1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
30 * Use is subject to license terms.
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
45 # include <libcfs/libcfs.h>
46 # include <linux/fs.h>
47 # include <linux/sched.h>
48 # include <linux/mm.h>
49 # include <linux/smp_lock.h>
50 # include <linux/quotaops.h>
51 # include <linux/highmem.h>
52 # include <linux/pagemap.h>
53 # include <linux/rbtree.h>
54 #else /* __KERNEL__ */
59 #include <sys/types.h>
61 #include <sys/queue.h>
73 # include <liblustre.h>
77 #include <obd_support.h>
78 #include <lustre_fid.h>
79 #include <lustre_lite.h>
80 #include <lustre_dlm.h>
81 #include <lustre_ver.h>
82 #include <lustre_mdc.h>
83 #include <cl_object.h>
88 #include "../llite/llite_internal.h"
90 #include "../liblustre/llite_lib.h"
93 const struct cl_req_operations ccc_req_ops;
96 * ccc_ prefix stands for "Common Client Code".
99 static cfs_mem_cache_t *ccc_lock_kmem;
100 static cfs_mem_cache_t *ccc_object_kmem;
101 static cfs_mem_cache_t *ccc_thread_kmem;
102 static cfs_mem_cache_t *ccc_session_kmem;
103 static cfs_mem_cache_t *ccc_req_kmem;
105 static struct lu_kmem_descr ccc_caches[] = {
107 .ckd_cache = &ccc_lock_kmem,
108 .ckd_name = "ccc_lock_kmem",
109 .ckd_size = sizeof (struct ccc_lock)
112 .ckd_cache = &ccc_object_kmem,
113 .ckd_name = "ccc_object_kmem",
114 .ckd_size = sizeof (struct ccc_object)
117 .ckd_cache = &ccc_thread_kmem,
118 .ckd_name = "ccc_thread_kmem",
119 .ckd_size = sizeof (struct ccc_thread_info),
122 .ckd_cache = &ccc_session_kmem,
123 .ckd_name = "ccc_session_kmem",
124 .ckd_size = sizeof (struct ccc_session)
127 .ckd_cache = &ccc_req_kmem,
128 .ckd_name = "ccc_req_kmem",
129 .ckd_size = sizeof (struct ccc_req)
136 /*****************************************************************************
138 * Vvp device and device type functions.
142 void *ccc_key_init(const struct lu_context *ctx,
143 struct lu_context_key *key)
145 struct ccc_thread_info *info;
147 OBD_SLAB_ALLOC_PTR_GFP(info, ccc_thread_kmem, CFS_ALLOC_IO);
149 info = ERR_PTR(-ENOMEM);
153 void ccc_key_fini(const struct lu_context *ctx,
154 struct lu_context_key *key, void *data)
156 struct ccc_thread_info *info = data;
157 OBD_SLAB_FREE_PTR(info, ccc_thread_kmem);
160 void *ccc_session_key_init(const struct lu_context *ctx,
161 struct lu_context_key *key)
163 struct ccc_session *session;
165 OBD_SLAB_ALLOC_PTR_GFP(session, ccc_session_kmem, CFS_ALLOC_IO);
167 session = ERR_PTR(-ENOMEM);
171 void ccc_session_key_fini(const struct lu_context *ctx,
172 struct lu_context_key *key, void *data)
174 struct ccc_session *session = data;
175 OBD_SLAB_FREE_PTR(session, ccc_session_kmem);
178 struct lu_context_key ccc_key = {
179 .lct_tags = LCT_CL_THREAD,
180 .lct_init = ccc_key_init,
181 .lct_fini = ccc_key_fini
184 struct lu_context_key ccc_session_key = {
185 .lct_tags = LCT_SESSION,
186 .lct_init = ccc_session_key_init,
187 .lct_fini = ccc_session_key_fini
191 /* type constructor/destructor: ccc_type_{init,fini,start,stop}(). */
192 // LU_TYPE_INIT_FINI(ccc, &ccc_key, &ccc_session_key);
194 int ccc_device_init(const struct lu_env *env, struct lu_device *d,
195 const char *name, struct lu_device *next)
197 struct ccc_device *vdv;
202 vdv->cdv_next = lu2cl_dev(next);
204 LASSERT(d->ld_site != NULL && next->ld_type != NULL);
205 next->ld_site = d->ld_site;
206 rc = next->ld_type->ldt_ops->ldto_device_init(
207 env, next, next->ld_type->ldt_name, NULL);
210 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
215 struct lu_device *ccc_device_fini(const struct lu_env *env,
218 return cl2lu_dev(lu2ccc_dev(d)->cdv_next);
221 struct lu_device *ccc_device_alloc(const struct lu_env *env,
222 struct lu_device_type *t,
223 struct lustre_cfg *cfg,
224 const struct lu_device_operations *luops,
225 const struct cl_device_operations *clops)
227 struct ccc_device *vdv;
228 struct lu_device *lud;
229 struct cl_site *site;
235 RETURN(ERR_PTR(-ENOMEM));
237 lud = &vdv->cdv_cl.cd_lu_dev;
238 cl_device_init(&vdv->cdv_cl, t);
239 ccc2lu_dev(vdv)->ld_ops = luops;
240 vdv->cdv_cl.cd_ops = clops;
244 rc = cl_site_init(site, &vdv->cdv_cl);
246 rc = lu_site_init_finish(&site->cs_lu);
248 LASSERT(lud->ld_site == NULL);
249 CERROR("Cannot init lu_site, rc %d.\n", rc);
255 ccc_device_free(env, lud);
261 struct lu_device *ccc_device_free(const struct lu_env *env,
264 struct ccc_device *vdv = lu2ccc_dev(d);
265 struct cl_site *site = lu2cl_site(d->ld_site);
266 struct lu_device *next = cl2lu_dev(vdv->cdv_next);
268 if (d->ld_site != NULL) {
272 cl_device_fini(lu2cl_dev(d));
277 int ccc_req_init(const struct lu_env *env, struct cl_device *dev,
283 OBD_SLAB_ALLOC_PTR_GFP(vrq, ccc_req_kmem, CFS_ALLOC_IO);
285 cl_req_slice_add(req, &vrq->crq_cl, dev, &ccc_req_ops);
293 * An `emergency' environment used by ccc_inode_fini() when cl_env_get()
294 * fails. Access to this environment is serialized by ccc_inode_fini_guard
297 static struct lu_env *ccc_inode_fini_env = NULL;
300 * A mutex serializing calls to slp_inode_fini() under extreme memory
301 * pressure, when environments cannot be allocated.
303 static DEFINE_MUTEX(ccc_inode_fini_guard);
304 static int dummy_refcheck;
306 int ccc_global_init(struct lu_device_type *device_type)
310 result = lu_kmem_init(ccc_caches);
312 result = lu_device_type_init(device_type);
313 ccc_inode_fini_env = cl_env_alloc(&dummy_refcheck,
314 LCT_REMEMBER|LCT_NOREF);
315 if (IS_ERR(ccc_inode_fini_env))
316 result = PTR_ERR(ccc_inode_fini_env);
318 ccc_inode_fini_env->le_ctx.lc_cookie = 0x4;
323 void ccc_global_fini(struct lu_device_type *device_type)
325 if (ccc_inode_fini_env != NULL) {
326 cl_env_put(ccc_inode_fini_env, &dummy_refcheck);
327 ccc_inode_fini_env = NULL;
329 lu_device_type_fini(device_type);
330 lu_kmem_fini(ccc_caches);
333 /*****************************************************************************
339 struct lu_object *ccc_object_alloc(const struct lu_env *env,
340 const struct lu_object_header *_,
341 struct lu_device *dev,
342 const struct cl_object_operations *clops,
343 const struct lu_object_operations *luops)
345 struct ccc_object *vob;
346 struct lu_object *obj;
348 OBD_SLAB_ALLOC_PTR_GFP(vob, ccc_object_kmem, CFS_ALLOC_IO);
350 struct cl_object_header *hdr;
353 hdr = &vob->cob_header;
354 cl_object_header_init(hdr);
355 lu_object_init(obj, &hdr->coh_lu, dev);
356 lu_object_add_top(&hdr->coh_lu, obj);
358 vob->cob_cl.co_ops = clops;
365 int ccc_object_init0(const struct lu_env *env,
366 struct ccc_object *vob,
367 const struct cl_object_conf *conf)
369 vob->cob_inode = conf->coc_inode;
370 vob->cob_transient_pages = 0;
374 int ccc_object_init(const struct lu_env *env, struct lu_object *obj,
375 const struct lu_object_conf *conf)
377 struct ccc_device *dev = lu2ccc_dev(obj->lo_dev);
378 struct ccc_object *vob = lu2ccc(obj);
379 struct lu_object *below;
380 struct lu_device *under;
383 under = &dev->cdv_next->cd_lu_dev;
384 below = under->ld_ops->ldo_object_alloc(env, obj->lo_header, under);
386 const struct cl_object_conf *cconf;
388 cconf = lu2cl_conf(conf);
389 CFS_INIT_LIST_HEAD(&vob->cob_pending_list);
390 lu_object_add(obj, below);
391 result = ccc_object_init0(env, vob, cconf);
397 void ccc_object_free(const struct lu_env *env, struct lu_object *obj)
399 struct ccc_object *vob = lu2ccc(obj);
402 lu_object_header_fini(obj->lo_header);
403 OBD_SLAB_FREE_PTR(vob, ccc_object_kmem);
406 int ccc_lock_init(const struct lu_env *env,
407 struct cl_object *obj, struct cl_lock *lock,
408 const struct cl_io *_,
409 const struct cl_lock_operations *lkops)
411 struct ccc_lock *clk;
414 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
416 OBD_SLAB_ALLOC_PTR_GFP(clk, ccc_lock_kmem, CFS_ALLOC_IO);
418 cl_lock_slice_add(lock, &clk->clk_cl, obj, lkops);
425 int ccc_attr_set(const struct lu_env *env, struct cl_object *obj,
426 const struct cl_attr *attr, unsigned valid)
431 int ccc_object_glimpse(const struct lu_env *env,
432 const struct cl_object *obj, struct ost_lvb *lvb)
434 struct inode *inode = ccc_object_inode(obj);
437 lvb->lvb_mtime = cl_inode_mtime(inode);
438 lvb->lvb_atime = cl_inode_atime(inode);
439 lvb->lvb_ctime = cl_inode_ctime(inode);
445 int ccc_conf_set(const struct lu_env *env, struct cl_object *obj,
446 const struct cl_object_conf *conf)
448 /* TODO: destroy all pages attached to this object. */
452 /*****************************************************************************
458 cfs_page_t *ccc_page_vmpage(const struct lu_env *env,
459 const struct cl_page_slice *slice)
461 return cl2vm_page(slice);
464 int ccc_page_is_under_lock(const struct lu_env *env,
465 const struct cl_page_slice *slice,
468 struct ccc_io *vio = ccc_env_io(env);
469 struct cl_lock_descr *desc = &ccc_env_info(env)->cti_descr;
470 struct cl_page *page = slice->cpl_page;
476 if (io->ci_type == CIT_READ || io->ci_type == CIT_WRITE ||
477 io->ci_type == CIT_FAULT) {
478 if (vio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)
481 desc->cld_start = page->cp_index;
482 desc->cld_end = page->cp_index;
483 desc->cld_obj = page->cp_obj;
484 desc->cld_mode = CLM_READ;
485 result = cl_queue_match(&io->ci_lockset.cls_done,
493 int ccc_fail(const struct lu_env *env, const struct cl_page_slice *slice)
502 void ccc_transient_page_verify(const struct cl_page *page)
506 void ccc_transient_page_own(const struct lu_env *env,
507 const struct cl_page_slice *slice,
510 ccc_transient_page_verify(slice->cpl_page);
513 void ccc_transient_page_assume(const struct lu_env *env,
514 const struct cl_page_slice *slice,
517 ccc_transient_page_verify(slice->cpl_page);
520 void ccc_transient_page_unassume(const struct lu_env *env,
521 const struct cl_page_slice *slice,
524 ccc_transient_page_verify(slice->cpl_page);
527 void ccc_transient_page_disown(const struct lu_env *env,
528 const struct cl_page_slice *slice,
531 ccc_transient_page_verify(slice->cpl_page);
534 void ccc_transient_page_discard(const struct lu_env *env,
535 const struct cl_page_slice *slice,
538 struct cl_page *page = slice->cpl_page;
540 ccc_transient_page_verify(slice->cpl_page);
543 * For transient pages, remove it from the radix tree.
545 cl_page_delete(env, page);
548 int ccc_transient_page_prep(const struct lu_env *env,
549 const struct cl_page_slice *slice,
553 /* transient page should always be sent. */
557 /*****************************************************************************
563 void ccc_lock_fini(const struct lu_env *env, struct cl_lock_slice *slice)
565 struct ccc_lock *clk = cl2ccc_lock(slice);
567 CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
568 OBD_SLAB_FREE_PTR(clk, ccc_lock_kmem);
571 int ccc_lock_enqueue(const struct lu_env *env,
572 const struct cl_lock_slice *slice,
573 struct cl_io *_, __u32 enqflags)
575 CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
579 int ccc_lock_unuse(const struct lu_env *env, const struct cl_lock_slice *slice)
581 CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
585 int ccc_lock_wait(const struct lu_env *env, const struct cl_lock_slice *slice)
587 CLOBINVRNT(env, slice->cls_obj, ccc_object_invariant(slice->cls_obj));
592 * Implementation of cl_lock_operations::clo_fits_into() methods for ccc
593 * layer. This function is executed every time io finds an existing lock in
594 * the lock cache while creating new lock. This function has to decide whether
595 * cached lock "fits" into io.
597 * \param slice lock to be checked
599 * \param io IO that wants a lock.
601 * \see lov_lock_fits_into().
603 int ccc_lock_fits_into(const struct lu_env *env,
604 const struct cl_lock_slice *slice,
605 const struct cl_lock_descr *need,
606 const struct cl_io *io)
608 const struct cl_lock *lock = slice->cls_lock;
609 const struct cl_lock_descr *descr = &lock->cll_descr;
610 const struct ccc_io *cio = ccc_env_io(env);
615 * Work around DLM peculiarity: it assumes that glimpse
616 * (LDLM_FL_HAS_INTENT) lock is always LCK_PR, and returns reads lock
617 * when asked for LCK_PW lock with LDLM_FL_HAS_INTENT flag set. Make
618 * sure that glimpse doesn't get CLM_WRITE top-lock, so that it
619 * doesn't enqueue CLM_WRITE sub-locks.
621 if (cio->cui_glimpse)
622 result = descr->cld_mode != CLM_WRITE;
624 * Also, don't match incomplete write locks for read, otherwise read
625 * would enqueue missing sub-locks in the write mode.
627 * XXX this is a candidate for generic locking policy, to be moved
628 * into cl_lock_lookup().
630 else if (need->cld_mode != descr->cld_mode)
631 result = lock->cll_state >= CLS_ENQUEUED;
638 * Implements cl_lock_operations::clo_state() method for ccc layer, invoked
639 * whenever lock state changes. Transfers object attributes, that might be
640 * updated as a result of lock acquiring into inode.
642 void ccc_lock_state(const struct lu_env *env,
643 const struct cl_lock_slice *slice,
644 enum cl_lock_state state)
646 struct cl_lock *lock;
647 struct cl_object *obj;
649 struct cl_attr *attr;
652 lock = slice->cls_lock;
655 * Refresh inode attributes when the lock is moving into CLS_HELD
656 * state, and only when this is a result of real enqueue, rather than
657 * of finding lock in the cache.
659 if (state == CLS_HELD && lock->cll_state < CLS_HELD) {
662 obj = slice->cls_obj;
663 inode = ccc_object_inode(obj);
664 attr = &ccc_env_info(env)->cti_attr;
666 /* vmtruncate()->ll_truncate() first sets the i_size and then
667 * the kms under both a DLM lock and the
668 * ll_inode_size_lock(). If we don't get the
669 * ll_inode_size_lock() here we can match the DLM lock and
670 * reset i_size from the kms before the truncating path has
671 * updated the kms. generic_file_write can then trust the
672 * stale i_size when doing appending writes and effectively
673 * cancel the result of the truncate. Getting the
674 * ll_inode_size_lock() after the enqueue maintains the DLM
675 * -> ll_inode_size_lock() acquiring order. */
676 cl_isize_lock(inode, 0);
677 cl_object_attr_lock(obj);
678 rc = cl_object_attr_get(env, obj, attr);
680 if (lock->cll_descr.cld_start == 0 &&
681 lock->cll_descr.cld_end == CL_PAGE_EOF) {
682 cl_isize_write(inode, attr->cat_kms);
683 CDEBUG(D_INODE, DFID" updating i_size %llu\n",
684 PFID(lu_object_fid(&obj->co_lu)),
685 (__u64)cl_isize_read(inode));
687 cl_inode_mtime(inode) = attr->cat_mtime;
688 cl_inode_atime(inode) = attr->cat_atime;
689 cl_inode_ctime(inode) = attr->cat_ctime;
691 CL_LOCK_DEBUG(D_ERROR, env, lock, "attr_get: %i\n", rc);
692 cl_object_attr_unlock(obj);
693 cl_isize_unlock(inode, 0);
698 /*****************************************************************************
704 void ccc_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
706 struct cl_io *io = ios->cis_io;
708 CLOBINVRNT(env, io->ci_obj, ccc_object_invariant(io->ci_obj));
711 int ccc_io_one_lock_index(const struct lu_env *env, struct cl_io *io,
712 __u32 enqflags, enum cl_lock_mode mode,
713 pgoff_t start, pgoff_t end)
715 struct ccc_io *vio = ccc_env_io(env);
716 struct cl_lock_descr *descr = &vio->cui_link.cill_descr;
717 struct cl_object *obj = io->ci_obj;
719 CLOBINVRNT(env, obj, ccc_object_invariant(obj));
722 CDEBUG(D_VFSTRACE, "lock: %i [%lu, %lu]\n", mode, start, end);
724 memset(&vio->cui_link, 0, sizeof vio->cui_link);
725 descr->cld_mode = mode;
726 descr->cld_obj = obj;
727 descr->cld_start = start;
728 descr->cld_end = end;
730 vio->cui_link.cill_enq_flags = enqflags;
731 cl_io_lock_add(env, io, &vio->cui_link);
735 int ccc_io_one_lock(const struct lu_env *env, struct cl_io *io,
736 __u32 enqflags, enum cl_lock_mode mode,
737 loff_t start, loff_t end)
739 struct cl_object *obj = io->ci_obj;
741 return ccc_io_one_lock_index(env, io, enqflags, mode,
742 cl_index(obj, start), cl_index(obj, end));
745 void ccc_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
747 CLOBINVRNT(env, ios->cis_io->ci_obj,
748 ccc_object_invariant(ios->cis_io->ci_obj));
751 static void ccc_object_size_lock(struct cl_object *obj, int vfslock)
753 struct inode *inode = ccc_object_inode(obj);
756 cl_isize_lock(inode, 0);
757 cl_object_attr_lock(obj);
760 static void ccc_object_size_unlock(struct cl_object *obj, int vfslock)
762 struct inode *inode = ccc_object_inode(obj);
764 cl_object_attr_unlock(obj);
766 cl_isize_unlock(inode, 0);
770 * Helper function that if necessary adjusts file size (inode->i_size), when
771 * position at the offset \a pos is accessed. File size can be arbitrary stale
772 * on a Lustre client, but client at least knows KMS. If accessed area is
773 * inside [0, KMS], set file size to KMS, otherwise glimpse file size.
775 * Locking: cl_isize_lock is used to serialize changes to inode size and to
776 * protect consistency between inode size and cl_object
777 * attributes. cl_object_size_lock() protects consistency between cl_attr's of
778 * top-object and sub-objects.
780 * In page fault path cl_isize_lock cannot be taken, client has to live with
781 * the resulting races.
783 int ccc_prep_size(const struct lu_env *env, struct cl_object *obj,
784 struct cl_io *io, loff_t start, size_t count, int vfslock,
787 struct cl_attr *attr = &ccc_env_info(env)->cti_attr;
788 struct inode *inode = ccc_object_inode(obj);
789 loff_t pos = start + count - 1;
794 * Consistency guarantees: following possibilities exist for the
795 * relation between region being accessed and real file size at this
798 * (A): the region is completely inside of the file;
800 * (B-x): x bytes of region are inside of the file, the rest is
803 * (C): the region is completely outside of the file.
805 * This classification is stable under DLM lock already acquired by
806 * the caller, because to change the class, other client has to take
807 * DLM lock conflicting with our lock. Also, any updates to ->i_size
808 * by other threads on this client are serialized by
809 * ll_inode_size_lock(). This guarantees that short reads are handled
810 * correctly in the face of concurrent writes and truncates.
812 ccc_object_size_lock(obj, vfslock);
813 result = cl_object_attr_get(env, obj, attr);
818 * A glimpse is necessary to determine whether we
819 * return a short read (B) or some zeroes at the end
822 ccc_object_size_unlock(obj, vfslock);
823 result = cl_glimpse_lock(env, io, inode, obj);
824 if (result == 0 && exceed != NULL) {
825 /* If objective page index exceed end-of-file
826 * page index, return directly. Do not expect
827 * kernel will check such case correctly.
828 * linux-2.6.18-128.1.1 miss to do that.
830 size_t size = cl_isize_read(inode);
831 unsigned long cur_index = start >> CFS_PAGE_SHIFT;
833 if ((size == 0 && cur_index != 0) ||
834 (((size - 1) >> CFS_PAGE_SHIFT) < cur_index))
840 * region is within kms and, hence, within real file
841 * size (A). We need to increase i_size to cover the
842 * read region so that generic_file_read() will do its
843 * job, but that doesn't mean the kms size is
844 * _correct_, it is only the _minimum_ size. If
845 * someone does a stat they will get the correct size
846 * which will always be >= the kms value here.
850 * XXX in a page fault path, change inode size without
851 * ll_inode_size_lock() held! there is a race
852 * condition with truncate path. (see ll_extent_lock)
855 * XXX i_size_write() is not used because it is not
856 * safe to take the ll_inode_size_lock() due to a
857 * potential lock inversion (bug 6077). And since
858 * it's not safe to use i_size_write() without a
859 * covering mutex we do the assignment directly. It
860 * is not critical that the size be correct.
862 if (cl_isize_read(inode) < kms) {
864 cl_isize_write(inode, kms);
866 cl_isize_write_nolock(inode, kms);
870 ccc_object_size_unlock(obj, vfslock);
874 /*****************************************************************************
876 * Transfer operations.
880 void ccc_req_completion(const struct lu_env *env,
881 const struct cl_req_slice *slice, int ioret)
885 vrq = cl2ccc_req(slice);
886 OBD_SLAB_FREE_PTR(vrq, ccc_req_kmem);
890 * Implementation of struct cl_req_operations::cro_attr_set() for ccc
891 * layer. ccc is responsible for
897 * - o_fid (filled with inode number?!)
903 * - and IO epoch (stored in o_easize),
907 void ccc_req_attr_set(const struct lu_env *env,
908 const struct cl_req_slice *slice,
909 const struct cl_object *obj,
910 struct cl_req_attr *attr, obd_valid flags)
914 obd_flag valid_flags;
917 inode = ccc_object_inode(obj);
918 valid_flags = OBD_MD_FLTYPE|OBD_MD_FLATIME;
920 if (flags != (obd_valid)~0ULL)
921 valid_flags |= OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME;
923 LASSERT(attr->cra_capa == NULL);
924 attr->cra_capa = cl_capa_lookup(inode,
925 slice->crs_req->crq_type);
928 if (slice->crs_req->crq_type == CRT_WRITE) {
929 if (flags & OBD_MD_FLEPOCH) {
930 oa->o_valid |= OBD_MD_FLEPOCH;
931 oa->o_easize = cl_i2info(inode)->lli_ioepoch;
932 valid_flags |= OBD_MD_FLMTIME|OBD_MD_FLCTIME|
933 OBD_MD_FLUID|OBD_MD_FLGID|
934 OBD_MD_FLFID|OBD_MD_FLGENER;
937 obdo_from_inode(oa, inode, valid_flags & flags);
940 const struct cl_req_operations ccc_req_ops = {
941 .cro_attr_set = ccc_req_attr_set,
942 .cro_completion = ccc_req_completion
945 /* Setattr helpers */
946 int cl_setattr_do_truncate(struct inode *inode, loff_t size,
947 struct obd_capa *capa)
956 env = cl_env_get(&refcheck);
958 RETURN(PTR_ERR(env));
960 io = &ccc_env_info(env)->cti_io;
961 io->ci_obj = cl_i2info(inode)->lli_clob;
962 io->u.ci_truncate.tr_size = size;
963 io->u.ci_truncate.tr_capa = capa;
964 if (cl_io_init(env, io, CIT_TRUNC, io->ci_obj) == 0)
965 result = cl_io_loop(env, io);
967 result = io->ci_result;
969 cl_env_put(env, &refcheck);
973 int cl_setattr_ost(struct inode *inode, struct obd_capa *capa)
975 struct cl_inode_info *lli = cl_i2info(inode);
976 struct lov_stripe_md *lsm = lli->lli_smd;
979 struct obd_info oinfo = { { { 0 } } };
984 oa->o_id = lsm->lsm_object_id;
985 oa->o_gr = lsm->lsm_object_gr;
986 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
988 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
989 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
990 OBD_MD_FLFID | OBD_MD_FLGENER |
993 obdo_from_inode(oa, inode, flags);
998 /* XXX: this looks unnecessary now. */
999 rc = obd_setattr_rqset(cl_i2sbi(inode)->ll_dt_exp, &oinfo,
1002 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1011 /*****************************************************************************
1017 struct lu_device *ccc2lu_dev(struct ccc_device *vdv)
1019 return &vdv->cdv_cl.cd_lu_dev;
1022 struct ccc_device *lu2ccc_dev(const struct lu_device *d)
1024 return container_of0(d, struct ccc_device, cdv_cl.cd_lu_dev);
1027 struct ccc_device *cl2ccc_dev(const struct cl_device *d)
1029 return container_of0(d, struct ccc_device, cdv_cl);
1032 struct lu_object *ccc2lu(struct ccc_object *vob)
1034 return &vob->cob_cl.co_lu;
1037 struct ccc_object *lu2ccc(const struct lu_object *obj)
1039 return container_of0(obj, struct ccc_object, cob_cl.co_lu);
1042 struct ccc_object *cl2ccc(const struct cl_object *obj)
1044 return container_of0(obj, struct ccc_object, cob_cl);
1047 struct ccc_lock *cl2ccc_lock(const struct cl_lock_slice *slice)
1049 return container_of(slice, struct ccc_lock, clk_cl);
1052 struct ccc_io *cl2ccc_io(const struct lu_env *env,
1053 const struct cl_io_slice *slice)
1057 cio = container_of(slice, struct ccc_io, cui_cl);
1058 LASSERT(cio == ccc_env_io(env));
1062 struct ccc_req *cl2ccc_req(const struct cl_req_slice *slice)
1064 return container_of0(slice, struct ccc_req, crq_cl);
1067 cfs_page_t *cl2vm_page(const struct cl_page_slice *slice)
1069 return cl2ccc_page(slice)->cpg_page;
1072 /*****************************************************************************
1077 int ccc_object_invariant(const struct cl_object *obj)
1079 struct inode *inode = ccc_object_inode(obj);
1080 struct cl_inode_info *lli = cl_i2info(inode);
1082 return (S_ISREG(cl_inode_mode(inode)) ||
1083 /* i_mode of unlinked inode is zeroed. */
1084 cl_inode_mode(inode) == 0) && lli->lli_clob == obj;
1087 struct inode *ccc_object_inode(const struct cl_object *obj)
1089 return cl2ccc(obj)->cob_inode;
1093 * Returns a pointer to cl_page associated with \a vmpage, without acquiring
1094 * additional reference to the resulting page. This is an unsafe version of
1095 * cl_vmpage_page() that can only be used under vmpage lock.
1097 struct cl_page *ccc_vmpage_page_transient(cfs_page_t *vmpage)
1099 KLASSERT(PageLocked(vmpage));
1100 return (struct cl_page *)vmpage->private;
1104 * Initializes or updates CLIO part when new meta-data arrives from the
1107 * - allocates cl_object if necessary,
1108 * - updated layout, if object was already here.
1110 int cl_inode_init(struct inode *inode, struct lustre_md *md)
1113 struct cl_inode_info *lli;
1114 struct cl_object *clob;
1115 struct lu_site *site;
1117 const struct cl_object_conf conf = {
1126 /* LASSERT(inode->i_state & I_NEW); */
1127 LASSERT(md->body->valid & OBD_MD_FLID);
1129 if (!S_ISREG(cl_inode_mode(inode)))
1132 env = cl_env_get(&refcheck);
1134 return PTR_ERR(env);
1136 site = cl_i2sbi(inode)->ll_site;
1137 lli = cl_i2info(inode);
1138 fid = &lli->lli_fid;
1139 LASSERT(fid_is_sane(fid));
1141 if (lli->lli_clob == NULL) {
1142 clob = cl_object_find(env, lu2cl_dev(site->ls_top_dev),
1144 if (!IS_ERR(clob)) {
1146 * No locking is necessary, as new inode is
1147 * locked by I_NEW bit.
1149 * XXX not true for call from ll_update_inode().
1151 lli->lli_clob = clob;
1152 lu_object_ref_add(&clob->co_lu, "inode", inode);
1154 result = PTR_ERR(clob);
1156 result = cl_conf_set(env, lli->lli_clob, &conf);
1157 cl_env_put(env, &refcheck);
1160 CERROR("Failure to initialize cl object "DFID": %d\n",
1165 void cl_inode_fini(struct inode *inode)
1168 struct cl_inode_info *lli = cl_i2info(inode);
1169 struct cl_object *clob = lli->lli_clob;
1174 struct lu_object_header *head = clob->co_lu.lo_header;
1177 cookie = cl_env_reenter();
1178 env = cl_env_get(&refcheck);
1179 emergency = IS_ERR(env);
1181 mutex_lock(&ccc_inode_fini_guard);
1182 LASSERT(ccc_inode_fini_env != NULL);
1183 cl_env_implant(ccc_inode_fini_env, &refcheck);
1184 env = ccc_inode_fini_env;
1187 * cl_object cache is a slave to inode cache (which, in turn
1188 * is a slave to dentry cache), don't keep cl_object in memory
1189 * when its master is evicted.
1191 cl_object_kill(env, clob);
1192 lu_object_ref_del(&clob->co_lu, "inode", inode);
1193 /* XXX temporary: this is racy */
1194 LASSERT(atomic_read(&head->loh_ref) == 1);
1195 cl_object_put(env, clob);
1196 lli->lli_clob = NULL;
1198 cl_env_unplant(ccc_inode_fini_env, &refcheck);
1199 mutex_unlock(&ccc_inode_fini_guard);
1201 cl_env_put(env, &refcheck);
1202 cl_env_reexit(cookie);
1207 * return IF_* type for given lu_dirent entry.
1208 * IF_* flag shld be converted to particular OS file type in
1209 * platform llite module.
1211 __u16 ll_dirent_type_get(struct lu_dirent *ent)
1214 struct luda_type *lt;
1217 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
1218 const unsigned align = sizeof(struct luda_type) - 1;
1220 len = le16_to_cpu(ent->lde_namelen);
1221 len = (len + align) & ~align;
1222 lt = (void *) ent->lde_name + len;
1223 type = CFS_IFTODT(le16_to_cpu(lt->lt_type));