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, 2013, 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>
63 #include "llite_internal.h"
65 static const struct cl_req_operations ccc_req_ops;
68 * ccc_ prefix stands for "Common Client Code".
71 static struct kmem_cache *ccc_lock_kmem;
72 static struct kmem_cache *ccc_object_kmem;
73 static struct kmem_cache *ccc_thread_kmem;
74 static struct kmem_cache *ccc_session_kmem;
75 static struct kmem_cache *ccc_req_kmem;
77 static struct lu_kmem_descr ccc_caches[] = {
79 .ckd_cache = &ccc_lock_kmem,
80 .ckd_name = "ccc_lock_kmem",
81 .ckd_size = sizeof (struct ccc_lock)
84 .ckd_cache = &ccc_object_kmem,
85 .ckd_name = "ccc_object_kmem",
86 .ckd_size = sizeof (struct ccc_object)
89 .ckd_cache = &ccc_thread_kmem,
90 .ckd_name = "ccc_thread_kmem",
91 .ckd_size = sizeof (struct ccc_thread_info),
94 .ckd_cache = &ccc_session_kmem,
95 .ckd_name = "ccc_session_kmem",
96 .ckd_size = sizeof (struct ccc_session)
99 .ckd_cache = &ccc_req_kmem,
100 .ckd_name = "ccc_req_kmem",
101 .ckd_size = sizeof (struct ccc_req)
108 /*****************************************************************************
110 * Vvp device and device type functions.
114 void *ccc_key_init(const struct lu_context *ctx, struct lu_context_key *key)
116 struct ccc_thread_info *info;
118 OBD_SLAB_ALLOC_PTR_GFP(info, ccc_thread_kmem, GFP_NOFS);
120 info = ERR_PTR(-ENOMEM);
124 void ccc_key_fini(const struct lu_context *ctx,
125 struct lu_context_key *key, void *data)
127 struct ccc_thread_info *info = data;
128 OBD_SLAB_FREE_PTR(info, ccc_thread_kmem);
131 void *ccc_session_key_init(const struct lu_context *ctx,
132 struct lu_context_key *key)
134 struct ccc_session *session;
136 OBD_SLAB_ALLOC_PTR_GFP(session, ccc_session_kmem, GFP_NOFS);
138 session = ERR_PTR(-ENOMEM);
142 void ccc_session_key_fini(const struct lu_context *ctx,
143 struct lu_context_key *key, void *data)
145 struct ccc_session *session = data;
146 OBD_SLAB_FREE_PTR(session, ccc_session_kmem);
149 struct lu_context_key ccc_key = {
150 .lct_tags = LCT_CL_THREAD,
151 .lct_init = ccc_key_init,
152 .lct_fini = ccc_key_fini
155 struct lu_context_key ccc_session_key = {
156 .lct_tags = LCT_SESSION,
157 .lct_init = ccc_session_key_init,
158 .lct_fini = ccc_session_key_fini
162 /* type constructor/destructor: ccc_type_{init,fini,start,stop}(). */
163 // LU_TYPE_INIT_FINI(ccc, &ccc_key, &ccc_session_key);
165 int ccc_device_init(const struct lu_env *env, struct lu_device *d,
166 const char *name, struct lu_device *next)
168 struct ccc_device *vdv;
173 vdv->cdv_next = lu2cl_dev(next);
175 LASSERT(d->ld_site != NULL && next->ld_type != NULL);
176 next->ld_site = d->ld_site;
177 rc = next->ld_type->ldt_ops->ldto_device_init(
178 env, next, next->ld_type->ldt_name, NULL);
181 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
186 struct lu_device *ccc_device_fini(const struct lu_env *env,
189 return cl2lu_dev(lu2ccc_dev(d)->cdv_next);
192 struct lu_device *ccc_device_alloc(const struct lu_env *env,
193 struct lu_device_type *t,
194 struct lustre_cfg *cfg,
195 const struct lu_device_operations *luops,
196 const struct cl_device_operations *clops)
198 struct ccc_device *vdv;
199 struct lu_device *lud;
200 struct cl_site *site;
206 RETURN(ERR_PTR(-ENOMEM));
208 lud = &vdv->cdv_cl.cd_lu_dev;
209 cl_device_init(&vdv->cdv_cl, t);
210 ccc2lu_dev(vdv)->ld_ops = luops;
211 vdv->cdv_cl.cd_ops = clops;
215 rc = cl_site_init(site, &vdv->cdv_cl);
217 rc = lu_site_init_finish(&site->cs_lu);
219 LASSERT(lud->ld_site == NULL);
220 CERROR("Cannot init lu_site, rc %d.\n", rc);
226 ccc_device_free(env, lud);
232 struct lu_device *ccc_device_free(const struct lu_env *env,
235 struct ccc_device *vdv = lu2ccc_dev(d);
236 struct cl_site *site = lu2cl_site(d->ld_site);
237 struct lu_device *next = cl2lu_dev(vdv->cdv_next);
239 if (d->ld_site != NULL) {
243 cl_device_fini(lu2cl_dev(d));
248 int ccc_req_init(const struct lu_env *env, struct cl_device *dev,
254 OBD_SLAB_ALLOC_PTR_GFP(vrq, ccc_req_kmem, GFP_NOFS);
256 cl_req_slice_add(req, &vrq->crq_cl, dev, &ccc_req_ops);
264 * An `emergency' environment used by ccc_inode_fini() when cl_env_get()
265 * fails. Access to this environment is serialized by ccc_inode_fini_guard
268 static struct lu_env *ccc_inode_fini_env = NULL;
271 * A mutex serializing calls to slp_inode_fini() under extreme memory
272 * pressure, when environments cannot be allocated.
274 static DEFINE_MUTEX(ccc_inode_fini_guard);
275 static int dummy_refcheck;
277 int ccc_global_init(struct lu_device_type *device_type)
281 result = lu_kmem_init(ccc_caches);
285 result = lu_device_type_init(device_type);
289 ccc_inode_fini_env = cl_env_alloc(&dummy_refcheck,
290 LCT_REMEMBER|LCT_NOREF);
291 if (IS_ERR(ccc_inode_fini_env)) {
292 result = PTR_ERR(ccc_inode_fini_env);
296 ccc_inode_fini_env->le_ctx.lc_cookie = 0x4;
299 lu_device_type_fini(device_type);
301 lu_kmem_fini(ccc_caches);
305 void ccc_global_fini(struct lu_device_type *device_type)
307 if (ccc_inode_fini_env != NULL) {
308 cl_env_put(ccc_inode_fini_env, &dummy_refcheck);
309 ccc_inode_fini_env = NULL;
311 lu_device_type_fini(device_type);
312 lu_kmem_fini(ccc_caches);
315 /*****************************************************************************
321 struct lu_object *ccc_object_alloc(const struct lu_env *env,
322 const struct lu_object_header *unused,
323 struct lu_device *dev,
324 const struct cl_object_operations *clops,
325 const struct lu_object_operations *luops)
327 struct ccc_object *vob;
328 struct lu_object *obj;
330 OBD_SLAB_ALLOC_PTR_GFP(vob, ccc_object_kmem, GFP_NOFS);
332 struct cl_object_header *hdr;
335 hdr = &vob->cob_header;
336 cl_object_header_init(hdr);
337 hdr->coh_page_bufsize = cfs_size_round(sizeof(struct cl_page));
339 lu_object_init(obj, &hdr->coh_lu, dev);
340 lu_object_add_top(&hdr->coh_lu, obj);
342 vob->cob_cl.co_ops = clops;
349 int ccc_object_init0(const struct lu_env *env,
350 struct ccc_object *vob,
351 const struct cl_object_conf *conf)
353 vob->cob_inode = conf->coc_inode;
354 atomic_set(&vob->cob_transient_pages, 0);
355 cl_object_page_init(&vob->cob_cl, sizeof(struct ccc_page));
359 int ccc_object_init(const struct lu_env *env, struct lu_object *obj,
360 const struct lu_object_conf *conf)
362 struct ccc_device *dev = lu2ccc_dev(obj->lo_dev);
363 struct ccc_object *vob = lu2ccc(obj);
364 struct lu_object *below;
365 struct lu_device *under;
368 under = &dev->cdv_next->cd_lu_dev;
369 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header, under);
371 const struct cl_object_conf *cconf;
373 cconf = lu2cl_conf(conf);
374 INIT_LIST_HEAD(&vob->cob_pending_list);
375 lu_object_add(obj, below);
376 result = ccc_object_init0(env, vob, cconf);
382 void ccc_object_free(const struct lu_env *env, struct lu_object *obj)
384 struct ccc_object *vob = lu2ccc(obj);
387 lu_object_header_fini(obj->lo_header);
388 OBD_SLAB_FREE_PTR(vob, ccc_object_kmem);
391 int ccc_lock_init(const struct lu_env *env,
392 struct cl_object *obj, struct cl_lock *lock,
393 const struct cl_io *unused,
394 const struct cl_lock_operations *lkops)
396 struct ccc_lock *clk;
399 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
401 OBD_SLAB_ALLOC_PTR_GFP(clk, ccc_lock_kmem, GFP_NOFS);
403 cl_lock_slice_add(lock, &clk->clk_cl, obj, lkops);
410 int ccc_attr_set(const struct lu_env *env, struct cl_object *obj,
411 const struct cl_attr *attr, unsigned valid)
416 int ccc_object_glimpse(const struct lu_env *env,
417 const struct cl_object *obj, struct ost_lvb *lvb)
419 struct inode *inode = ccc_object_inode(obj);
422 lvb->lvb_mtime = cl_inode_mtime(inode);
423 lvb->lvb_atime = cl_inode_atime(inode);
424 lvb->lvb_ctime = cl_inode_ctime(inode);
426 * LU-417: Add dirty pages block count lest i_blocks reports 0, some
427 * "cp" or "tar" on remote node may think it's a completely sparse file
430 if (lvb->lvb_size > 0 && lvb->lvb_blocks == 0)
431 lvb->lvb_blocks = dirty_cnt(inode);
437 int ccc_conf_set(const struct lu_env *env, struct cl_object *obj,
438 const struct cl_object_conf *conf)
440 /* TODO: destroy all pages attached to this object. */
444 static void ccc_object_size_lock(struct cl_object *obj)
446 struct inode *inode = ccc_object_inode(obj);
448 cl_isize_lock(inode);
449 cl_object_attr_lock(obj);
452 static void ccc_object_size_unlock(struct cl_object *obj)
454 struct inode *inode = ccc_object_inode(obj);
456 cl_object_attr_unlock(obj);
457 cl_isize_unlock(inode);
460 /*****************************************************************************
466 int ccc_fail(const struct lu_env *env, const struct cl_page_slice *slice)
475 void ccc_transient_page_verify(const struct cl_page *page)
479 int ccc_transient_page_own(const struct lu_env *env,
480 const struct cl_page_slice *slice,
481 struct cl_io *unused,
484 ccc_transient_page_verify(slice->cpl_page);
488 void ccc_transient_page_assume(const struct lu_env *env,
489 const struct cl_page_slice *slice,
490 struct cl_io *unused)
492 ccc_transient_page_verify(slice->cpl_page);
495 void ccc_transient_page_unassume(const struct lu_env *env,
496 const struct cl_page_slice *slice,
497 struct cl_io *unused)
499 ccc_transient_page_verify(slice->cpl_page);
502 void ccc_transient_page_disown(const struct lu_env *env,
503 const struct cl_page_slice *slice,
504 struct cl_io *unused)
506 ccc_transient_page_verify(slice->cpl_page);
509 void ccc_transient_page_discard(const struct lu_env *env,
510 const struct cl_page_slice *slice,
511 struct cl_io *unused)
513 struct cl_page *page = slice->cpl_page;
515 ccc_transient_page_verify(slice->cpl_page);
518 * For transient pages, remove it from the radix tree.
520 cl_page_delete(env, page);
523 int ccc_transient_page_prep(const struct lu_env *env,
524 const struct cl_page_slice *slice,
525 struct cl_io *unused)
528 /* transient page should always be sent. */
532 /*****************************************************************************
538 void ccc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice)
540 struct ccc_lock *clk = cl2ccc_lock(slice);
541 OBD_SLAB_FREE_PTR(clk, ccc_lock_kmem);
544 int ccc_lock_enqueue(const struct lu_env *env,
545 const struct cl_lock_slice *slice,
546 struct cl_io *unused, struct cl_sync_io *anchor)
548 CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
552 /*****************************************************************************
558 void ccc_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
560 struct cl_io *io = ios->cis_io;
562 CLOBINVRNT(env, io->ci_obj, ccc_object_invariant(io->ci_obj));
565 int ccc_io_one_lock_index(const struct lu_env *env, struct cl_io *io,
566 __u32 enqflags, enum cl_lock_mode mode,
567 pgoff_t start, pgoff_t end)
569 struct ccc_io *cio = ccc_env_io(env);
570 struct cl_lock_descr *descr = &cio->cui_link.cill_descr;
571 struct cl_object *obj = io->ci_obj;
573 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
576 CDEBUG(D_VFSTRACE, "lock: %d [%lu, %lu]\n", mode, start, end);
578 memset(&cio->cui_link, 0, sizeof cio->cui_link);
580 if (cio->cui_fd && (cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
581 descr->cld_mode = CLM_GROUP;
582 descr->cld_gid = cio->cui_fd->fd_grouplock.cg_gid;
584 descr->cld_mode = mode;
586 descr->cld_obj = obj;
587 descr->cld_start = start;
588 descr->cld_end = end;
589 descr->cld_enq_flags = enqflags;
591 cl_io_lock_add(env, io, &cio->cui_link);
595 void ccc_io_update_iov(const struct lu_env *env,
596 struct ccc_io *cio, struct cl_io *io)
599 size_t size = io->u.ci_rw.crw_count;
601 cio->cui_iov_olen = 0;
602 if (!cl_is_normalio(env, io) || cio->cui_tot_nrsegs == 0)
605 for (i = 0; i < cio->cui_tot_nrsegs; i++) {
606 struct iovec *iv = &cio->cui_iov[i];
608 if (iv->iov_len < size)
611 if (iv->iov_len > size) {
612 cio->cui_iov_olen = iv->iov_len;
619 cio->cui_nrsegs = i + 1;
620 LASSERTF(cio->cui_tot_nrsegs >= cio->cui_nrsegs,
621 "tot_nrsegs: %lu, nrsegs: %lu\n",
622 cio->cui_tot_nrsegs, cio->cui_nrsegs);
625 int ccc_io_one_lock(const struct lu_env *env, struct cl_io *io,
626 __u32 enqflags, enum cl_lock_mode mode,
627 loff_t start, loff_t end)
629 struct cl_object *obj = io->ci_obj;
630 return ccc_io_one_lock_index(env, io, enqflags, mode,
631 cl_index(obj, start), cl_index(obj, end));
634 void ccc_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
636 CLOBINVRNT(env, ios->cis_io->ci_obj,
637 ccc_object_invariant(ios->cis_io->ci_obj));
640 void ccc_io_advance(const struct lu_env *env,
641 const struct cl_io_slice *ios,
644 struct ccc_io *cio = cl2ccc_io(env, ios);
645 struct cl_io *io = ios->cis_io;
646 struct cl_object *obj = ios->cis_io->ci_obj;
648 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
650 if (!cl_is_normalio(env, io))
653 LASSERT(cio->cui_tot_nrsegs >= cio->cui_nrsegs);
654 LASSERT(cio->cui_tot_count >= nob);
656 cio->cui_iov += cio->cui_nrsegs;
657 cio->cui_tot_nrsegs -= cio->cui_nrsegs;
658 cio->cui_tot_count -= nob;
661 if (cio->cui_iov_olen > 0) {
665 cio->cui_tot_nrsegs++;
666 iv = &cio->cui_iov[0];
667 if (io->ci_continue) {
668 iv->iov_base += iv->iov_len;
669 LASSERT(cio->cui_iov_olen > iv->iov_len);
670 iv->iov_len = cio->cui_iov_olen - iv->iov_len;
672 /* restore the iov_len, in case of restart io. */
673 iv->iov_len = cio->cui_iov_olen;
675 cio->cui_iov_olen = 0;
680 * Helper function that if necessary adjusts file size (inode->i_size), when
681 * position at the offset \a pos is accessed. File size can be arbitrary stale
682 * on a Lustre client, but client at least knows KMS. If accessed area is
683 * inside [0, KMS], set file size to KMS, otherwise glimpse file size.
685 * Locking: cl_isize_lock is used to serialize changes to inode size and to
686 * protect consistency between inode size and cl_object
687 * attributes. cl_object_size_lock() protects consistency between cl_attr's of
688 * top-object and sub-objects.
690 int ccc_prep_size(const struct lu_env *env, struct cl_object *obj,
691 struct cl_io *io, loff_t start, size_t count, int *exceed)
693 struct cl_attr *attr = ccc_env_thread_attr(env);
694 struct inode *inode = ccc_object_inode(obj);
695 loff_t pos = start + count - 1;
700 * Consistency guarantees: following possibilities exist for the
701 * relation between region being accessed and real file size at this
704 * (A): the region is completely inside of the file;
706 * (B-x): x bytes of region are inside of the file, the rest is
709 * (C): the region is completely outside of the file.
711 * This classification is stable under DLM lock already acquired by
712 * the caller, because to change the class, other client has to take
713 * DLM lock conflicting with our lock. Also, any updates to ->i_size
714 * by other threads on this client are serialized by
715 * ll_inode_size_lock(). This guarantees that short reads are handled
716 * correctly in the face of concurrent writes and truncates.
718 ccc_object_size_lock(obj);
719 result = cl_object_attr_get(env, obj, attr);
724 * A glimpse is necessary to determine whether we
725 * return a short read (B) or some zeroes at the end
728 ccc_object_size_unlock(obj);
729 result = cl_glimpse_lock(env, io, inode, obj, 0);
730 if (result == 0 && exceed != NULL) {
731 /* If objective page index exceed end-of-file
732 * page index, return directly. Do not expect
733 * kernel will check such case correctly.
734 * linux-2.6.18-128.1.1 miss to do that.
736 loff_t size = cl_isize_read(inode);
737 unsigned long cur_index = start >>
740 if ((size == 0 && cur_index != 0) ||
741 (((size - 1) >> PAGE_CACHE_SHIFT) <
748 * region is within kms and, hence, within real file
749 * size (A). We need to increase i_size to cover the
750 * read region so that generic_file_read() will do its
751 * job, but that doesn't mean the kms size is
752 * _correct_, it is only the _minimum_ size. If
753 * someone does a stat they will get the correct size
754 * which will always be >= the kms value here.
757 if (cl_isize_read(inode) < kms) {
758 cl_isize_write_nolock(inode, kms);
760 DFID" updating i_size "LPU64"\n",
761 PFID(lu_object_fid(&obj->co_lu)),
762 (__u64)cl_isize_read(inode));
767 ccc_object_size_unlock(obj);
771 /*****************************************************************************
773 * Transfer operations.
777 void ccc_req_completion(const struct lu_env *env,
778 const struct cl_req_slice *slice, int ioret)
783 cl_stats_tally(slice->crs_dev, slice->crs_req->crq_type, ioret);
785 vrq = cl2ccc_req(slice);
786 OBD_SLAB_FREE_PTR(vrq, ccc_req_kmem);
790 * Implementation of struct cl_req_operations::cro_attr_set() for ccc
791 * layer. ccc is responsible for
809 void ccc_req_attr_set(const struct lu_env *env,
810 const struct cl_req_slice *slice,
811 const struct cl_object *obj,
812 struct cl_req_attr *attr, obd_valid flags)
816 obd_flag valid_flags;
819 inode = ccc_object_inode(obj);
820 valid_flags = OBD_MD_FLTYPE;
822 if ((flags & OBD_MD_FLOSSCAPA) != 0) {
823 LASSERT(attr->cra_capa == NULL);
824 attr->cra_capa = cl_capa_lookup(inode,
825 slice->crs_req->crq_type);
828 if (slice->crs_req->crq_type == CRT_WRITE) {
829 if (flags & OBD_MD_FLEPOCH) {
830 oa->o_valid |= OBD_MD_FLEPOCH;
831 oa->o_ioepoch = cl_i2info(inode)->lli_ioepoch;
832 valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
833 OBD_MD_FLUID | OBD_MD_FLGID;
836 obdo_from_inode(oa, inode, valid_flags & flags);
837 obdo_set_parent_fid(oa, &cl_i2info(inode)->lli_fid);
838 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_INVALID_PFID))
840 memcpy(attr->cra_jobid, cl_i2info(inode)->lli_jobid,
844 static const struct cl_req_operations ccc_req_ops = {
845 .cro_attr_set = ccc_req_attr_set,
846 .cro_completion = ccc_req_completion
849 int cl_setattr_ost(struct inode *inode, const struct iattr *attr,
850 struct obd_capa *capa)
859 env = cl_env_get(&refcheck);
861 RETURN(PTR_ERR(env));
863 io = ccc_env_thread_io(env);
864 io->ci_obj = cl_i2info(inode)->lli_clob;
866 io->u.ci_setattr.sa_attr.lvb_atime = LTIME_S(attr->ia_atime);
867 io->u.ci_setattr.sa_attr.lvb_mtime = LTIME_S(attr->ia_mtime);
868 io->u.ci_setattr.sa_attr.lvb_ctime = LTIME_S(attr->ia_ctime);
869 io->u.ci_setattr.sa_attr.lvb_size = attr->ia_size;
870 io->u.ci_setattr.sa_valid = attr->ia_valid;
871 io->u.ci_setattr.sa_capa = capa;
874 if (cl_io_init(env, io, CIT_SETATTR, io->ci_obj) == 0) {
875 struct ccc_io *cio = ccc_env_io(env);
877 if (attr->ia_valid & ATTR_FILE)
878 /* populate the file descriptor for ftruncate to honor
879 * group lock - see LU-787 */
880 cio->cui_fd = cl_iattr2fd(inode, attr);
882 result = cl_io_loop(env, io);
884 result = io->ci_result;
887 if (unlikely(io->ci_need_restart))
889 /* HSM import case: file is released, cannot be restored
890 * no need to fail except if restore registration failed
892 if (result == -ENODATA && io->ci_restore_needed &&
893 io->ci_result != -ENODATA)
895 cl_env_put(env, &refcheck);
899 /*****************************************************************************
905 struct lu_device *ccc2lu_dev(struct ccc_device *vdv)
907 return &vdv->cdv_cl.cd_lu_dev;
910 struct ccc_device *lu2ccc_dev(const struct lu_device *d)
912 return container_of0(d, struct ccc_device, cdv_cl.cd_lu_dev);
915 struct ccc_device *cl2ccc_dev(const struct cl_device *d)
917 return container_of0(d, struct ccc_device, cdv_cl);
920 struct lu_object *ccc2lu(struct ccc_object *vob)
922 return &vob->cob_cl.co_lu;
925 struct ccc_object *lu2ccc(const struct lu_object *obj)
927 return container_of0(obj, struct ccc_object, cob_cl.co_lu);
930 struct ccc_object *cl2ccc(const struct cl_object *obj)
932 return container_of0(obj, struct ccc_object, cob_cl);
935 struct ccc_lock *cl2ccc_lock(const struct cl_lock_slice *slice)
937 return container_of(slice, struct ccc_lock, clk_cl);
940 struct ccc_io *cl2ccc_io(const struct lu_env *env,
941 const struct cl_io_slice *slice)
945 cio = container_of(slice, struct ccc_io, cui_cl);
946 LASSERT(cio == ccc_env_io(env));
950 struct ccc_req *cl2ccc_req(const struct cl_req_slice *slice)
952 return container_of0(slice, struct ccc_req, crq_cl);
955 struct page *cl2vm_page(const struct cl_page_slice *slice)
957 return cl2ccc_page(slice)->cpg_page;
960 /*****************************************************************************
965 int ccc_object_invariant(const struct cl_object *obj)
967 struct inode *inode = ccc_object_inode(obj);
968 struct cl_inode_info *lli = cl_i2info(inode);
970 return (S_ISREG(cl_inode_mode(inode)) ||
971 /* i_mode of unlinked inode is zeroed. */
972 cl_inode_mode(inode) == 0) && lli->lli_clob == obj;
975 struct inode *ccc_object_inode(const struct cl_object *obj)
977 return cl2ccc(obj)->cob_inode;
981 * Returns a pointer to cl_page associated with \a vmpage, without acquiring
982 * additional reference to the resulting page. This is an unsafe version of
983 * cl_vmpage_page() that can only be used under vmpage lock.
985 struct cl_page *ccc_vmpage_page_transient(struct page *vmpage)
987 KLASSERT(PageLocked(vmpage));
988 return (struct cl_page *)vmpage->private;
992 * Initialize or update CLIO structures for regular files when new
993 * meta-data arrives from the server.
995 * \param inode regular file inode
996 * \param md new file metadata from MDS
997 * - allocates cl_object if necessary,
998 * - updated layout, if object was already here.
1000 int cl_file_inode_init(struct inode *inode, struct lustre_md *md)
1003 struct cl_inode_info *lli;
1004 struct cl_object *clob;
1005 struct lu_site *site;
1007 struct cl_object_conf conf = {
1016 LASSERT(md->body->mbo_valid & OBD_MD_FLID);
1017 LASSERT(S_ISREG(cl_inode_mode(inode)));
1019 env = cl_env_get(&refcheck);
1021 return PTR_ERR(env);
1023 site = cl_i2sbi(inode)->ll_site;
1024 lli = cl_i2info(inode);
1025 fid = &lli->lli_fid;
1026 LASSERT(fid_is_sane(fid));
1028 if (lli->lli_clob == NULL) {
1029 /* clob is slave of inode, empty lli_clob means for new inode,
1030 * there is no clob in cache with the given fid, so it is
1031 * unnecessary to perform lookup-alloc-lookup-insert, just
1032 * alloc and insert directly. */
1033 LASSERT(inode->i_state & I_NEW);
1034 conf.coc_lu.loc_flags = LOC_F_NEW;
1035 clob = cl_object_find(env, lu2cl_dev(site->ls_top_dev),
1037 if (!IS_ERR(clob)) {
1039 * No locking is necessary, as new inode is
1040 * locked by I_NEW bit.
1042 lli->lli_clob = clob;
1043 lli->lli_has_smd = lsm_has_objects(md->lsm);
1044 lu_object_ref_add(&clob->co_lu, "inode", inode);
1046 result = PTR_ERR(clob);
1048 result = cl_conf_set(env, lli->lli_clob, &conf);
1051 cl_env_put(env, &refcheck);
1054 CERROR("Failure to initialize cl object "DFID": %d\n",
1060 * Wait for others drop their references of the object at first, then we drop
1061 * the last one, which will lead to the object be destroyed immediately.
1062 * Must be called after cl_object_kill() against this object.
1064 * The reason we want to do this is: destroying top object will wait for sub
1065 * objects being destroyed first, so we can't let bottom layer (e.g. from ASTs)
1066 * to initiate top object destroying which may deadlock. See bz22520.
1068 static void cl_object_put_last(struct lu_env *env, struct cl_object *obj)
1070 struct lu_object_header *header = obj->co_lu.lo_header;
1071 wait_queue_t waiter;
1073 if (unlikely(atomic_read(&header->loh_ref) != 1)) {
1074 struct lu_site *site = obj->co_lu.lo_dev->ld_site;
1075 struct lu_site_bkt_data *bkt;
1077 bkt = lu_site_bkt_from_fid(site, &header->loh_fid);
1079 init_waitqueue_entry_current(&waiter);
1080 add_wait_queue(&bkt->lsb_marche_funebre, &waiter);
1083 set_current_state(TASK_UNINTERRUPTIBLE);
1084 if (atomic_read(&header->loh_ref) == 1)
1086 waitq_wait(&waiter, TASK_UNINTERRUPTIBLE);
1089 set_current_state(TASK_RUNNING);
1090 remove_wait_queue(&bkt->lsb_marche_funebre, &waiter);
1093 cl_object_put(env, obj);
1096 void cl_inode_fini(struct inode *inode)
1099 struct cl_inode_info *lli = cl_i2info(inode);
1100 struct cl_object *clob = lli->lli_clob;
1107 cookie = cl_env_reenter();
1108 env = cl_env_get(&refcheck);
1109 emergency = IS_ERR(env);
1111 mutex_lock(&ccc_inode_fini_guard);
1112 LASSERT(ccc_inode_fini_env != NULL);
1113 cl_env_implant(ccc_inode_fini_env, &refcheck);
1114 env = ccc_inode_fini_env;
1117 * cl_object cache is a slave to inode cache (which, in turn
1118 * is a slave to dentry cache), don't keep cl_object in memory
1119 * when its master is evicted.
1121 cl_object_kill(env, clob);
1122 lu_object_ref_del(&clob->co_lu, "inode", inode);
1123 cl_object_put_last(env, clob);
1124 lli->lli_clob = NULL;
1126 cl_env_unplant(ccc_inode_fini_env, &refcheck);
1127 mutex_unlock(&ccc_inode_fini_guard);
1129 cl_env_put(env, &refcheck);
1130 cl_env_reexit(cookie);
1135 * return IF_* type for given lu_dirent entry.
1136 * IF_* flag shld be converted to particular OS file type in
1137 * platform llite module.
1139 __u16 ll_dirent_type_get(struct lu_dirent *ent)
1142 struct luda_type *lt;
1145 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
1146 const unsigned align = sizeof(struct luda_type) - 1;
1148 len = le16_to_cpu(ent->lde_namelen);
1149 len = (len + align) & ~align;
1150 lt = (void *)ent->lde_name + len;
1151 type = IFTODT(le16_to_cpu(lt->lt_type));
1157 * build inode number from passed @fid */
1158 __u64 cl_fid_build_ino(const struct lu_fid *fid, int api32)
1160 if (BITS_PER_LONG == 32 || api32)
1161 RETURN(fid_flatten32(fid));
1163 RETURN(fid_flatten(fid));
1167 * build inode generation from passed @fid. If our FID overflows the 32-bit
1168 * inode number then return a non-zero generation to distinguish them. */
1169 __u32 cl_fid_build_gen(const struct lu_fid *fid)
1174 if (fid_is_igif(fid)) {
1175 gen = lu_igif_gen(fid);
1179 gen = (fid_flatten(fid) >> 32);
1183 /* lsm is unreliable after hsm implementation as layout can be changed at
1184 * any time. This is only to support old, non-clio-ized interfaces. It will
1185 * cause deadlock if clio operations are called with this extra layout refcount
1186 * because in case the layout changed during the IO, ll_layout_refresh() will
1187 * have to wait for the refcount to become zero to destroy the older layout.
1189 * Notice that the lsm returned by this function may not be valid unless called
1190 * inside layout lock - MDS_INODELOCK_LAYOUT. */
1191 struct lov_stripe_md *ccc_inode_lsm_get(struct inode *inode)
1193 return lov_lsm_get(cl_i2info(inode)->lli_clob);
1196 void inline ccc_inode_lsm_put(struct inode *inode, struct lov_stripe_md *lsm)
1198 lov_lsm_put(cl_i2info(inode)->lli_clob, lsm);