj)
365 struct inode *inode = ccc_object_inode(obj);
367 cl_object_attr_unlock(obj);
368 ll_inode_size_unlock(inode);
371 /*****************************************************************************
377 int ccc_fail(const struct lu_env *env, const struct cl_page_slice *slice)
386 void ccc_transient_page_verify(const struct cl_page *page)
390 int ccc_transient_page_own(const struct lu_env *env,
391 const struct cl_page_slice *slice,
392 struct cl_io *unused,
395 ccc_transient_page_verify(slice->cpl_page);
399 void ccc_transient_page_assume(const struct lu_env *env,
400 const struct cl_page_slice *slice,
401 struct cl_io *unused)
403 ccc_transient_page_verify(slice->cpl_page);
406 void ccc_transient_page_unassume(const struct lu_env *env,
407 const struct cl_page_slice *slice,
408 struct cl_io *unused)
410 ccc_transient_page_verify(slice->cpl_page);
413 void ccc_transient_page_disown(const struct lu_env *env,
414 const struct cl_page_slice *slice,
415 struct cl_io *unused)
417 ccc_transient_page_verify(slice->cpl_page);
420 void ccc_transient_page_discard(const struct lu_env *env,
421 const struct cl_page_slice *slice,
422 struct cl_io *unused)
424 struct cl_page *page = slice->cpl_page;
426 ccc_transient_page_verify(slice->cpl_page);
429 * For transient pages, remove it from the radix tree.
431 cl_page_delete(env, page);
434 int ccc_transient_page_prep(const struct lu_env *env,
435 const struct cl_page_slice *slice,
436 struct cl_io *unused)
439 /* transient page should always be sent. */
443 /*****************************************************************************
449 void ccc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice)
451 struct ccc_lock *clk = cl2ccc_lock(slice);
452 OBD_SLAB_FREE_PTR(clk, ccc_lock_kmem);
455 int ccc_lock_enqueue(const struct lu_env *env,
456 const struct cl_lock_slice *slice,
457 struct cl_io *unused, struct cl_sync_io *anchor)
459 CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
463 /*****************************************************************************
469 void ccc_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
471 struct cl_io *io = ios->cis_io;
473 CLOBINVRNT(env, io->ci_obj, ccc_object_invariant(io->ci_obj));
476 int ccc_io_one_lock_index(const struct lu_env *env, struct cl_io *io,
477 __u32 enqflags, enum cl_lock_mode mode,
478 pgoff_t start, pgoff_t end)
480 struct ccc_io *cio = ccc_env_io(env);
481 struct cl_lock_descr *descr = &cio->cui_link.cill_descr;
482 struct cl_object *obj = io->ci_obj;
484 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
487 CDEBUG(D_VFSTRACE, "lock: %d [%lu, %lu]\n", mode, start, end);
489 memset(&cio->cui_link, 0, sizeof cio->cui_link);
491 if (cio->cui_fd && (cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
492 descr->cld_mode = CLM_GROUP;
493 descr->cld_gid = cio->cui_fd->fd_grouplock.cg_gid;
495 descr->cld_mode = mode;
497 descr->cld_obj = obj;
498 descr->cld_start = start;
499 descr->cld_end = end;
500 descr->cld_enq_flags = enqflags;
502 cl_io_lock_add(env, io, &cio->cui_link);
506 void ccc_io_update_iov(const struct lu_env *env,
507 struct ccc_io *cio, struct cl_io *io)
510 size_t size = io->u.ci_rw.crw_count;
512 cio->cui_iov_olen = 0;
513 if (!cl_is_normalio(env, io) || cio->cui_tot_nrsegs == 0)
516 for (i = 0; i < cio->cui_tot_nrsegs; i++) {
517 struct iovec *iv = &cio->cui_iov[i];
519 if (iv->iov_len < size)
522 if (iv->iov_len > size) {
523 cio->cui_iov_olen = iv->iov_len;
530 cio->cui_nrsegs = i + 1;
531 LASSERTF(cio->cui_tot_nrsegs >= cio->cui_nrsegs,
532 "tot_nrsegs: %lu, nrsegs: %lu\n",
533 cio->cui_tot_nrsegs, cio->cui_nrsegs);
536 int ccc_io_one_lock(const struct lu_env *env, struct cl_io *io,
537 __u32 enqflags, enum cl_lock_mode mode,
538 loff_t start, loff_t end)
540 struct cl_object *obj = io->ci_obj;
541 return ccc_io_one_lock_index(env, io, enqflags, mode,
542 cl_index(obj, start), cl_index(obj, end));
545 void ccc_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
547 CLOBINVRNT(env, ios->cis_io->ci_obj,
548 ccc_object_invariant(ios->cis_io->ci_obj));
551 void ccc_io_advance(const struct lu_env *env,
552 const struct cl_io_slice *ios,
555 struct ccc_io *cio = cl2ccc_io(env, ios);
556 struct cl_io *io = ios->cis_io;
557 struct cl_object *obj = ios->cis_io->ci_obj;
559 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
561 if (!cl_is_normalio(env, io))
564 LASSERT(cio->cui_tot_nrsegs >= cio->cui_nrsegs);
565 LASSERT(cio->cui_tot_count >= nob);
567 cio->cui_iov += cio->cui_nrsegs;
568 cio->cui_tot_nrsegs -= cio->cui_nrsegs;
569 cio->cui_tot_count -= nob;
572 if (cio->cui_iov_olen > 0) {
576 cio->cui_tot_nrsegs++;
577 iv = &cio->cui_iov[0];
578 if (io->ci_continue) {
579 iv->iov_base += iv->iov_len;
580 LASSERT(cio->cui_iov_olen > iv->iov_len);
581 iv->iov_len = cio->cui_iov_olen - iv->iov_len;
583 /* restore the iov_len, in case of restart io. */
584 iv->iov_len = cio->cui_iov_olen;
586 cio->cui_iov_olen = 0;
591 * Helper function that if necessary adjusts file size (inode->i_size), when
592 * position at the offset \a pos is accessed. File size can be arbitrary stale
593 * on a Lustre client, but client at least knows KMS. If accessed area is
594 * inside [0, KMS], set file size to KMS, otherwise glimpse file size.
596 * Locking: cl_isize_lock is used to serialize changes to inode size and to
597 * protect consistency between inode size and cl_object
598 * attributes. cl_object_size_lock() protects consistency between cl_attr's of
599 * top-object and sub-objects.
601 int ccc_prep_size(const struct lu_env *env, struct cl_object *obj,
602 struct cl_io *io, loff_t start, size_t count, int *exceed)
604 struct cl_attr *attr = ccc_env_thread_attr(env);
605 struct inode *inode = ccc_object_inode(obj);
606 loff_t pos = start + count - 1;
611 * Consistency guarantees: following possibilities exist for the
612 * relation between region being accessed and real file size at this
615 * (A): the region is completely inside of the file;
617 * (B-x): x bytes of region are inside of the file, the rest is
620 * (C): the region is completely outside of the file.
622 * This classification is stable under DLM lock already acquired by
623 * the caller, because to change the class, other client has to take
624 * DLM lock conflicting with our lock. Also, any updates to ->i_size
625 * by other threads on this client are serialized by
626 * ll_inode_size_lock(). This guarantees that short reads are handled
627 * correctly in the face of concurrent writes and truncates.
629 ccc_object_size_lock(obj);
630 result = cl_object_attr_get(env, obj, attr);
635 * A glimpse is necessary to determine whether we
636 * return a short read (B) or some zeroes at the end
639 ccc_object_size_unlock(obj);
640 result = cl_glimpse_lock(env, io, inode, obj, 0);
641 if (result == 0 && exceed != NULL) {
642 /* If objective page index exceed end-of-file
643 * page index, return directly. Do not expect
644 * kernel will check such case correctly.
645 * linux-2.6.18-128.1.1 miss to do that.
647 loff_t size = i_size_read(inode);
648 unsigned long cur_index = start >>
651 if ((size == 0 && cur_index != 0) ||
652 (((size - 1) >> PAGE_CACHE_SHIFT) <
659 * region is within kms and, hence, within real file
660 * size (A). We need to increase i_size to cover the
661 * read region so that generic_file_read() will do its
662 * job, but that doesn't mean the kms size is
663 * _correct_, it is only the _minimum_ size. If
664 * someone does a stat they will get the correct size
665 * which will always be >= the kms value here.
668 if (i_size_read(inode) < kms) {
669 i_size_write(inode, kms);
671 DFID" updating i_size "LPU64"\n",
672 PFID(lu_object_fid(&obj->co_lu)),
673 (__u64)i_size_read(inode));
678 ccc_object_size_unlock(obj);
682 /*****************************************************************************
684 * Transfer operations.
688 void ccc_req_completion(const struct lu_env *env,
689 const struct cl_req_slice *slice, int ioret)
694 cl_stats_tally(slice->crs_dev, slice->crs_req->crq_type, ioret);
696 vrq = cl2ccc_req(slice);
697 OBD_SLAB_FREE_PTR(vrq, ccc_req_kmem);
701 * Implementation of struct cl_req_operations::cro_attr_set() for ccc
702 * layer. ccc is responsible for
720 void ccc_req_attr_set(const struct lu_env *env,
721 const struct cl_req_slice *slice,
722 const struct cl_object *obj,
723 struct cl_req_attr *attr, obd_valid flags)
727 obd_flag valid_flags;
730 inode = ccc_object_inode(obj);
731 valid_flags = OBD_MD_FLTYPE;
733 if ((flags & OBD_MD_FLOSSCAPA) != 0) {
734 LASSERT(attr->cra_capa == NULL);
735 attr->cra_capa = cl_capa_lookup(inode,
736 slice->crs_req->crq_type);
739 if (slice->crs_req->crq_type == CRT_WRITE) {
740 if (flags & OBD_MD_FLEPOCH) {
741 oa->o_valid |= OBD_MD_FLEPOCH;
742 oa->o_ioepoch = ll_i2info(inode)->lli_ioepoch;
743 valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
744 OBD_MD_FLUID | OBD_MD_FLGID;
747 obdo_from_inode(oa, inode, valid_flags & flags);
748 obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
749 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_INVALID_PFID))
751 memcpy(attr->cra_jobid, ll_i2info(inode)->lli_jobid,
755 static const struct cl_req_operations ccc_req_ops = {
756 .cro_attr_set = ccc_req_attr_set,
757 .cro_completion = ccc_req_completion
760 int cl_setattr_ost(struct inode *inode, const struct iattr *attr,
761 struct obd_capa *capa)
770 env = cl_env_get(&refcheck);
772 RETURN(PTR_ERR(env));
774 io = ccc_env_thread_io(env);
775 io->ci_obj = ll_i2info(inode)->lli_clob;
777 io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime);
778 io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime);
779 io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime);
780 io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size;
781 io->u.ci_setattr.sa_valid = attr->ia_valid;
782 io->u.ci_setattr.sa_capa = capa;
785 if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0) {
786 struct ccc_io *cio = ccc_env_io(env);
788 if (attr->ia_valid & ATTR_FILE)
789 /* populate the file descriptor for ftruncate to honor
790 * group lock - see LU-787 */
791 cio->cui_fd = LUSTRE_FPRIVATE(attr->ia_file);
793 result = cl_io_loop(env, io);
795 result = io->ci_result;
798 if (unlikely(io->ci_need_restart))
800 /* HSM import case: file is released, cannot be restored
801 * no need to fail except if restore registration failed
803 if (result == -ENODATA && io->ci_restore_needed &&
804 io->ci_result != -ENODATA)
806 cl_env_put(env, &refcheck);
810 /*****************************************************************************
816 struct lu_object *ccc2lu(struct ccc_object *vob)
818 return &vob->cob_cl.co_lu;
821 struct ccc_object *lu2ccc(const struct lu_object *obj)
823 return container_of0(obj, struct ccc_object, cob_cl.co_lu);
826 struct ccc_object *cl2ccc(const struct cl_object *obj)
828 return container_of0(obj, struct ccc_object, cob_cl);
831 struct ccc_lock *cl2ccc_lock(const struct cl_lock_slice *slice)
833 return container_of(slice, struct ccc_lock, clk_cl);
836 struct ccc_io *cl2ccc_io(const struct lu_env *env,
837 const struct cl_io_slice *slice)
841 cio = container_of(slice, struct ccc_io, cui_cl);
842 LASSERT(cio == ccc_env_io(env));
846 struct ccc_req *cl2ccc_req(const struct cl_req_slice *slice)
848 return container_of0(slice, struct ccc_req, crq_cl);
851 struct page *cl2vm_page(const struct cl_page_slice *slice)
853 return cl2ccc_page(slice)->cpg_page;
856 /*****************************************************************************
861 int ccc_object_invariant(const struct cl_object *obj)
863 struct inode *inode = ccc_object_inode(obj);
864 struct ll_inode_info *lli = ll_i2info(inode);
866 return (S_ISREG(inode->i_mode) || inode->i_mode == 0) &&
867 lli->lli_clob == obj;
870 struct inode *ccc_object_inode(const struct cl_object *obj)
872 return cl2ccc(obj)->cob_inode;
876 * Returns a pointer to cl_page associated with \a vmpage, without acquiring
877 * additional reference to the resulting page. This is an unsafe version of
878 * cl_vmpage_page() that can only be used under vmpage lock.
880 struct cl_page *ccc_vmpage_page_transient(struct page *vmpage)
882 KLASSERT(PageLocked(vmpage));
883 return (struct cl_page *)vmpage->private;
887 * Initialize or update CLIO structures for regular files when new
888 * meta-data arrives from the server.
890 * \param inode regular file inode
891 * \param md new file metadata from MDS
892 * - allocates cl_object if necessary,
893 * - updated layout, if object was already here.
895 int cl_file_inode_init(struct inode *inode, struct lustre_md *md)
898 struct ll_inode_info *lli;
899 struct cl_object *clob;
900 struct lu_site *site;
902 struct cl_object_conf conf = {
911 LASSERT(md->body->mbo_valid & OBD_MD_FLID);
912 LASSERT(S_ISREG(inode->i_mode));
914 env = cl_env_get(&refcheck);
918 site = ll_i2sbi(inode)->ll_site;
919 lli = ll_i2info(inode);
921 LASSERT(fid_is_sane(fid));
923 if (lli->lli_clob == NULL) {
924 /* clob is slave of inode, empty lli_clob means for new inode,
925 * there is no clob in cache with the given fid, so it is
926 * unnecessary to perform lookup-alloc-lookup-insert, just
927 * alloc and insert directly. */
928 LASSERT(inode->i_state & I_NEW);
929 conf.coc_lu.loc_flags = LOC_F_NEW;
930 clob = cl_object_find(env, lu2cl_dev(site->ls_top_dev),
934 * No locking is necessary, as new inode is
935 * locked by I_NEW bit.
937 lli->lli_clob = clob;
938 lli->lli_has_smd = lsm_has_objects(md->lsm);
939 lu_object_ref_add(&clob->co_lu, "inode", inode);
941 result = PTR_ERR(clob);
943 result = cl_conf_set(env, lli->lli_clob, &conf);
946 cl_env_put(env, &refcheck);
949 CERROR("Failure to initialize cl object "DFID": %d\n",
955 * Wait for others drop their references of the object at first, then we drop
956 * the last one, which will lead to the object be destroyed immediately.
957 * Must be called after cl_object_kill() against this object.
959 * The reason we want to do this is: destroying top object will wait for sub
960 * objects being destroyed first, so we can't let bottom layer (e.g. from ASTs)
961 * to initiate top object destroying which may deadlock. See bz22520.
963 static void cl_object_put_last(struct lu_env *env, struct cl_object *obj)
965 struct lu_object_header *header = obj->co_lu.lo_header;
968 if (unlikely(atomic_read(&header->loh_ref) != 1)) {
969 struct lu_site *site = obj->co_lu.lo_dev->ld_site;
970 struct lu_site_bkt_data *bkt;
972 bkt = lu_site_bkt_from_fid(site, &header->loh_fid);
974 init_waitqueue_entry_current(&waiter);
975 add_wait_queue(&bkt->lsb_marche_funebre, &waiter);
978 set_current_state(TASK_UNINTERRUPTIBLE);
979 if (atomic_read(&header->loh_ref) == 1)
981 waitq_wait(&waiter, TASK_UNINTERRUPTIBLE);
984 set_current_state(TASK_RUNNING);
985 remove_wait_queue(&bkt->lsb_marche_funebre, &waiter);
988 cl_object_put(env, obj);
991 void cl_inode_fini(struct inode *inode)
994 struct ll_inode_info *lli = ll_i2info(inode);
995 struct cl_object *clob = lli->lli_clob;
1002 cookie = cl_env_reenter();
1003 env = cl_env_get(&refcheck);
1004 emergency = IS_ERR(env);
1006 mutex_lock(&ccc_inode_fini_guard);
1007 LASSERT(ccc_inode_fini_env != NULL);
1008 cl_env_implant(ccc_inode_fini_env, &refcheck);
1009 env = ccc_inode_fini_env;
1012 * cl_object cache is a slave to inode cache (which, in turn
1013 * is a slave to dentry cache), don't keep cl_object in memory
1014 * when its master is evicted.
1016 cl_object_kill(env, clob);
1017 lu_object_ref_del(&clob->co_lu, "inode", inode);
1018 cl_object_put_last(env, clob);
1019 lli->lli_clob = NULL;
1021 cl_env_unplant(ccc_inode_fini_env, &refcheck);
1022 mutex_unlock(&ccc_inode_fini_guard);
1024 cl_env_put(env, &refcheck);
1025 cl_env_reexit(cookie);
1030 * return IF_* type for given lu_dirent entry.
1031 * IF_* flag shld be converted to particular OS file type in
1032 * platform llite module.
1034 __u16 ll_dirent_type_get(struct lu_dirent *ent)
1037 struct luda_type *lt;
1040 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
1041 const unsigned align = sizeof(struct luda_type) - 1;
1043 len = le16_to_cpu(ent->lde_namelen);
1044 len = (len + align) & ~align;
1045 lt = (void *)ent->lde_name + len;
1046 type = IFTODT(le16_to_cpu(lt->lt_type));
1052 * build inode number from passed @fid */
1053 __u64 cl_fid_build_ino(const struct lu_fid *fid, int api32)
1055 if (BITS_PER_LONG == 32 || api32)
1056 RETURN(fid_flatten32(fid));
1058 RETURN(fid_flatten(fid));
1062 * build inode generation from passed @fid. If our FID overflows the 32-bit
1063 * inode number then return a non-zero generation to distinguish them. */
1064 __u32 cl_fid_build_gen(const struct lu_fid *fid)
1069 if (fid_is_igif(fid)) {
1070 gen = lu_igif_gen(fid);
1074 gen = (fid_flatten(fid) >> 32);
1078 /* lsm is unreliable after hsm implementation as layout can be changed at
1079 * any time. This is only to support old, non-clio-ized interfaces. It will
1080 * cause deadlock if clio operations are called with this extra layout refcount
1081 * because in case the layout changed during the IO, ll_layout_refresh() will
1082 * have to wait for the refcount to become zero to destroy the older layout.
1084 * Notice that the lsm returned by this function may not be valid unless called
1085 * inside layout lock - MDS_INODELOCK_LAYOUT. */
1086 struct lov_stripe_md *ccc_inode_lsm_get(struct inode *inode)
1088 return lov_lsm_get(ll_i2info(inode)->lli_clob);
1091 void inline ccc_inode_lsm_put(struct inode *inode, struct lov_stripe_md *lsm)
1093 lov_lsm_put(ll_i2info(inode)->lli_clob, lsm);