diff options
Diffstat (limited to 'drivers/staging/lustre/lustre/obdclass')
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/cl_io.c | 430 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/cl_lock.c | 2086 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/cl_object.c | 303 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/cl_page.c | 659 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/class_obd.c | 5 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/debug.c | 4 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/genops.c | 1 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/linux/linux-module.c | 4 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/llog.c | 1 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/lprocfs_status.c | 72 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/lu_object.c | 9 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/lustre_peer.c | 3 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/obd_config.c | 26 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/obd_mount.c | 15 | ||||
-rw-r--r-- | drivers/staging/lustre/lustre/obdclass/obdo.c | 3 |
15 files changed, 628 insertions, 2993 deletions
diff --git a/drivers/staging/lustre/lustre/obdclass/cl_io.c b/drivers/staging/lustre/lustre/obdclass/cl_io.c index f5128b4f176f..583fb5f33889 100644 --- a/drivers/staging/lustre/lustre/obdclass/cl_io.c +++ b/drivers/staging/lustre/lustre/obdclass/cl_io.c @@ -36,6 +36,7 @@ * Client IO. * * Author: Nikita Danilov <nikita.danilov@sun.com> + * Author: Jinshan Xiong <jinshan.xiong@intel.com> */ #define DEBUG_SUBSYSTEM S_CLASS @@ -132,6 +133,7 @@ void cl_io_fini(const struct lu_env *env, struct cl_io *io) case CIT_WRITE: break; case CIT_FAULT: + break; case CIT_FSYNC: LASSERT(!io->ci_need_restart); break; @@ -159,7 +161,6 @@ static int cl_io_init0(const struct lu_env *env, struct cl_io *io, io->ci_type = iot; INIT_LIST_HEAD(&io->ci_lockset.cls_todo); - INIT_LIST_HEAD(&io->ci_lockset.cls_curr); INIT_LIST_HEAD(&io->ci_lockset.cls_done); INIT_LIST_HEAD(&io->ci_layers); @@ -241,37 +242,7 @@ static int cl_lock_descr_sort(const struct cl_lock_descr *d0, const struct cl_lock_descr *d1) { return lu_fid_cmp(lu_object_fid(&d0->cld_obj->co_lu), - lu_object_fid(&d1->cld_obj->co_lu)) ?: - __diff_normalize(d0->cld_start, d1->cld_start); -} - -static int cl_lock_descr_cmp(const struct cl_lock_descr *d0, - const struct cl_lock_descr *d1) -{ - int ret; - - ret = lu_fid_cmp(lu_object_fid(&d0->cld_obj->co_lu), - lu_object_fid(&d1->cld_obj->co_lu)); - if (ret) - return ret; - if (d0->cld_end < d1->cld_start) - return -1; - if (d0->cld_start > d0->cld_end) - return 1; - return 0; -} - -static void cl_lock_descr_merge(struct cl_lock_descr *d0, - const struct cl_lock_descr *d1) -{ - d0->cld_start = min(d0->cld_start, d1->cld_start); - d0->cld_end = max(d0->cld_end, d1->cld_end); - - if (d1->cld_mode == CLM_WRITE && d0->cld_mode != CLM_WRITE) - d0->cld_mode = CLM_WRITE; - - if (d1->cld_mode == CLM_GROUP && d0->cld_mode != CLM_GROUP) - d0->cld_mode = CLM_GROUP; + lu_object_fid(&d1->cld_obj->co_lu)); } /* @@ -320,33 +291,35 @@ static void cl_io_locks_sort(struct cl_io *io) } while (!done); } -/** - * Check whether \a queue contains locks matching \a need. - * - * \retval +ve there is a matching lock in the \a queue - * \retval 0 there are no matching locks in the \a queue - */ -int cl_queue_match(const struct list_head *queue, - const struct cl_lock_descr *need) +static void cl_lock_descr_merge(struct cl_lock_descr *d0, + const struct cl_lock_descr *d1) { - struct cl_io_lock_link *scan; + d0->cld_start = min(d0->cld_start, d1->cld_start); + d0->cld_end = max(d0->cld_end, d1->cld_end); - list_for_each_entry(scan, queue, cill_linkage) { - if (cl_lock_descr_match(&scan->cill_descr, need)) - return 1; - } - return 0; + if (d1->cld_mode == CLM_WRITE && d0->cld_mode != CLM_WRITE) + d0->cld_mode = CLM_WRITE; + + if (d1->cld_mode == CLM_GROUP && d0->cld_mode != CLM_GROUP) + d0->cld_mode = CLM_GROUP; } -EXPORT_SYMBOL(cl_queue_match); -static int cl_queue_merge(const struct list_head *queue, - const struct cl_lock_descr *need) +static int cl_lockset_merge(const struct cl_lockset *set, + const struct cl_lock_descr *need) { struct cl_io_lock_link *scan; - list_for_each_entry(scan, queue, cill_linkage) { - if (cl_lock_descr_cmp(&scan->cill_descr, need)) + list_for_each_entry(scan, &set->cls_todo, cill_linkage) { + if (!cl_object_same(scan->cill_descr.cld_obj, need->cld_obj)) continue; + + /* Merge locks for the same object because ldlm lock server + * may expand the lock extent, otherwise there is a deadlock + * case if two conflicted locks are queueud for the same object + * and lock server expands one lock to overlap the another. + * The side effect is that it can generate a multi-stripe lock + * that may cause casacading problem + */ cl_lock_descr_merge(&scan->cill_descr, need); CDEBUG(D_VFSTRACE, "lock: %d: [%lu, %lu]\n", scan->cill_descr.cld_mode, scan->cill_descr.cld_start, @@ -356,87 +329,20 @@ static int cl_queue_merge(const struct list_head *queue, return 0; } -static int cl_lockset_match(const struct cl_lockset *set, - const struct cl_lock_descr *need) -{ - return cl_queue_match(&set->cls_curr, need) || - cl_queue_match(&set->cls_done, need); -} - -static int cl_lockset_merge(const struct cl_lockset *set, - const struct cl_lock_descr *need) -{ - return cl_queue_merge(&set->cls_todo, need) || - cl_lockset_match(set, need); -} - -static int cl_lockset_lock_one(const struct lu_env *env, - struct cl_io *io, struct cl_lockset *set, - struct cl_io_lock_link *link) -{ - struct cl_lock *lock; - int result; - - lock = cl_lock_request(env, io, &link->cill_descr, "io", io); - - if (!IS_ERR(lock)) { - link->cill_lock = lock; - list_move(&link->cill_linkage, &set->cls_curr); - if (!(link->cill_descr.cld_enq_flags & CEF_ASYNC)) { - result = cl_wait(env, lock); - if (result == 0) - list_move(&link->cill_linkage, &set->cls_done); - } else - result = 0; - } else - result = PTR_ERR(lock); - return result; -} - -static void cl_lock_link_fini(const struct lu_env *env, struct cl_io *io, - struct cl_io_lock_link *link) -{ - struct cl_lock *lock = link->cill_lock; - - list_del_init(&link->cill_linkage); - if (lock) { - cl_lock_release(env, lock, "io", io); - link->cill_lock = NULL; - } - if (link->cill_fini) - link->cill_fini(env, link); -} - static int cl_lockset_lock(const struct lu_env *env, struct cl_io *io, struct cl_lockset *set) { struct cl_io_lock_link *link; struct cl_io_lock_link *temp; - struct cl_lock *lock; int result; result = 0; list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage) { - if (!cl_lockset_match(set, &link->cill_descr)) { - /* XXX some locking to guarantee that locks aren't - * expanded in between. - */ - result = cl_lockset_lock_one(env, io, set, link); - if (result != 0) - break; - } else - cl_lock_link_fini(env, io, link); - } - if (result == 0) { - list_for_each_entry_safe(link, temp, - &set->cls_curr, cill_linkage) { - lock = link->cill_lock; - result = cl_wait(env, lock); - if (result == 0) - list_move(&link->cill_linkage, &set->cls_done); - else - break; - } + result = cl_lock_request(env, io, &link->cill_lock); + if (result < 0) + break; + + list_move(&link->cill_linkage, &set->cls_done); } return result; } @@ -492,16 +398,19 @@ void cl_io_unlock(const struct lu_env *env, struct cl_io *io) set = &io->ci_lockset; - list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage) - cl_lock_link_fini(env, io, link); - - list_for_each_entry_safe(link, temp, &set->cls_curr, cill_linkage) - cl_lock_link_fini(env, io, link); + list_for_each_entry_safe(link, temp, &set->cls_todo, cill_linkage) { + list_del_init(&link->cill_linkage); + if (link->cill_fini) + link->cill_fini(env, link); + } list_for_each_entry_safe(link, temp, &set->cls_done, cill_linkage) { - cl_unuse(env, link->cill_lock); - cl_lock_link_fini(env, io, link); + list_del_init(&link->cill_linkage); + cl_lock_release(env, &link->cill_lock); + if (link->cill_fini) + link->cill_fini(env, link); } + cl_io_for_each_reverse(scan, io) { if (scan->cis_iop->op[io->ci_type].cio_unlock) scan->cis_iop->op[io->ci_type].cio_unlock(env, scan); @@ -595,9 +504,9 @@ int cl_io_lock_add(const struct lu_env *env, struct cl_io *io, { int result; - if (cl_lockset_merge(&io->ci_lockset, &link->cill_descr)) + if (cl_lockset_merge(&io->ci_lockset, &link->cill_descr)) { result = 1; - else { + } else { list_add(&link->cill_linkage, &io->ci_lockset.cls_todo); result = 0; } @@ -627,8 +536,9 @@ int cl_io_lock_alloc_add(const struct lu_env *env, struct cl_io *io, result = cl_io_lock_add(env, io, link); if (result) /* lock match */ link->cill_fini(env, link); - } else + } else { result = -ENOMEM; + } return result; } @@ -692,42 +602,6 @@ cl_io_slice_page(const struct cl_io_slice *ios, struct cl_page *page) } /** - * True iff \a page is within \a io range. - */ -static int cl_page_in_io(const struct cl_page *page, const struct cl_io *io) -{ - int result = 1; - loff_t start; - loff_t end; - pgoff_t idx; - - idx = page->cp_index; - switch (io->ci_type) { - case CIT_READ: - case CIT_WRITE: - /* - * check that [start, end) and [pos, pos + count) extents - * overlap. - */ - if (!cl_io_is_append(io)) { - const struct cl_io_rw_common *crw = &(io->u.ci_rw); - - start = cl_offset(page->cp_obj, idx); - end = cl_offset(page->cp_obj, idx + 1); - result = crw->crw_pos < end && - start < crw->crw_pos + crw->crw_count; - } - break; - case CIT_FAULT: - result = io->u.ci_fault.ft_index == idx; - break; - default: - LBUG(); - } - return result; -} - -/** * Called by read io, when page has to be read from the server. * * \see cl_io_operations::cio_read_page() @@ -742,7 +616,6 @@ int cl_io_read_page(const struct lu_env *env, struct cl_io *io, LINVRNT(io->ci_type == CIT_READ || io->ci_type == CIT_FAULT); LINVRNT(cl_page_is_owned(page, io)); LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED); - LINVRNT(cl_page_in_io(page, io)); LINVRNT(cl_io_invariant(io)); queue = &io->ci_queue; @@ -769,7 +642,7 @@ int cl_io_read_page(const struct lu_env *env, struct cl_io *io, break; } } - if (result == 0) + if (result == 0 && queue->c2_qin.pl_nr > 0) result = cl_io_submit_rw(env, io, CRT_READ, queue); /* * Unlock unsent pages in case of error. @@ -781,77 +654,29 @@ int cl_io_read_page(const struct lu_env *env, struct cl_io *io, EXPORT_SYMBOL(cl_io_read_page); /** - * Called by write io to prepare page to receive data from user buffer. + * Commit a list of contiguous pages into writeback cache. * - * \see cl_io_operations::cio_prepare_write() + * \returns 0 if all pages committed, or errcode if error occurred. + * \see cl_io_operations::cio_commit_async() */ -int cl_io_prepare_write(const struct lu_env *env, struct cl_io *io, - struct cl_page *page, unsigned from, unsigned to) +int cl_io_commit_async(const struct lu_env *env, struct cl_io *io, + struct cl_page_list *queue, int from, int to, + cl_commit_cbt cb) { const struct cl_io_slice *scan; int result = 0; - LINVRNT(io->ci_type == CIT_WRITE); - LINVRNT(cl_page_is_owned(page, io)); - LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED); - LINVRNT(cl_io_invariant(io)); - LASSERT(cl_page_in_io(page, io)); - - cl_io_for_each_reverse(scan, io) { - if (scan->cis_iop->cio_prepare_write) { - const struct cl_page_slice *slice; - - slice = cl_io_slice_page(scan, page); - result = scan->cis_iop->cio_prepare_write(env, scan, - slice, - from, to); - if (result != 0) - break; - } - } - return result; -} -EXPORT_SYMBOL(cl_io_prepare_write); - -/** - * Called by write io after user data were copied into a page. - * - * \see cl_io_operations::cio_commit_write() - */ -int cl_io_commit_write(const struct lu_env *env, struct cl_io *io, - struct cl_page *page, unsigned from, unsigned to) -{ - const struct cl_io_slice *scan; - int result = 0; - - LINVRNT(io->ci_type == CIT_WRITE); - LINVRNT(io->ci_state == CIS_IO_GOING || io->ci_state == CIS_LOCKED); - LINVRNT(cl_io_invariant(io)); - /* - * XXX Uh... not nice. Top level cl_io_commit_write() call (vvp->lov) - * already called cl_page_cache_add(), moving page into CPS_CACHED - * state. Better (and more general) way of dealing with such situation - * is needed. - */ - LASSERT(cl_page_is_owned(page, io) || page->cp_parent); - LASSERT(cl_page_in_io(page, io)); - cl_io_for_each(scan, io) { - if (scan->cis_iop->cio_commit_write) { - const struct cl_page_slice *slice; - - slice = cl_io_slice_page(scan, page); - result = scan->cis_iop->cio_commit_write(env, scan, - slice, - from, to); - if (result != 0) - break; - } + if (!scan->cis_iop->cio_commit_async) + continue; + result = scan->cis_iop->cio_commit_async(env, scan, queue, + from, to, cb); + if (result != 0) + break; } - LINVRNT(result <= 0); return result; } -EXPORT_SYMBOL(cl_io_commit_write); +EXPORT_SYMBOL(cl_io_commit_async); /** * Submits a list of pages for immediate io. @@ -869,13 +694,10 @@ int cl_io_submit_rw(const struct lu_env *env, struct cl_io *io, const struct cl_io_slice *scan; int result = 0; - LINVRNT(crt < ARRAY_SIZE(scan->cis_iop->req_op)); - cl_io_for_each(scan, io) { - if (!scan->cis_iop->req_op[crt].cio_submit) + if (!scan->cis_iop->cio_submit) continue; - result = scan->cis_iop->req_op[crt].cio_submit(env, scan, crt, - queue); + result = scan->cis_iop->cio_submit(env, scan, crt, queue); if (result != 0) break; } @@ -887,6 +709,9 @@ int cl_io_submit_rw(const struct lu_env *env, struct cl_io *io, } EXPORT_SYMBOL(cl_io_submit_rw); +static void cl_page_list_assume(const struct lu_env *env, + struct cl_io *io, struct cl_page_list *plist); + /** * Submit a sync_io and wait for the IO to be finished, or error happens. * If \a timeout is zero, it means to wait for the IO unconditionally. @@ -904,7 +729,7 @@ int cl_io_submit_sync(const struct lu_env *env, struct cl_io *io, pg->cp_sync_io = anchor; } - cl_sync_io_init(anchor, queue->c2_qin.pl_nr); + cl_sync_io_init(anchor, queue->c2_qin.pl_nr, &cl_sync_io_end); rc = cl_io_submit_rw(env, io, iot, queue); if (rc == 0) { /* @@ -915,12 +740,12 @@ int cl_io_submit_sync(const struct lu_env *env, struct cl_io *io, */ cl_page_list_for_each(pg, &queue->c2_qin) { pg->cp_sync_io = NULL; - cl_sync_io_note(anchor, 1); + cl_sync_io_note(env, anchor, 1); } /* wait for the IO to be finished. */ - rc = cl_sync_io_wait(env, io, &queue->c2_qout, - anchor, timeout); + rc = cl_sync_io_wait(env, anchor, timeout); + cl_page_list_assume(env, io, &queue->c2_qout); } else { LASSERT(list_empty(&queue->c2_qout.pl_pages)); cl_page_list_for_each(pg, &queue->c2_qin) @@ -931,26 +756,6 @@ int cl_io_submit_sync(const struct lu_env *env, struct cl_io *io, EXPORT_SYMBOL(cl_io_submit_sync); /** - * Cancel an IO which has been submitted by cl_io_submit_rw. - */ -static int cl_io_cancel(const struct lu_env *env, struct cl_io *io, - struct cl_page_list *queue) -{ - struct cl_page *page; - int result = 0; - - CERROR("Canceling ongoing page transmission\n"); - cl_page_list_for_each(page, queue) { - int rc; - - LINVRNT(cl_page_in_io(page, io)); - rc = cl_page_cancel(env, page); - result = result ?: rc; - } - return result; -} - -/** * Main io loop. * * Pumps io through iterations calling @@ -1072,8 +877,8 @@ EXPORT_SYMBOL(cl_page_list_add); /** * Removes a page from a page list. */ -static void cl_page_list_del(const struct lu_env *env, - struct cl_page_list *plist, struct cl_page *page) +void cl_page_list_del(const struct lu_env *env, struct cl_page_list *plist, + struct cl_page *page) { LASSERT(plist->pl_nr > 0); LINVRNT(plist->pl_owner == current); @@ -1086,6 +891,7 @@ static void cl_page_list_del(const struct lu_env *env, lu_ref_del_at(&page->cp_reference, &page->cp_queue_ref, "queue", plist); cl_page_put(env, page); } +EXPORT_SYMBOL(cl_page_list_del); /** * Moves a page from one page list to another. @@ -1106,6 +912,24 @@ void cl_page_list_move(struct cl_page_list *dst, struct cl_page_list *src, EXPORT_SYMBOL(cl_page_list_move); /** + * Moves a page from one page list to the head of another list. + */ +void cl_page_list_move_head(struct cl_page_list *dst, struct cl_page_list *src, + struct cl_page *page) +{ + LASSERT(src->pl_nr > 0); + LINVRNT(dst->pl_owner == current); + LINVRNT(src->pl_owner == current); + + list_move(&page->cp_batch, &dst->pl_pages); + --src->pl_nr; + ++dst->pl_nr; + lu_ref_set_at(&page->cp_reference, &page->cp_queue_ref, "queue", + src, dst); +} +EXPORT_SYMBOL(cl_page_list_move_head); + +/** * splice the cl_page_list, just as list head does */ void cl_page_list_splice(struct cl_page_list *list, struct cl_page_list *head) @@ -1162,8 +986,7 @@ EXPORT_SYMBOL(cl_page_list_disown); /** * Releases pages from queue. */ -static void cl_page_list_fini(const struct lu_env *env, - struct cl_page_list *plist) +void cl_page_list_fini(const struct lu_env *env, struct cl_page_list *plist) { struct cl_page *page; struct cl_page *temp; @@ -1174,6 +997,7 @@ static void cl_page_list_fini(const struct lu_env *env, cl_page_list_del(env, plist, page); LASSERT(plist->pl_nr == 0); } +EXPORT_SYMBOL(cl_page_list_fini); /** * Assumes all pages in a queue. @@ -1260,7 +1084,7 @@ EXPORT_SYMBOL(cl_2queue_init_page); /** * Returns top-level io. * - * \see cl_object_top(), cl_page_top(). + * \see cl_object_top() */ struct cl_io *cl_io_top(struct cl_io *io) { @@ -1323,19 +1147,14 @@ static int cl_req_init(const struct lu_env *env, struct cl_req *req, int result; result = 0; - page = cl_page_top(page); - do { - list_for_each_entry(slice, &page->cp_layers, cpl_linkage) { - dev = lu2cl_dev(slice->cpl_obj->co_lu.lo_dev); - if (dev->cd_ops->cdo_req_init) { - result = dev->cd_ops->cdo_req_init(env, - dev, req); - if (result != 0) - break; - } + list_for_each_entry(slice, &page->cp_layers, cpl_linkage) { + dev = lu2cl_dev(slice->cpl_obj->co_lu.lo_dev); + if (dev->cd_ops->cdo_req_init) { + result = dev->cd_ops->cdo_req_init(env, dev, req); + if (result != 0) + break; } - page = page->cp_child; - } while (page && result == 0); + } return result; } @@ -1384,14 +1203,16 @@ struct cl_req *cl_req_alloc(const struct lu_env *env, struct cl_page *page, if (req->crq_o) { req->crq_nrobjs = nr_objects; result = cl_req_init(env, req, page); - } else + } else { result = -ENOMEM; + } if (result != 0) { cl_req_completion(env, req, result); req = ERR_PTR(result); } - } else + } else { req = ERR_PTR(-ENOMEM); + } return req; } EXPORT_SYMBOL(cl_req_alloc); @@ -1406,8 +1227,6 @@ void cl_req_page_add(const struct lu_env *env, struct cl_req_obj *rqo; int i; - page = cl_page_top(page); - LASSERT(list_empty(&page->cp_flight)); LASSERT(!page->cp_req); @@ -1438,8 +1257,6 @@ void cl_req_page_done(const struct lu_env *env, struct cl_page *page) { struct cl_req *req = page->cp_req; - page = cl_page_top(page); - LASSERT(!list_empty(&page->cp_flight)); LASSERT(req->crq_nrpages > 0); @@ -1511,25 +1328,39 @@ void cl_req_attr_set(const struct lu_env *env, struct cl_req *req, } EXPORT_SYMBOL(cl_req_attr_set); +/* cl_sync_io_callback assumes the caller must call cl_sync_io_wait() to + * wait for the IO to finish. + */ +void cl_sync_io_end(const struct lu_env *env, struct cl_sync_io *anchor) +{ + wake_up_all(&anchor->csi_waitq); + + /* it's safe to nuke or reuse anchor now */ + atomic_set(&anchor->csi_barrier, 0); +} +EXPORT_SYMBOL(cl_sync_io_end); /** - * Initialize synchronous io wait anchor, for transfer of \a nrpages pages. + * Initialize synchronous io wait anchor */ -void cl_sync_io_init(struct cl_sync_io *anchor, int nrpages) +void cl_sync_io_init(struct cl_sync_io *anchor, int nr, + void (*end)(const struct lu_env *, struct cl_sync_io *)) { + memset(anchor, 0, sizeof(*anchor)); init_waitqueue_head(&anchor->csi_waitq); - atomic_set(&anchor->csi_sync_nr, nrpages); - atomic_set(&anchor->csi_barrier, nrpages > 0); + atomic_set(&anchor->csi_sync_nr, nr); + atomic_set(&anchor->csi_barrier, nr > 0); anchor->csi_sync_rc = 0; + anchor->csi_end_io = end; + LASSERT(end); } EXPORT_SYMBOL(cl_sync_io_init); /** - * Wait until all transfer completes. Transfer completion routine has to call - * cl_sync_io_note() for every page. + * Wait until all IO completes. Transfer completion routine has to call + * cl_sync_io_note() for every entity. */ -int cl_sync_io_wait(const struct lu_env *env, struct cl_io *io, - struct cl_page_list *queue, struct cl_sync_io *anchor, +int cl_sync_io_wait(const struct lu_env *env, struct cl_sync_io *anchor, long timeout) { struct l_wait_info lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(timeout), @@ -1542,11 +1373,9 @@ int cl_sync_io_wait(const struct lu_env *env, struct cl_io *io, atomic_read(&anchor->csi_sync_nr) == 0, &lwi); if (rc < 0) { - CERROR("SYNC IO failed with error: %d, try to cancel %d remaining pages\n", + CERROR("IO failed: %d, still wait for %d remaining entries\n", rc, atomic_read(&anchor->csi_sync_nr)); - (void)cl_io_cancel(env, io, queue); - lwi = (struct l_wait_info) { 0 }; (void)l_wait_event(anchor->csi_waitq, atomic_read(&anchor->csi_sync_nr) == 0, @@ -1555,14 +1384,12 @@ int cl_sync_io_wait(const struct lu_env *env, struct cl_io *io, rc = anchor->csi_sync_rc; } LASSERT(atomic_read(&anchor->csi_sync_nr) == 0); - cl_page_list_assume(env, io, queue); /* wait until cl_sync_io_note() has done wakeup */ while (unlikely(atomic_read(&anchor->csi_barrier) != 0)) { cpu_relax(); } - POISON(anchor, 0x5a, sizeof(*anchor)); return rc; } EXPORT_SYMBOL(cl_sync_io_wait); @@ -1570,7 +1397,8 @@ EXPORT_SYMBOL(cl_sync_io_wait); /** * Indicate that transfer of a single page completed. */ -void cl_sync_io_note(struct cl_sync_io *anchor, int ioret) +void cl_sync_io_note(const struct lu_env *env, struct cl_sync_io *anchor, + int ioret) { if (anchor->csi_sync_rc == 0 && ioret < 0) anchor->csi_sync_rc = ioret; @@ -1581,9 +1409,9 @@ void cl_sync_io_note(struct cl_sync_io *anchor, int ioret) */ LASSERT(atomic_read(&anchor->csi_sync_nr) > 0); if (atomic_dec_and_test(&anchor->csi_sync_nr)) { - wake_up_all(&anchor->csi_waitq); - /* it's safe to nuke or reuse anchor now */ - atomic_set(&anchor->csi_barrier, 0); + LASSERT(anchor->csi_end_io); + anchor->csi_end_io(env, anchor); + /* Can't access anchor any more */ } } EXPORT_SYMBOL(cl_sync_io_note); diff --git a/drivers/staging/lustre/lustre/obdclass/cl_lock.c b/drivers/staging/lustre/lustre/obdclass/cl_lock.c index aec644eb4db9..26a576b63a72 100644 --- a/drivers/staging/lustre/lustre/obdclass/cl_lock.c +++ b/drivers/staging/lustre/lustre/obdclass/cl_lock.c @@ -36,6 +36,7 @@ * Client Extent Lock. * * Author: Nikita Danilov <nikita.danilov@sun.com> + * Author: Jinshan Xiong <jinshan.xiong@intel.com> */ #define DEBUG_SUBSYSTEM S_CLASS @@ -47,138 +48,18 @@ #include "../include/cl_object.h" #include "cl_internal.h" -/** Lock class of cl_lock::cll_guard */ -static struct lock_class_key cl_lock_guard_class; -static struct kmem_cache *cl_lock_kmem; - -static struct lu_kmem_descr cl_lock_caches[] = { - { - .ckd_cache = &cl_lock_kmem, - .ckd_name = "cl_lock_kmem", - .ckd_size = sizeof (struct cl_lock) - }, - { - .ckd_cache = NULL - } -}; - -#define CS_LOCK_INC(o, item) -#define CS_LOCK_DEC(o, item) -#define CS_LOCKSTATE_INC(o, state) -#define CS_LOCKSTATE_DEC(o, state) - -/** - * Basic lock invariant that is maintained at all times. Caller either has a - * reference to \a lock, or somehow assures that \a lock cannot be freed. - * - * \see cl_lock_invariant() - */ -static int cl_lock_invariant_trusted(const struct lu_env *env, - const struct cl_lock *lock) -{ - return ergo(lock->cll_state == CLS_FREEING, lock->cll_holds == 0) && - atomic_read(&lock->cll_ref) >= lock->cll_holds && - lock->cll_holds >= lock->cll_users && - lock->cll_holds >= 0 && - lock->cll_users >= 0 && - lock->cll_depth >= 0; -} - -/** - * Stronger lock invariant, checking that caller has a reference on a lock. - * - * \see cl_lock_invariant_trusted() - */ -static int cl_lock_invariant(const struct lu_env *env, - const struct cl_lock *lock) -{ - int result; - - result = atomic_read(&lock->cll_ref) > 0 && - cl_lock_invariant_trusted(env, lock); - if (!result && env) - CL_LOCK_DEBUG(D_ERROR, env, lock, "invariant broken\n"); - return result; -} - -/** - * Returns lock "nesting": 0 for a top-lock and 1 for a sub-lock. - */ -static enum clt_nesting_level cl_lock_nesting(const struct cl_lock *lock) -{ - return cl_object_header(lock->cll_descr.cld_obj)->coh_nesting; -} - -/** - * Returns a set of counters for this lock, depending on a lock nesting. - */ -static struct cl_thread_counters *cl_lock_counters(const struct lu_env *env, - const struct cl_lock *lock) -{ - struct cl_thread_info *info; - enum clt_nesting_level nesting; - - info = cl_env_info(env); - nesting = cl_lock_nesting(lock); - LASSERT(nesting < ARRAY_SIZE(info->clt_counters)); - return &info->clt_counters[nesting]; -} - static void cl_lock_trace0(int level, const struct lu_env *env, const char *prefix, const struct cl_lock *lock, const char *func, const int line) { struct cl_object_header *h = cl_object_header(lock->cll_descr.cld_obj); - CDEBUG(level, "%s: %p@(%d %p %d %d %d %d %d %lx)(%p/%d/%d) at %s():%d\n", - prefix, lock, atomic_read(&lock->cll_ref), - lock->cll_guarder, lock->cll_depth, - lock->cll_state, lock->cll_error, lock->cll_holds, - lock->cll_users, lock->cll_flags, - env, h->coh_nesting, cl_lock_nr_mutexed(env), - func, line); + CDEBUG(level, "%s: %p (%p/%d) at %s():%d\n", + prefix, lock, env, h->coh_nesting, func, line); } - -#define cl_lock_trace(level, env, prefix, lock) \ +#define cl_lock_trace(level, env, prefix, lock) \ cl_lock_trace0(level, env, prefix, lock, __func__, __LINE__) -#define RETIP ((unsigned long)__builtin_return_address(0)) - -#ifdef CONFIG_LOCKDEP -static struct lock_class_key cl_lock_key; - -static void cl_lock_lockdep_init(struct cl_lock *lock) -{ - lockdep_set_class_and_name(lock, &cl_lock_key, "EXT"); -} - -static void cl_lock_lockdep_acquire(const struct lu_env *env, - struct cl_lock *lock, __u32 enqflags) -{ - cl_lock_counters(env, lock)->ctc_nr_locks_acquired++; - lock_map_acquire(&lock->dep_map); -} - -static void cl_lock_lockdep_release(const struct lu_env *env, - struct cl_lock *lock) -{ - cl_lock_counters(env, lock)->ctc_nr_locks_acquired--; - lock_release(&lock->dep_map, 0, RETIP); -} - -#else /* !CONFIG_LOCKDEP */ - -static void cl_lock_lockdep_init(struct cl_lock *lock) -{} -static void cl_lock_lockdep_acquire(const struct lu_env *env, - struct cl_lock *lock, __u32 enqflags) -{} -static void cl_lock_lockdep_release(const struct lu_env *env, - struct cl_lock *lock) -{} - -#endif /* !CONFIG_LOCKDEP */ - /** * Adds lock slice to the compound lock. * @@ -199,62 +80,10 @@ void cl_lock_slice_add(struct cl_lock *lock, struct cl_lock_slice *slice, } EXPORT_SYMBOL(cl_lock_slice_add); -/** - * Returns true iff a lock with the mode \a has provides at least the same - * guarantees as a lock with the mode \a need. - */ -int cl_lock_mode_match(enum cl_lock_mode has, enum cl_lock_mode need) -{ - LINVRNT(need == CLM_READ || need == CLM_WRITE || - need == CLM_PHANTOM || need == CLM_GROUP); - LINVRNT(has == CLM_READ || has == CLM_WRITE || - has == CLM_PHANTOM || has == CLM_GROUP); - CLASSERT(CLM_PHANTOM < CLM_READ); - CLASSERT(CLM_READ < CLM_WRITE); - CLASSERT(CLM_WRITE < CLM_GROUP); - - if (has != CLM_GROUP) - return need <= has; - else - return need == has; -} -EXPORT_SYMBOL(cl_lock_mode_match); - -/** - * Returns true iff extent portions of lock descriptions match. - */ -int cl_lock_ext_match(const struct cl_lock_descr *has, - const struct cl_lock_descr *need) -{ - return - has->cld_start <= need->cld_start && - has->cld_end >= need->cld_end && - cl_lock_mode_match(has->cld_mode, need->cld_mode) && - (has->cld_mode != CLM_GROUP || has->cld_gid == need->cld_gid); -} -EXPORT_SYMBOL(cl_lock_ext_match); - -/** - * Returns true iff a lock with the description \a has provides at least the - * same guarantees as a lock with the description \a need. - */ -int cl_lock_descr_match(const struct cl_lock_descr *has, - const struct cl_lock_descr *need) +void cl_lock_fini(const struct lu_env *env, struct cl_lock *lock) { - return - cl_object_same(has->cld_obj, need->cld_obj) && - cl_lock_ext_match(has, need); -} -EXPORT_SYMBOL(cl_lock_descr_match); + cl_lock_trace(D_DLMTRACE, env, "destroy lock", lock); -static void cl_lock_free(const struct lu_env *env, struct cl_lock *lock) -{ - struct cl_object *obj = lock->cll_descr.cld_obj; - - LINVRNT(!cl_lock_is_mutexed(lock)); - - cl_lock_trace(D_DLMTRACE, env, "free lock", lock); - might_sleep(); while (!list_empty(&lock->cll_layers)) { struct cl_lock_slice *slice; @@ -263,350 +92,36 @@ static void cl_lock_free(const struct lu_env *env, struct cl_lock *lock) list_del_init(lock->cll_layers.next); slice->cls_ops->clo_fini(env, slice); } - CS_LOCK_DEC(obj, total); - CS_LOCKSTATE_DEC(obj, lock->cll_state); - lu_object_ref_del_at(&obj->co_lu, &lock->cll_obj_ref, "cl_lock", lock); - cl_object_put(env, obj); - lu_ref_fini(&lock->cll_reference); - lu_ref_fini(&lock->cll_holders); - mutex_destroy(&lock->cll_guard); - kmem_cache_free(cl_lock_kmem, lock); -} - -/** - * Releases a reference on a lock. - * - * When last reference is released, lock is returned to the cache, unless it - * is in cl_lock_state::CLS_FREEING state, in which case it is destroyed - * immediately. - * - * \see cl_object_put(), cl_page_put() - */ -void cl_lock_put(const struct lu_env *env, struct cl_lock *lock) -{ - struct cl_object *obj; - - LINVRNT(cl_lock_invariant(env, lock)); - obj = lock->cll_descr.cld_obj; - LINVRNT(obj); - - CDEBUG(D_TRACE, "releasing reference: %d %p %lu\n", - atomic_read(&lock->cll_ref), lock, RETIP); - - if (atomic_dec_and_test(&lock->cll_ref)) { - if (lock->cll_state == CLS_FREEING) { - LASSERT(list_empty(&lock->cll_linkage)); - cl_lock_free(env, lock); - } - CS_LOCK_DEC(obj, busy); - } -} -EXPORT_SYMBOL(cl_lock_put); - -/** - * Acquires an additional reference to a lock. - * - * This can be called only by caller already possessing a reference to \a - * lock. - * - * \see cl_object_get(), cl_page_get() - */ -void cl_lock_get(struct cl_lock *lock) -{ - LINVRNT(cl_lock_invariant(NULL, lock)); - CDEBUG(D_TRACE, "acquiring reference: %d %p %lu\n", - atomic_read(&lock->cll_ref), lock, RETIP); - atomic_inc(&lock->cll_ref); -} -EXPORT_SYMBOL(cl_lock_get); - -/** - * Acquires a reference to a lock. - * - * This is much like cl_lock_get(), except that this function can be used to - * acquire initial reference to the cached lock. Caller has to deal with all - * possible races. Use with care! - * - * \see cl_page_get_trust() - */ -void cl_lock_get_trust(struct cl_lock *lock) -{ - CDEBUG(D_TRACE, "acquiring trusted reference: %d %p %lu\n", - atomic_read(&lock->cll_ref), lock, RETIP); - if (atomic_inc_return(&lock->cll_ref) == 1) - CS_LOCK_INC(lock->cll_descr.cld_obj, busy); -} -EXPORT_SYMBOL(cl_lock_get_trust); - -/** - * Helper function destroying the lock that wasn't completely initialized. - * - * Other threads can acquire references to the top-lock through its - * sub-locks. Hence, it cannot be cl_lock_free()-ed immediately. - */ -static void cl_lock_finish(const struct lu_env *env, struct cl_lock *lock) -{ - cl_lock_mutex_get(env, lock); - cl_lock_cancel(env, lock); - cl_lock_delete(env, lock); - cl_lock_mutex_put(env, lock); - cl_lock_put(env, lock); -} - -static struct cl_lock *cl_lock_alloc(const struct lu_env *env, - struct cl_object *obj, - const struct cl_io *io, - const struct cl_lock_descr *descr) -{ - struct cl_lock *lock; - struct lu_object_header *head; - - lock = kmem_cache_zalloc(cl_lock_kmem, GFP_NOFS); - if (lock) { - atomic_set(&lock->cll_ref, 1); - lock->cll_descr = *descr; - lock->cll_state = CLS_NEW; - cl_object_get(obj); - lu_object_ref_add_at(&obj->co_lu, &lock->cll_obj_ref, "cl_lock", - lock); - INIT_LIST_HEAD(&lock->cll_layers); - INIT_LIST_HEAD(&lock->cll_linkage); - INIT_LIST_HEAD(&lock->cll_inclosure); - lu_ref_init(&lock->cll_reference); - lu_ref_init(&lock->cll_holders); - mutex_init(&lock->cll_guard); - lockdep_set_class(&lock->cll_guard, &cl_lock_guard_class); - init_waitqueue_head(&lock->cll_wq); - head = obj->co_lu.lo_header; - CS_LOCKSTATE_INC(obj, CLS_NEW); - CS_LOCK_INC(obj, total); - CS_LOCK_INC(obj, create); - cl_lock_lockdep_init(lock); - list_for_each_entry(obj, &head->loh_layers, co_lu.lo_linkage) { - int err; - - err = obj->co_ops->coo_lock_init(env, obj, lock, io); - if (err != 0) { - cl_lock_finish(env, lock); - lock = ERR_PTR(err); - break; - } - } - } else - lock = ERR_PTR(-ENOMEM); - return lock; -} - -/** - * Transfer the lock into INTRANSIT state and return the original state. - * - * \pre state: CLS_CACHED, CLS_HELD or CLS_ENQUEUED - * \post state: CLS_INTRANSIT - * \see CLS_INTRANSIT - */ -static enum cl_lock_state cl_lock_intransit(const struct lu_env *env, - struct cl_lock *lock) -{ - enum cl_lock_state state = lock->cll_state; - - LASSERT(cl_lock_is_mutexed(lock)); - LASSERT(state != CLS_INTRANSIT); - LASSERTF(state >= CLS_ENQUEUED && state <= CLS_CACHED, - "Malformed lock state %d.\n", state); - - cl_lock_state_set(env, lock, CLS_INTRANSIT); - lock->cll_intransit_owner = current; - cl_lock_hold_add(env, lock, "intransit", current); - return state; -} - -/** - * Exit the intransit state and restore the lock state to the original state - */ -static void cl_lock_extransit(const struct lu_env *env, struct cl_lock *lock, - enum cl_lock_state state) -{ - LASSERT(cl_lock_is_mutexed(lock)); - LASSERT(lock->cll_state == CLS_INTRANSIT); - LASSERT(state != CLS_INTRANSIT); - LASSERT(lock->cll_intransit_owner == current); - - lock->cll_intransit_owner = NULL; - cl_lock_state_set(env, lock, state); - cl_lock_unhold(env, lock, "intransit", current); -} - -/** - * Checking whether the lock is intransit state - */ -int cl_lock_is_intransit(struct cl_lock *lock) -{ - LASSERT(cl_lock_is_mutexed(lock)); - return lock->cll_state == CLS_INTRANSIT && - lock->cll_intransit_owner != current; -} -EXPORT_SYMBOL(cl_lock_is_intransit); -/** - * Returns true iff lock is "suitable" for given io. E.g., locks acquired by - * truncate and O_APPEND cannot be reused for read/non-append-write, as they - * cover multiple stripes and can trigger cascading timeouts. - */ -static int cl_lock_fits_into(const struct lu_env *env, - const struct cl_lock *lock, - const struct cl_lock_descr *need, - const struct cl_io *io) -{ - const struct cl_lock_slice *slice; - - LINVRNT(cl_lock_invariant_trusted(env, lock)); - list_for_each_entry(slice, &lock->cll_layers, cls_linkage) { - if (slice->cls_ops->clo_fits_into && - !slice->cls_ops->clo_fits_into(env, slice, need, io)) - return 0; - } - return 1; + POISON(lock, 0x5a, sizeof(*lock)); } +EXPORT_SYMBOL(cl_lock_fini); -static struct cl_lock *cl_lock_lookup(const struct lu_env *env, - struct cl_object *obj, - const struct cl_io *io, - const struct cl_lock_descr *need) +int cl_lock_init(const struct lu_env *env, struct cl_lock *lock, + const struct cl_io *io) { - struct cl_lock *lock; - struct cl_object_header *head; - - head = cl_object_header(obj); - assert_spin_locked(&head->coh_lock_guard); - CS_LOCK_INC(obj, lookup); - list_for_each_entry(lock, &head->coh_locks, cll_linkage) { - int matched; - - matched = cl_lock_ext_match(&lock->cll_descr, need) && - lock->cll_state < CLS_FREEING && - lock->cll_error == 0 && - !(lock->cll_flags & CLF_CANCELLED) && - cl_lock_fits_into(env, lock, need, io); - CDEBUG(D_DLMTRACE, "has: "DDESCR"(%d) need: "DDESCR": %d\n", - PDESCR(&lock->cll_descr), lock->cll_state, PDESCR(need), - matched); - if (matched) { - cl_lock_get_trust(lock); - CS_LOCK_INC(obj, hit); - return lock; - } - } - return NULL; -} - -/** - * Returns a lock matching description \a need. - * - * This is the main entry point into the cl_lock caching interface. First, a - * cache (implemented as a per-object linked list) is consulted. If lock is - * found there, it is returned immediately. Otherwise new lock is allocated - * and returned. In any case, additional reference to lock is acquired. - * - * \see cl_object_find(), cl_page_find() - */ -static struct cl_lock *cl_lock_find(const struct lu_env *env, - const struct cl_io *io, - const struct cl_lock_descr *need) -{ - struct cl_object_header *head; - struct cl_object *obj; - struct cl_lock *lock; - - obj = need->cld_obj; - head = cl_object_header(obj); - - spin_lock(&head->coh_lock_guard); - lock = cl_lock_lookup(env, obj, io, need); - spin_unlock(&head->coh_lock_guard); + struct cl_object *obj = lock->cll_descr.cld_obj; + struct cl_object *scan; + int result = 0; - if (!lock) { - lock = cl_lock_alloc(env, obj, io, need); - if (!IS_ERR(lock)) { - struct cl_lock *ghost; + /* Make sure cl_lock::cll_descr is initialized. */ + LASSERT(obj); - spin_lock(&head->coh_lock_guard); - ghost = cl_lock_lookup(env, obj, io, need); - if (!ghost) { - cl_lock_get_trust(lock); - list_add_tail(&lock->cll_linkage, - &head->coh_locks); - spin_unlock(&head->coh_lock_guard); - CS_LOCK_INC(obj, busy); - } else { - spin_unlock(&head->coh_lock_guard); - /* - * Other threads can acquire references to the - * top-lock through its sub-locks. Hence, it - * cannot be cl_lock_free()-ed immediately. - */ - cl_lock_finish(env, lock); - lock = ghost; - } + INIT_LIST_HEAD(&lock->cll_layers); + list_for_each_entry(scan, &obj->co_lu.lo_header->loh_layers, + co_lu.lo_linkage) { + result = scan->co_ops->coo_lock_init(env, scan, lock, io); + if (result != 0) { + cl_lock_fini(env, lock); + break; } } - return lock; -} -/** - * Returns existing lock matching given description. This is similar to - * cl_lock_find() except that no new lock is created, and returned lock is - * guaranteed to be in enum cl_lock_state::CLS_HELD state. - */ -struct cl_lock *cl_lock_peek(const struct lu_env *env, const struct cl_io *io, - const struct cl_lock_descr *need, - const char *scope, const void *source) -{ - struct cl_object_header *head; - struct cl_object *obj; - struct cl_lock *lock; - - obj = need->cld_obj; - head = cl_object_header(obj); - - do { - spin_lock(&head->coh_lock_guard); - lock = cl_lock_lookup(env, obj, io, need); - spin_unlock(&head->coh_lock_guard); - if (!lock) - return NULL; - - cl_lock_mutex_get(env, lock); - if (lock->cll_state == CLS_INTRANSIT) - /* Don't care return value. */ - cl_lock_state_wait(env, lock); - if (lock->cll_state == CLS_FREEING) { - cl_lock_mutex_put(env, lock); - cl_lock_put(env, lock); - lock = NULL; - } - } while (!lock); - - cl_lock_hold_add(env, lock, scope, source); - cl_lock_user_add(env, lock); - if (lock->cll_state == CLS_CACHED) - cl_use_try(env, lock, 1); - if (lock->cll_state == CLS_HELD) { - cl_lock_mutex_put(env, lock); - cl_lock_lockdep_acquire(env, lock, 0); - cl_lock_put(env, lock); - } else { - cl_unuse_try(env, lock); - cl_lock_unhold(env, lock, scope, source); - cl_lock_mutex_put(env, lock); - cl_lock_put(env, lock); - lock = NULL; - } - - return lock; + return result; } -EXPORT_SYMBOL(cl_lock_peek); +EXPORT_SYMBOL(cl_lock_init); /** - * Returns a slice within a lock, corresponding to the given layer in the + * Returns a slice with a lock, corresponding to the given layer in the * device stack. * * \see cl_page_at() @@ -616,8 +131,6 @@ const struct cl_lock_slice *cl_lock_at(const struct cl_lock *lock, { const struct cl_lock_slice *slice; - LINVRNT(cl_lock_invariant_trusted(NULL, lock)); - list_for_each_entry(slice, &lock->cll_layers, cls_linkage) { if (slice->cls_obj->co_lu.lo_dev->ld_type == dtype) return slice; @@ -626,1537 +139,96 @@ const struct cl_lock_slice *cl_lock_at(const struct cl_lock *lock, } EXPORT_SYMBOL(cl_lock_at); -static void cl_lock_mutex_tail(const struct lu_env *env, struct cl_lock *lock) -{ - struct cl_thread_counters *counters; - - counters = cl_lock_counters(env, lock); - lock->cll_depth++; - counters->ctc_nr_locks_locked++; - lu_ref_add(&counters->ctc_locks_locked, "cll_guard", lock); - cl_lock_trace(D_TRACE, env, "got mutex", lock); -} - -/** - * Locks cl_lock object. - * - * This is used to manipulate cl_lock fields, and to serialize state - * transitions in the lock state machine. - * - * \post cl_lock_is_mutexed(lock) - * - * \see cl_lock_mutex_put() - */ -void cl_lock_mutex_get(const struct lu_env *env, struct cl_lock *lock) -{ - LINVRNT(cl_lock_invariant(env, lock)); - - if (lock->cll_guarder == current) { - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(lock->cll_depth > 0); - } else { - struct cl_object_header *hdr; - struct cl_thread_info *info; - int i; - - LINVRNT(lock->cll_guarder != current); - hdr = cl_object_header(lock->cll_descr.cld_obj); - /* - * Check that mutices are taken in the bottom-to-top order. - */ - info = cl_env_info(env); - for (i = 0; i < hdr->coh_nesting; ++i) - LASSERT(info->clt_counters[i].ctc_nr_locks_locked == 0); - mutex_lock_nested(&lock->cll_guard, hdr->coh_nesting); - lock->cll_guarder = current; - LINVRNT(lock->cll_depth == 0); - } - cl_lock_mutex_tail(env, lock); -} -EXPORT_SYMBOL(cl_lock_mutex_get); - -/** - * Try-locks cl_lock object. - * - * \retval 0 \a lock was successfully locked - * - * \retval -EBUSY \a lock cannot be locked right now - * - * \post ergo(result == 0, cl_lock_is_mutexed(lock)) - * - * \see cl_lock_mutex_get() - */ -static int cl_lock_mutex_try(const struct lu_env *env, struct cl_lock *lock) -{ - int result; - - LINVRNT(cl_lock_invariant_trusted(env, lock)); - - result = 0; - if (lock->cll_guarder == current) { - LINVRNT(lock->cll_depth > 0); - cl_lock_mutex_tail(env, lock); - } else if (mutex_trylock(&lock->cll_guard)) { - LINVRNT(lock->cll_depth == 0); - lock->cll_guarder = current; - cl_lock_mutex_tail(env, lock); - } else - result = -EBUSY; - return result; -} - -/** - {* Unlocks cl_lock object. - * - * \pre cl_lock_is_mutexed(lock) - * - * \see cl_lock_mutex_get() - */ -void cl_lock_mutex_put(const struct lu_env *env, struct cl_lock *lock) -{ - struct cl_thread_counters *counters; - - LINVRNT(cl_lock_invariant(env, lock)); - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(lock->cll_guarder == current); - LINVRNT(lock->cll_depth > 0); - - counters = cl_lock_counters(env, lock); - LINVRNT(counters->ctc_nr_locks_locked > 0); - - cl_lock_trace(D_TRACE, env, "put mutex", lock); - lu_ref_del(&counters->ctc_locks_locked, "cll_guard", lock); - counters->ctc_nr_locks_locked--; - if (--lock->cll_depth == 0) { - lock->cll_guarder = NULL; - mutex_unlock(&lock->cll_guard); - } -} -EXPORT_SYMBOL(cl_lock_mutex_put); - -/** - * Returns true iff lock's mutex is owned by the current thread. - */ -int cl_lock_is_mutexed(struct cl_lock *lock) -{ - return lock->cll_guarder == current; -} -EXPORT_SYMBOL(cl_lock_is_mutexed); - -/** - * Returns number of cl_lock mutices held by the current thread (environment). - */ -int cl_lock_nr_mutexed(const struct lu_env *env) -{ - struct cl_thread_info *info; - int i; - int locked; - - /* - * NOTE: if summation across all nesting levels (currently 2) proves - * too expensive, a summary counter can be added to - * struct cl_thread_info. - */ - info = cl_env_info(env); - for (i = 0, locked = 0; i < ARRAY_SIZE(info->clt_counters); ++i) - locked += info->clt_counters[i].ctc_nr_locks_locked; - return locked; -} -EXPORT_SYMBOL(cl_lock_nr_mutexed); - -static void cl_lock_cancel0(const struct lu_env *env, struct cl_lock *lock) -{ - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - if (!(lock->cll_flags & CLF_CANCELLED)) { - const struct cl_lock_slice *slice; - - lock->cll_flags |= CLF_CANCELLED; - list_for_each_entry_reverse(slice, &lock->cll_layers, - cls_linkage) { - if (slice->cls_ops->clo_cancel) - slice->cls_ops->clo_cancel(env, slice); - } - } -} - -static void cl_lock_delete0(const struct lu_env *env, struct cl_lock *lock) -{ - struct cl_object_header *head; - const struct cl_lock_slice *slice; - - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - - if (lock->cll_state < CLS_FREEING) { - bool in_cache; - - LASSERT(lock->cll_state != CLS_INTRANSIT); - cl_lock_state_set(env, lock, CLS_FREEING); - - head = cl_object_header(lock->cll_descr.cld_obj); - - spin_lock(&head->coh_lock_guard); - in_cache = !list_empty(&lock->cll_linkage); - if (in_cache) - list_del_init(&lock->cll_linkage); - spin_unlock(&head->coh_lock_guard); - - if (in_cache) /* coh_locks cache holds a refcount. */ - cl_lock_put(env, lock); - - /* - * From now on, no new references to this lock can be acquired - * by cl_lock_lookup(). - */ - list_for_each_entry_reverse(slice, &lock->cll_layers, - cls_linkage) { - if (slice->cls_ops->clo_delete) - slice->cls_ops->clo_delete(env, slice); - } - /* - * From now on, no new references to this lock can be acquired - * by layer-specific means (like a pointer from struct - * ldlm_lock in osc, or a pointer from top-lock to sub-lock in - * lov). - * - * Lock will be finally freed in cl_lock_put() when last of - * existing references goes away. - */ - } -} - -/** - * Mod(ifie)s cl_lock::cll_holds counter for a given lock. Also, for a - * top-lock (nesting == 0) accounts for this modification in the per-thread - * debugging counters. Sub-lock holds can be released by a thread different - * from one that acquired it. - */ -static void cl_lock_hold_mod(const struct lu_env *env, struct cl_lock *lock, - int delta) -{ - struct cl_thread_counters *counters; - enum clt_nesting_level nesting; - - lock->cll_holds += delta; - nesting = cl_lock_nesting(lock); - if (nesting == CNL_TOP) { - counters = &cl_env_info(env)->clt_counters[CNL_TOP]; - counters->ctc_nr_held += delta; - LASSERT(counters->ctc_nr_held >= 0); - } -} - -/** - * Mod(ifie)s cl_lock::cll_users counter for a given lock. See - * cl_lock_hold_mod() for the explanation of the debugging code. - */ -static void cl_lock_used_mod(const struct lu_env *env, struct cl_lock *lock, - int delta) -{ - struct cl_thread_counters *counters; - enum clt_nesting_level nesting; - - lock->cll_users += delta; - nesting = cl_lock_nesting(lock); - if (nesting == CNL_TOP) { - counters = &cl_env_info(env)->clt_counters[CNL_TOP]; - counters->ctc_nr_used += delta; - LASSERT(counters->ctc_nr_used >= 0); - } -} - -void cl_lock_hold_release(const struct lu_env *env, struct cl_lock *lock, - const char *scope, const void *source) -{ - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - LASSERT(lock->cll_holds > 0); - - cl_lock_trace(D_DLMTRACE, env, "hold release lock", lock); - lu_ref_del(&lock->cll_holders, scope, source); - cl_lock_hold_mod(env, lock, -1); - if (lock->cll_holds == 0) { - CL_LOCK_ASSERT(lock->cll_state != CLS_HELD, env, lock); - if (lock->cll_descr.cld_mode == CLM_PHANTOM || - lock->cll_descr.cld_mode == CLM_GROUP || - lock->cll_state != CLS_CACHED) - /* - * If lock is still phantom or grouplock when user is - * done with it---destroy the lock. - */ - lock->cll_flags |= CLF_CANCELPEND|CLF_DOOMED; - if (lock->cll_flags & CLF_CANCELPEND) { - lock->cll_flags &= ~CLF_CANCELPEND; - cl_lock_cancel0(env, lock); - } - if (lock->cll_flags & CLF_DOOMED) { - /* no longer doomed: it's dead... Jim. */ - lock->cll_flags &= ~CLF_DOOMED; - cl_lock_delete0(env, lock); - } - } -} -EXPORT_SYMBOL(cl_lock_hold_release); - -/** - * Waits until lock state is changed. - * - * This function is called with cl_lock mutex locked, atomically releases - * mutex and goes to sleep, waiting for a lock state change (signaled by - * cl_lock_signal()), and re-acquires the mutex before return. - * - * This function is used to wait until lock state machine makes some progress - * and to emulate synchronous operations on top of asynchronous lock - * interface. - * - * \retval -EINTR wait was interrupted - * - * \retval 0 wait wasn't interrupted - * - * \pre cl_lock_is_mutexed(lock) - * - * \see cl_lock_signal() - */ -int cl_lock_state_wait(const struct lu_env *env, struct cl_lock *lock) -{ - wait_queue_t waiter; - sigset_t blocked; - int result; - - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - LASSERT(lock->cll_depth == 1); - LASSERT(lock->cll_state != CLS_FREEING); /* too late to wait */ - - cl_lock_trace(D_DLMTRACE, env, "state wait lock", lock); - result = lock->cll_error; - if (result == 0) { - /* To avoid being interrupted by the 'non-fatal' signals - * (SIGCHLD, for instance), we'd block them temporarily. - * LU-305 - */ - blocked = cfs_block_sigsinv(LUSTRE_FATAL_SIGS); - - init_waitqueue_entry(&waiter, current); - add_wait_queue(&lock->cll_wq, &waiter); - set_current_state(TASK_INTERRUPTIBLE); - cl_lock_mutex_put(env, lock); - - LASSERT(cl_lock_nr_mutexed(env) == 0); - - /* Returning ERESTARTSYS instead of EINTR so syscalls - * can be restarted if signals are pending here - */ - result = -ERESTARTSYS; - if (likely(!OBD_FAIL_CHECK(OBD_FAIL_LOCK_STATE_WAIT_INTR))) { - schedule(); - if (!cfs_signal_pending()) - result = 0; - } - - cl_lock_mutex_get(env, lock); - set_current_state(TASK_RUNNING); - remove_wait_queue(&lock->cll_wq, &waiter); - - /* Restore old blocked signals */ - cfs_restore_sigs(blocked); - } - return result; -} -EXPORT_SYMBOL(cl_lock_state_wait); - -static void cl_lock_state_signal(const struct lu_env *env, struct cl_lock *lock, - enum cl_lock_state state) -{ - const struct cl_lock_slice *slice; - - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - - list_for_each_entry(slice, &lock->cll_layers, cls_linkage) - if (slice->cls_ops->clo_state) - slice->cls_ops->clo_state(env, slice, state); - wake_up_all(&lock->cll_wq); -} - -/** - * Notifies waiters that lock state changed. - * - * Wakes up all waiters sleeping in cl_lock_state_wait(), also notifies all - * layers about state change by calling cl_lock_operations::clo_state() - * top-to-bottom. - */ -void cl_lock_signal(const struct lu_env *env, struct cl_lock *lock) -{ - cl_lock_trace(D_DLMTRACE, env, "state signal lock", lock); - cl_lock_state_signal(env, lock, lock->cll_state); -} -EXPORT_SYMBOL(cl_lock_signal); - -/** - * Changes lock state. - * - * This function is invoked to notify layers that lock state changed, possible - * as a result of an asynchronous event such as call-back reception. - * - * \post lock->cll_state == state - * - * \see cl_lock_operations::clo_state() - */ -void cl_lock_state_set(const struct lu_env *env, struct cl_lock *lock, - enum cl_lock_state state) -{ - LASSERT(lock->cll_state <= state || - (lock->cll_state == CLS_CACHED && - (state == CLS_HELD || /* lock found in cache */ - state == CLS_NEW || /* sub-lock canceled */ - state == CLS_INTRANSIT)) || - /* lock is in transit state */ - lock->cll_state == CLS_INTRANSIT); - - if (lock->cll_state != state) { - CS_LOCKSTATE_DEC(lock->cll_descr.cld_obj, lock->cll_state); - CS_LOCKSTATE_INC(lock->cll_descr.cld_obj, state); - - cl_lock_state_signal(env, lock, state); - lock->cll_state = state; - } -} -EXPORT_SYMBOL(cl_lock_state_set); - -static int cl_unuse_try_internal(const struct lu_env *env, struct cl_lock *lock) -{ - const struct cl_lock_slice *slice; - int result; - - do { - result = 0; - - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - LASSERT(lock->cll_state == CLS_INTRANSIT); - - result = -ENOSYS; - list_for_each_entry_reverse(slice, &lock->cll_layers, - cls_linkage) { - if (slice->cls_ops->clo_unuse) { - result = slice->cls_ops->clo_unuse(env, slice); - if (result != 0) - break; - } - } - LASSERT(result != -ENOSYS); - } while (result == CLO_REPEAT); - - return result; -} - -/** - * Yanks lock from the cache (cl_lock_state::CLS_CACHED state) by calling - * cl_lock_operations::clo_use() top-to-bottom to notify layers. - * @atomic = 1, it must unuse the lock to recovery the lock to keep the - * use process atomic - */ -int cl_use_try(const struct lu_env *env, struct cl_lock *lock, int atomic) +void cl_lock_cancel(const struct lu_env *env, struct cl_lock *lock) { const struct cl_lock_slice *slice; - int result; - enum cl_lock_state state; - - cl_lock_trace(D_DLMTRACE, env, "use lock", lock); - - LASSERT(lock->cll_state == CLS_CACHED); - if (lock->cll_error) - return lock->cll_error; - - result = -ENOSYS; - state = cl_lock_intransit(env, lock); - list_for_each_entry(slice, &lock->cll_layers, cls_linkage) { - if (slice->cls_ops->clo_use) { - result = slice->cls_ops->clo_use(env, slice); - if (result != 0) - break; - } - } - LASSERT(result != -ENOSYS); - - LASSERTF(lock->cll_state == CLS_INTRANSIT, "Wrong state %d.\n", - lock->cll_state); - - if (result == 0) { - state = CLS_HELD; - } else { - if (result == -ESTALE) { - /* - * ESTALE means sublock being cancelled - * at this time, and set lock state to - * be NEW here and ask the caller to repeat. - */ - state = CLS_NEW; - result = CLO_REPEAT; - } - - /* @atomic means back-off-on-failure. */ - if (atomic) { - int rc; - - rc = cl_unuse_try_internal(env, lock); - /* Vet the results. */ - if (rc < 0 && result > 0) - result = rc; - } + cl_lock_trace(D_DLMTRACE, env, "cancel lock", lock); + list_for_each_entry_reverse(slice, &lock->cll_layers, cls_linkage) { + if (slice->cls_ops->clo_cancel) + slice->cls_ops->clo_cancel(env, slice); } - cl_lock_extransit(env, lock, state); - return result; } -EXPORT_SYMBOL(cl_use_try); +EXPORT_SYMBOL(cl_lock_cancel); /** - * Helper for cl_enqueue_try() that calls ->clo_enqueue() across all layers - * top-to-bottom. + * Enqueue a lock. + * \param anchor: if we need to wait for resources before getting the lock, + * use @anchor for the purpose. + * \retval 0 enqueue successfully + * \retval <0 error code */ -static int cl_enqueue_kick(const struct lu_env *env, - struct cl_lock *lock, - struct cl_io *io, __u32 flags) +int cl_lock_enqueue(const struct lu_env *env, struct cl_io *io, + struct cl_lock *lock, struct cl_sync_io *anchor) { - int result; const struct cl_lock_slice *slice; + int rc = -ENOSYS; - result = -ENOSYS; list_for_each_entry(slice, &lock->cll_layers, cls_linkage) { - if (slice->cls_ops->clo_enqueue) { - result = slice->cls_ops->clo_enqueue(env, - slice, io, flags); - if (result != 0) - break; - } - } - LASSERT(result != -ENOSYS); - return result; -} - -/** - * Tries to enqueue a lock. - * - * This function is called repeatedly by cl_enqueue() until either lock is - * enqueued, or error occurs. This function does not block waiting for - * networking communication to complete. - * - * \post ergo(result == 0, lock->cll_state == CLS_ENQUEUED || - * lock->cll_state == CLS_HELD) - * - * \see cl_enqueue() cl_lock_operations::clo_enqueue() - * \see cl_lock_state::CLS_ENQUEUED - */ -int cl_enqueue_try(const struct lu_env *env, struct cl_lock *lock, - struct cl_io *io, __u32 flags) -{ - int result; - - cl_lock_trace(D_DLMTRACE, env, "enqueue lock", lock); - do { - LINVRNT(cl_lock_is_mutexed(lock)); - - result = lock->cll_error; - if (result != 0) - break; - - switch (lock->cll_state) { - case CLS_NEW: - cl_lock_state_set(env, lock, CLS_QUEUING); - /* fall-through */ - case CLS_QUEUING: - /* kick layers. */ - result = cl_enqueue_kick(env, lock, io, flags); - /* For AGL case, the cl_lock::cll_state may - * become CLS_HELD already. - */ - if (result == 0 && lock->cll_state == CLS_QUEUING) - cl_lock_state_set(env, lock, CLS_ENQUEUED); - break; - case CLS_INTRANSIT: - LASSERT(cl_lock_is_intransit(lock)); - result = CLO_WAIT; - break; - case CLS_CACHED: - /* yank lock from the cache. */ - result = cl_use_try(env, lock, 0); - break; - case CLS_ENQUEUED: - case CLS_HELD: - result = 0; - break; - default: - case CLS_FREEING: - /* - * impossible, only held locks with increased - * ->cll_holds can be enqueued, and they cannot be - * freed. - */ - LBUG(); - } - } while (result == CLO_REPEAT); - return result; -} -EXPORT_SYMBOL(cl_enqueue_try); - -/** - * Cancel the conflicting lock found during previous enqueue. - * - * \retval 0 conflicting lock has been canceled. - * \retval -ve error code. - */ -int cl_lock_enqueue_wait(const struct lu_env *env, - struct cl_lock *lock, - int keep_mutex) -{ - struct cl_lock *conflict; - int rc = 0; - - LASSERT(cl_lock_is_mutexed(lock)); - LASSERT(lock->cll_state == CLS_QUEUING); - LASSERT(lock->cll_conflict); - - conflict = lock->cll_conflict; - lock->cll_conflict = NULL; + if (!slice->cls_ops->clo_enqueue) + continue; - cl_lock_mutex_put(env, lock); - LASSERT(cl_lock_nr_mutexed(env) == 0); - - cl_lock_mutex_get(env, conflict); - cl_lock_trace(D_DLMTRACE, env, "enqueue wait", conflict); - cl_lock_cancel(env, conflict); - cl_lock_delete(env, conflict); - - while (conflict->cll_state != CLS_FREEING) { - rc = cl_lock_state_wait(env, conflict); + rc = slice->cls_ops->clo_enqueue(env, slice, io, anchor); if (rc != 0) break; - } - cl_lock_mutex_put(env, conflict); - lu_ref_del(&conflict->cll_reference, "cancel-wait", lock); - cl_lock_put(env, conflict); - - if (keep_mutex) - cl_lock_mutex_get(env, lock); - - LASSERT(rc <= 0); - return rc; -} -EXPORT_SYMBOL(cl_lock_enqueue_wait); - -static int cl_enqueue_locked(const struct lu_env *env, struct cl_lock *lock, - struct cl_io *io, __u32 enqflags) -{ - int result; - - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - LASSERT(lock->cll_holds > 0); - - cl_lock_user_add(env, lock); - do { - result = cl_enqueue_try(env, lock, io, enqflags); - if (result == CLO_WAIT) { - if (lock->cll_conflict) - result = cl_lock_enqueue_wait(env, lock, 1); - else - result = cl_lock_state_wait(env, lock); - if (result == 0) - continue; - } - break; - } while (1); - if (result != 0) - cl_unuse_try(env, lock); - LASSERT(ergo(result == 0 && !(enqflags & CEF_AGL), - lock->cll_state == CLS_ENQUEUED || - lock->cll_state == CLS_HELD)); - return result; -} - -/** - * Tries to unlock a lock. - * - * This function is called to release underlying resource: - * 1. for top lock, the resource is sublocks it held; - * 2. for sublock, the resource is the reference to dlmlock. - * - * cl_unuse_try is a one-shot operation, so it must NOT return CLO_WAIT. - * - * \see cl_unuse() cl_lock_operations::clo_unuse() - * \see cl_lock_state::CLS_CACHED - */ -int cl_unuse_try(const struct lu_env *env, struct cl_lock *lock) -{ - int result; - enum cl_lock_state state = CLS_NEW; - - cl_lock_trace(D_DLMTRACE, env, "unuse lock", lock); - - if (lock->cll_users > 1) { - cl_lock_user_del(env, lock); - return 0; - } - - /* Only if the lock is in CLS_HELD or CLS_ENQUEUED state, it can hold - * underlying resources. - */ - if (!(lock->cll_state == CLS_HELD || lock->cll_state == CLS_ENQUEUED)) { - cl_lock_user_del(env, lock); - return 0; - } - - /* - * New lock users (->cll_users) are not protecting unlocking - * from proceeding. From this point, lock eventually reaches - * CLS_CACHED, is reinitialized to CLS_NEW or fails into - * CLS_FREEING. - */ - state = cl_lock_intransit(env, lock); - - result = cl_unuse_try_internal(env, lock); - LASSERT(lock->cll_state == CLS_INTRANSIT); - LASSERT(result != CLO_WAIT); - cl_lock_user_del(env, lock); - if (result == 0 || result == -ESTALE) { - /* - * Return lock back to the cache. This is the only - * place where lock is moved into CLS_CACHED state. - * - * If one of ->clo_unuse() methods returned -ESTALE, lock - * cannot be placed into cache and has to be - * re-initialized. This happens e.g., when a sub-lock was - * canceled while unlocking was in progress. - */ - if (state == CLS_HELD && result == 0) - state = CLS_CACHED; - else - state = CLS_NEW; - cl_lock_extransit(env, lock, state); - - /* - * Hide -ESTALE error. - * If the lock is a glimpse lock, and it has multiple - * stripes. Assuming that one of its sublock returned -ENAVAIL, - * and other sublocks are matched write locks. In this case, - * we can't set this lock to error because otherwise some of - * its sublocks may not be canceled. This causes some dirty - * pages won't be written to OSTs. -jay - */ - result = 0; - } else { - CERROR("result = %d, this is unlikely!\n", result); - state = CLS_NEW; - cl_lock_extransit(env, lock, state); - } - return result ?: lock->cll_error; -} -EXPORT_SYMBOL(cl_unuse_try); - -static void cl_unuse_locked(const struct lu_env *env, struct cl_lock *lock) -{ - int result; - - result = cl_unuse_try(env, lock); - if (result) - CL_LOCK_DEBUG(D_ERROR, env, lock, "unuse return %d\n", result); -} - -/** - * Unlocks a lock. - */ -void cl_unuse(const struct lu_env *env, struct cl_lock *lock) -{ - cl_lock_mutex_get(env, lock); - cl_unuse_locked(env, lock); - cl_lock_mutex_put(env, lock); - cl_lock_lockdep_release(env, lock); -} -EXPORT_SYMBOL(cl_unuse); - -/** - * Tries to wait for a lock. - * - * This function is called repeatedly by cl_wait() until either lock is - * granted, or error occurs. This function does not block waiting for network - * communication to complete. - * - * \see cl_wait() cl_lock_operations::clo_wait() - * \see cl_lock_state::CLS_HELD - */ -int cl_wait_try(const struct lu_env *env, struct cl_lock *lock) -{ - const struct cl_lock_slice *slice; - int result; - - cl_lock_trace(D_DLMTRACE, env, "wait lock try", lock); - do { - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - LASSERTF(lock->cll_state == CLS_QUEUING || - lock->cll_state == CLS_ENQUEUED || - lock->cll_state == CLS_HELD || - lock->cll_state == CLS_INTRANSIT, - "lock state: %d\n", lock->cll_state); - LASSERT(lock->cll_users > 0); - LASSERT(lock->cll_holds > 0); - - result = lock->cll_error; - if (result != 0) - break; - - if (cl_lock_is_intransit(lock)) { - result = CLO_WAIT; - break; - } - - if (lock->cll_state == CLS_HELD) - /* nothing to do */ - break; - - result = -ENOSYS; - list_for_each_entry(slice, &lock->cll_layers, cls_linkage) { - if (slice->cls_ops->clo_wait) { - result = slice->cls_ops->clo_wait(env, slice); - if (result != 0) - break; - } - } - LASSERT(result != -ENOSYS); - if (result == 0) { - LASSERT(lock->cll_state != CLS_INTRANSIT); - cl_lock_state_set(env, lock, CLS_HELD); - } - } while (result == CLO_REPEAT); - return result; -} -EXPORT_SYMBOL(cl_wait_try); - -/** - * Waits until enqueued lock is granted. - * - * \pre current thread or io owns a hold on the lock - * \pre ergo(result == 0, lock->cll_state == CLS_ENQUEUED || - * lock->cll_state == CLS_HELD) - * - * \post ergo(result == 0, lock->cll_state == CLS_HELD) - */ -int cl_wait(const struct lu_env *env, struct cl_lock *lock) -{ - int result; - - cl_lock_mutex_get(env, lock); - - LINVRNT(cl_lock_invariant(env, lock)); - LASSERTF(lock->cll_state == CLS_ENQUEUED || lock->cll_state == CLS_HELD, - "Wrong state %d\n", lock->cll_state); - LASSERT(lock->cll_holds > 0); - - do { - result = cl_wait_try(env, lock); - if (result == CLO_WAIT) { - result = cl_lock_state_wait(env, lock); - if (result == 0) - continue; - } - break; - } while (1); - if (result < 0) { - cl_unuse_try(env, lock); - cl_lock_lockdep_release(env, lock); - } - cl_lock_trace(D_DLMTRACE, env, "wait lock", lock); - cl_lock_mutex_put(env, lock); - LASSERT(ergo(result == 0, lock->cll_state == CLS_HELD)); - return result; -} -EXPORT_SYMBOL(cl_wait); - -/** - * Executes cl_lock_operations::clo_weigh(), and sums results to estimate lock - * value. - */ -unsigned long cl_lock_weigh(const struct lu_env *env, struct cl_lock *lock) -{ - const struct cl_lock_slice *slice; - unsigned long pound; - unsigned long ounce; - - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - - pound = 0; - list_for_each_entry_reverse(slice, &lock->cll_layers, cls_linkage) { - if (slice->cls_ops->clo_weigh) { - ounce = slice->cls_ops->clo_weigh(env, slice); - pound += ounce; - if (pound < ounce) /* over-weight^Wflow */ - pound = ~0UL; - } - } - return pound; -} -EXPORT_SYMBOL(cl_lock_weigh); - -/** - * Notifies layers that lock description changed. - * - * The server can grant client a lock different from one that was requested - * (e.g., larger in extent). This method is called when actually granted lock - * description becomes known to let layers to accommodate for changed lock - * description. - * - * \see cl_lock_operations::clo_modify() - */ -int cl_lock_modify(const struct lu_env *env, struct cl_lock *lock, - const struct cl_lock_descr *desc) -{ - const struct cl_lock_slice *slice; - struct cl_object *obj = lock->cll_descr.cld_obj; - struct cl_object_header *hdr = cl_object_header(obj); - int result; - - cl_lock_trace(D_DLMTRACE, env, "modify lock", lock); - /* don't allow object to change */ - LASSERT(obj == desc->cld_obj); - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - - list_for_each_entry_reverse(slice, &lock->cll_layers, cls_linkage) { - if (slice->cls_ops->clo_modify) { - result = slice->cls_ops->clo_modify(env, slice, desc); - if (result != 0) - return result; - } - } - CL_LOCK_DEBUG(D_DLMTRACE, env, lock, " -> "DDESCR"@"DFID"\n", - PDESCR(desc), PFID(lu_object_fid(&desc->cld_obj->co_lu))); - /* - * Just replace description in place. Nothing more is needed for - * now. If locks were indexed according to their extent and/or mode, - * that index would have to be updated here. - */ - spin_lock(&hdr->coh_lock_guard); - lock->cll_descr = *desc; - spin_unlock(&hdr->coh_lock_guard); - return 0; -} -EXPORT_SYMBOL(cl_lock_modify); - -/** - * Initializes lock closure with a given origin. - * - * \see cl_lock_closure - */ -void cl_lock_closure_init(const struct lu_env *env, - struct cl_lock_closure *closure, - struct cl_lock *origin, int wait) -{ - LINVRNT(cl_lock_is_mutexed(origin)); - LINVRNT(cl_lock_invariant(env, origin)); - - INIT_LIST_HEAD(&closure->clc_list); - closure->clc_origin = origin; - closure->clc_wait = wait; - closure->clc_nr = 0; -} -EXPORT_SYMBOL(cl_lock_closure_init); - -/** - * Builds a closure of \a lock. - * - * Building of a closure consists of adding initial lock (\a lock) into it, - * and calling cl_lock_operations::clo_closure() methods of \a lock. These - * methods might call cl_lock_closure_build() recursively again, adding more - * locks to the closure, etc. - * - * \see cl_lock_closure - */ -int cl_lock_closure_build(const struct lu_env *env, struct cl_lock *lock, - struct cl_lock_closure *closure) -{ - const struct cl_lock_slice *slice; - int result; - - LINVRNT(cl_lock_is_mutexed(closure->clc_origin)); - LINVRNT(cl_lock_invariant(env, closure->clc_origin)); - - result = cl_lock_enclosure(env, lock, closure); - if (result == 0) { - list_for_each_entry(slice, &lock->cll_layers, cls_linkage) { - if (slice->cls_ops->clo_closure) { - result = slice->cls_ops->clo_closure(env, slice, - closure); - if (result != 0) - break; - } - } - } - if (result != 0) - cl_lock_disclosure(env, closure); - return result; -} -EXPORT_SYMBOL(cl_lock_closure_build); - -/** - * Adds new lock to a closure. - * - * Try-locks \a lock and if succeeded, adds it to the closure (never more than - * once). If try-lock failed, returns CLO_REPEAT, after optionally waiting - * until next try-lock is likely to succeed. - */ -int cl_lock_enclosure(const struct lu_env *env, struct cl_lock *lock, - struct cl_lock_closure *closure) -{ - int result = 0; - - cl_lock_trace(D_DLMTRACE, env, "enclosure lock", lock); - if (!cl_lock_mutex_try(env, lock)) { - /* - * If lock->cll_inclosure is not empty, lock is already in - * this closure. - */ - if (list_empty(&lock->cll_inclosure)) { - cl_lock_get_trust(lock); - lu_ref_add(&lock->cll_reference, "closure", closure); - list_add(&lock->cll_inclosure, &closure->clc_list); - closure->clc_nr++; - } else - cl_lock_mutex_put(env, lock); - result = 0; - } else { - cl_lock_disclosure(env, closure); - if (closure->clc_wait) { - cl_lock_get_trust(lock); - lu_ref_add(&lock->cll_reference, "closure-w", closure); - cl_lock_mutex_put(env, closure->clc_origin); - - LASSERT(cl_lock_nr_mutexed(env) == 0); - cl_lock_mutex_get(env, lock); - cl_lock_mutex_put(env, lock); - - cl_lock_mutex_get(env, closure->clc_origin); - lu_ref_del(&lock->cll_reference, "closure-w", closure); - cl_lock_put(env, lock); - } - result = CLO_REPEAT; - } - return result; -} -EXPORT_SYMBOL(cl_lock_enclosure); - -/** Releases mutices of enclosed locks. */ -void cl_lock_disclosure(const struct lu_env *env, - struct cl_lock_closure *closure) -{ - struct cl_lock *scan; - struct cl_lock *temp; - - cl_lock_trace(D_DLMTRACE, env, "disclosure lock", closure->clc_origin); - list_for_each_entry_safe(scan, temp, &closure->clc_list, - cll_inclosure) { - list_del_init(&scan->cll_inclosure); - cl_lock_mutex_put(env, scan); - lu_ref_del(&scan->cll_reference, "closure", closure); - cl_lock_put(env, scan); - closure->clc_nr--; - } - LASSERT(closure->clc_nr == 0); -} -EXPORT_SYMBOL(cl_lock_disclosure); - -/** Finalizes a closure. */ -void cl_lock_closure_fini(struct cl_lock_closure *closure) -{ - LASSERT(closure->clc_nr == 0); - LASSERT(list_empty(&closure->clc_list)); -} -EXPORT_SYMBOL(cl_lock_closure_fini); - -/** - * Destroys this lock. Notifies layers (bottom-to-top) that lock is being - * destroyed, then destroy the lock. If there are holds on the lock, postpone - * destruction until all holds are released. This is called when a decision is - * made to destroy the lock in the future. E.g., when a blocking AST is - * received on it, or fatal communication error happens. - * - * Caller must have a reference on this lock to prevent a situation, when - * deleted lock lingers in memory for indefinite time, because nobody calls - * cl_lock_put() to finish it. - * - * \pre atomic_read(&lock->cll_ref) > 0 - * \pre ergo(cl_lock_nesting(lock) == CNL_TOP, - * cl_lock_nr_mutexed(env) == 1) - * [i.e., if a top-lock is deleted, mutices of no other locks can be - * held, as deletion of sub-locks might require releasing a top-lock - * mutex] - * - * \see cl_lock_operations::clo_delete() - * \see cl_lock::cll_holds - */ -void cl_lock_delete(const struct lu_env *env, struct cl_lock *lock) -{ - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - LASSERT(ergo(cl_lock_nesting(lock) == CNL_TOP, - cl_lock_nr_mutexed(env) == 1)); - - cl_lock_trace(D_DLMTRACE, env, "delete lock", lock); - if (lock->cll_holds == 0) - cl_lock_delete0(env, lock); - else - lock->cll_flags |= CLF_DOOMED; -} -EXPORT_SYMBOL(cl_lock_delete); - -/** - * Mark lock as irrecoverably failed, and mark it for destruction. This - * happens when, e.g., server fails to grant a lock to us, or networking - * time-out happens. - * - * \pre atomic_read(&lock->cll_ref) > 0 - * - * \see clo_lock_delete() - * \see cl_lock::cll_holds - */ -void cl_lock_error(const struct lu_env *env, struct cl_lock *lock, int error) -{ - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - - if (lock->cll_error == 0 && error != 0) { - cl_lock_trace(D_DLMTRACE, env, "set lock error", lock); - lock->cll_error = error; - cl_lock_signal(env, lock); - cl_lock_cancel(env, lock); - cl_lock_delete(env, lock); - } -} -EXPORT_SYMBOL(cl_lock_error); - -/** - * Cancels this lock. Notifies layers - * (bottom-to-top) that lock is being cancelled, then destroy the lock. If - * there are holds on the lock, postpone cancellation until - * all holds are released. - * - * Cancellation notification is delivered to layers at most once. - * - * \see cl_lock_operations::clo_cancel() - * \see cl_lock::cll_holds - */ -void cl_lock_cancel(const struct lu_env *env, struct cl_lock *lock) -{ - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - - cl_lock_trace(D_DLMTRACE, env, "cancel lock", lock); - if (lock->cll_holds == 0) - cl_lock_cancel0(env, lock); - else - lock->cll_flags |= CLF_CANCELPEND; -} -EXPORT_SYMBOL(cl_lock_cancel); - -/** - * Finds an existing lock covering given index and optionally different from a - * given \a except lock. - */ -struct cl_lock *cl_lock_at_pgoff(const struct lu_env *env, - struct cl_object *obj, pgoff_t index, - struct cl_lock *except, - int pending, int canceld) -{ - struct cl_object_header *head; - struct cl_lock *scan; - struct cl_lock *lock; - struct cl_lock_descr *need; - - head = cl_object_header(obj); - need = &cl_env_info(env)->clt_descr; - lock = NULL; - - need->cld_mode = CLM_READ; /* CLM_READ matches both READ & WRITE, but - * not PHANTOM - */ - need->cld_start = need->cld_end = index; - need->cld_enq_flags = 0; - - spin_lock(&head->coh_lock_guard); - /* It is fine to match any group lock since there could be only one - * with a uniq gid and it conflicts with all other lock modes too - */ - list_for_each_entry(scan, &head->coh_locks, cll_linkage) { - if (scan != except && - (scan->cll_descr.cld_mode == CLM_GROUP || - cl_lock_ext_match(&scan->cll_descr, need)) && - scan->cll_state >= CLS_HELD && - scan->cll_state < CLS_FREEING && - /* - * This check is racy as the lock can be canceled right - * after it is done, but this is fine, because page exists - * already. - */ - (canceld || !(scan->cll_flags & CLF_CANCELLED)) && - (pending || !(scan->cll_flags & CLF_CANCELPEND))) { - /* Don't increase cs_hit here since this - * is just a helper function. - */ - cl_lock_get_trust(scan); - lock = scan; - break; } - } - spin_unlock(&head->coh_lock_guard); - return lock; -} -EXPORT_SYMBOL(cl_lock_at_pgoff); - -/** - * Calculate the page offset at the layer of @lock. - * At the time of this writing, @page is top page and @lock is sub lock. - */ -static pgoff_t pgoff_at_lock(struct cl_page *page, struct cl_lock *lock) -{ - struct lu_device_type *dtype; - const struct cl_page_slice *slice; - - dtype = lock->cll_descr.cld_obj->co_lu.lo_dev->ld_type; - slice = cl_page_at(page, dtype); - return slice->cpl_page->cp_index; + return rc; } +EXPORT_SYMBOL(cl_lock_enqueue); /** - * Check if page @page is covered by an extra lock or discard it. + * Main high-level entry point of cl_lock interface that finds existing or + * enqueues new lock matching given description. */ -static int check_and_discard_cb(const struct lu_env *env, struct cl_io *io, - struct cl_page *page, void *cbdata) +int cl_lock_request(const struct lu_env *env, struct cl_io *io, + struct cl_lock *lock) { - struct cl_thread_info *info = cl_env_info(env); - struct cl_lock *lock = cbdata; - pgoff_t index = pgoff_at_lock(page, lock); + struct cl_sync_io *anchor = NULL; + __u32 enq_flags = lock->cll_descr.cld_enq_flags; + int rc; - if (index >= info->clt_fn_index) { - struct cl_lock *tmp; + rc = cl_lock_init(env, lock, io); + if (rc < 0) + return rc; - /* refresh non-overlapped index */ - tmp = cl_lock_at_pgoff(env, lock->cll_descr.cld_obj, index, - lock, 1, 0); - if (tmp) { - /* Cache the first-non-overlapped index so as to skip - * all pages within [index, clt_fn_index). This - * is safe because if tmp lock is canceled, it will - * discard these pages. - */ - info->clt_fn_index = tmp->cll_descr.cld_end + 1; - if (tmp->cll_descr.cld_end == CL_PAGE_EOF) - info->clt_fn_index = CL_PAGE_EOF; - cl_lock_put(env, tmp); - } else if (cl_page_own(env, io, page) == 0) { - /* discard the page */ - cl_page_unmap(env, io, page); - cl_page_discard(env, io, page); - cl_page_disown(env, io, page); - } else { - LASSERT(page->cp_state == CPS_FREEING); - } + if ((enq_flags & CEF_ASYNC) && !(enq_flags & CEF_AGL)) { + anchor = &cl_env_info(env)->clt_anchor; + cl_sync_io_init(anchor, 1, cl_sync_io_end); } - info->clt_next_index = index + 1; - return CLP_GANG_OKAY; -} + rc = cl_lock_enqueue(env, io, lock, anchor); -static int discard_cb(const struct lu_env *env, struct cl_io *io, - struct cl_page *page, void *cbdata) -{ - struct cl_thread_info *info = cl_env_info(env); - struct cl_lock *lock = cbdata; + if (anchor) { + int rc2; - LASSERT(lock->cll_descr.cld_mode >= CLM_WRITE); - KLASSERT(ergo(page->cp_type == CPT_CACHEABLE, - !PageWriteback(cl_page_vmpage(env, page)))); - KLASSERT(ergo(page->cp_type == CPT_CACHEABLE, - !PageDirty(cl_page_vmpage(env, page)))); - - info->clt_next_index = pgoff_at_lock(page, lock) + 1; - if (cl_page_own(env, io, page) == 0) { - /* discard the page */ - cl_page_unmap(env, io, page); - cl_page_discard(env, io, page); - cl_page_disown(env, io, page); - } else { - LASSERT(page->cp_state == CPS_FREEING); + /* drop the reference count held at initialization time */ + cl_sync_io_note(env, anchor, 0); + rc2 = cl_sync_io_wait(env, anchor, 0); + if (rc2 < 0 && rc == 0) + rc = rc2; } - return CLP_GANG_OKAY; -} + if (rc < 0) + cl_lock_release(env, lock); -/** - * Discard pages protected by the given lock. This function traverses radix - * tree to find all covering pages and discard them. If a page is being covered - * by other locks, it should remain in cache. - * - * If error happens on any step, the process continues anyway (the reasoning - * behind this being that lock cancellation cannot be delayed indefinitely). - */ -int cl_lock_discard_pages(const struct lu_env *env, struct cl_lock *lock) -{ - struct cl_thread_info *info = cl_env_info(env); - struct cl_io *io = &info->clt_io; - struct cl_lock_descr *descr = &lock->cll_descr; - cl_page_gang_cb_t cb; - int res; - int result; - - LINVRNT(cl_lock_invariant(env, lock)); - - io->ci_obj = cl_object_top(descr->cld_obj); - io->ci_ignore_layout = 1; - result = cl_io_init(env, io, CIT_MISC, io->ci_obj); - if (result != 0) - goto out; - - cb = descr->cld_mode == CLM_READ ? check_and_discard_cb : discard_cb; - info->clt_fn_index = info->clt_next_index = descr->cld_start; - do { - res = cl_page_gang_lookup(env, descr->cld_obj, io, - info->clt_next_index, descr->cld_end, - cb, (void *)lock); - if (info->clt_next_index > descr->cld_end) - break; - - if (res == CLP_GANG_RESCHED) - cond_resched(); - } while (res != CLP_GANG_OKAY); -out: - cl_io_fini(env, io); - return result; -} -EXPORT_SYMBOL(cl_lock_discard_pages); - -/** - * Eliminate all locks for a given object. - * - * Caller has to guarantee that no lock is in active use. - * - * \param cancel when this is set, cl_locks_prune() cancels locks before - * destroying. - */ -void cl_locks_prune(const struct lu_env *env, struct cl_object *obj, int cancel) -{ - struct cl_object_header *head; - struct cl_lock *lock; - - head = cl_object_header(obj); - /* - * If locks are destroyed without cancellation, all pages must be - * already destroyed (as otherwise they will be left unprotected). - */ - LASSERT(ergo(!cancel, - !head->coh_tree.rnode && head->coh_pages == 0)); - - spin_lock(&head->coh_lock_guard); - while (!list_empty(&head->coh_locks)) { - lock = container_of(head->coh_locks.next, - struct cl_lock, cll_linkage); - cl_lock_get_trust(lock); - spin_unlock(&head->coh_lock_guard); - lu_ref_add(&lock->cll_reference, "prune", current); - -again: - cl_lock_mutex_get(env, lock); - if (lock->cll_state < CLS_FREEING) { - LASSERT(lock->cll_users <= 1); - if (unlikely(lock->cll_users == 1)) { - struct l_wait_info lwi = { 0 }; - - cl_lock_mutex_put(env, lock); - l_wait_event(lock->cll_wq, - lock->cll_users == 0, - &lwi); - goto again; - } - - if (cancel) - cl_lock_cancel(env, lock); - cl_lock_delete(env, lock); - } - cl_lock_mutex_put(env, lock); - lu_ref_del(&lock->cll_reference, "prune", current); - cl_lock_put(env, lock); - spin_lock(&head->coh_lock_guard); - } - spin_unlock(&head->coh_lock_guard); -} -EXPORT_SYMBOL(cl_locks_prune); - -static struct cl_lock *cl_lock_hold_mutex(const struct lu_env *env, - const struct cl_io *io, - const struct cl_lock_descr *need, - const char *scope, const void *source) -{ - struct cl_lock *lock; - - while (1) { - lock = cl_lock_find(env, io, need); - if (IS_ERR(lock)) - break; - cl_lock_mutex_get(env, lock); - if (lock->cll_state < CLS_FREEING && - !(lock->cll_flags & CLF_CANCELLED)) { - cl_lock_hold_mod(env, lock, 1); - lu_ref_add(&lock->cll_holders, scope, source); - lu_ref_add(&lock->cll_reference, scope, source); - break; - } - cl_lock_mutex_put(env, lock); - cl_lock_put(env, lock); - } - return lock; -} - -/** - * Returns a lock matching \a need description with a reference and a hold on - * it. - * - * This is much like cl_lock_find(), except that cl_lock_hold() additionally - * guarantees that lock is not in the CLS_FREEING state on return. - */ -struct cl_lock *cl_lock_hold(const struct lu_env *env, const struct cl_io *io, - const struct cl_lock_descr *need, - const char *scope, const void *source) -{ - struct cl_lock *lock; - - lock = cl_lock_hold_mutex(env, io, need, scope, source); - if (!IS_ERR(lock)) - cl_lock_mutex_put(env, lock); - return lock; -} -EXPORT_SYMBOL(cl_lock_hold); - -/** - * Main high-level entry point of cl_lock interface that finds existing or - * enqueues new lock matching given description. - */ -struct cl_lock *cl_lock_request(const struct lu_env *env, struct cl_io *io, - const struct cl_lock_descr *need, - const char *scope, const void *source) -{ - struct cl_lock *lock; - int rc; - __u32 enqflags = need->cld_enq_flags; - - do { - lock = cl_lock_hold_mutex(env, io, need, scope, source); - if (IS_ERR(lock)) - break; - - rc = cl_enqueue_locked(env, lock, io, enqflags); - if (rc == 0) { - if (cl_lock_fits_into(env, lock, need, io)) { - if (!(enqflags & CEF_AGL)) { - cl_lock_mutex_put(env, lock); - cl_lock_lockdep_acquire(env, lock, - enqflags); - break; - } - rc = 1; - } - cl_unuse_locked(env, lock); - } - cl_lock_trace(D_DLMTRACE, env, - rc <= 0 ? "enqueue failed" : "agl succeed", lock); - cl_lock_hold_release(env, lock, scope, source); - cl_lock_mutex_put(env, lock); - lu_ref_del(&lock->cll_reference, scope, source); - cl_lock_put(env, lock); - if (rc > 0) { - LASSERT(enqflags & CEF_AGL); - lock = NULL; - } else if (rc != 0) { - lock = ERR_PTR(rc); - } - } while (rc == 0); - return lock; + return rc; } EXPORT_SYMBOL(cl_lock_request); /** - * Adds a hold to a known lock. - */ -void cl_lock_hold_add(const struct lu_env *env, struct cl_lock *lock, - const char *scope, const void *source) -{ - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - LASSERT(lock->cll_state != CLS_FREEING); - - cl_lock_hold_mod(env, lock, 1); - cl_lock_get(lock); - lu_ref_add(&lock->cll_holders, scope, source); - lu_ref_add(&lock->cll_reference, scope, source); -} -EXPORT_SYMBOL(cl_lock_hold_add); - -/** - * Releases a hold and a reference on a lock, on which caller acquired a - * mutex. - */ -void cl_lock_unhold(const struct lu_env *env, struct cl_lock *lock, - const char *scope, const void *source) -{ - LINVRNT(cl_lock_invariant(env, lock)); - cl_lock_hold_release(env, lock, scope, source); - lu_ref_del(&lock->cll_reference, scope, source); - cl_lock_put(env, lock); -} -EXPORT_SYMBOL(cl_lock_unhold); - -/** * Releases a hold and a reference on a lock, obtained by cl_lock_hold(). */ -void cl_lock_release(const struct lu_env *env, struct cl_lock *lock, - const char *scope, const void *source) +void cl_lock_release(const struct lu_env *env, struct cl_lock *lock) { - LINVRNT(cl_lock_invariant(env, lock)); cl_lock_trace(D_DLMTRACE, env, "release lock", lock); - cl_lock_mutex_get(env, lock); - cl_lock_hold_release(env, lock, scope, source); - cl_lock_mutex_put(env, lock); - lu_ref_del(&lock->cll_reference, scope, source); - cl_lock_put(env, lock); + cl_lock_cancel(env, lock); + cl_lock_fini(env, lock); } EXPORT_SYMBOL(cl_lock_release); -void cl_lock_user_add(const struct lu_env *env, struct cl_lock *lock) -{ - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - - cl_lock_used_mod(env, lock, 1); -} -EXPORT_SYMBOL(cl_lock_user_add); - -void cl_lock_user_del(const struct lu_env *env, struct cl_lock *lock) -{ - LINVRNT(cl_lock_is_mutexed(lock)); - LINVRNT(cl_lock_invariant(env, lock)); - LASSERT(lock->cll_users > 0); - - cl_lock_used_mod(env, lock, -1); - if (lock->cll_users == 0) - wake_up_all(&lock->cll_wq); -} -EXPORT_SYMBOL(cl_lock_user_del); - const char *cl_lock_mode_name(const enum cl_lock_mode mode) { static const char *names[] = { - [CLM_PHANTOM] = "P", [CLM_READ] = "R", [CLM_WRITE] = "W", [CLM_GROUP] = "G" @@ -2189,10 +261,8 @@ void cl_lock_print(const struct lu_env *env, void *cookie, lu_printer_t printer, const struct cl_lock *lock) { const struct cl_lock_slice *slice; - (*printer)(env, cookie, "lock@%p[%d %d %d %d %d %08lx] ", - lock, atomic_read(&lock->cll_ref), - lock->cll_state, lock->cll_error, lock->cll_holds, - lock->cll_users, lock->cll_flags); + + (*printer)(env, cookie, "lock@%p", lock); cl_lock_descr_print(env, cookie, printer, &lock->cll_descr); (*printer)(env, cookie, " {\n"); @@ -2207,13 +277,3 @@ void cl_lock_print(const struct lu_env *env, void *cookie, (*printer)(env, cookie, "} lock@%p\n", lock); } EXPORT_SYMBOL(cl_lock_print); - -int cl_lock_init(void) -{ - return lu_kmem_init(cl_lock_caches); -} - -void cl_lock_fini(void) -{ - lu_kmem_fini(cl_lock_caches); -} diff --git a/drivers/staging/lustre/lustre/obdclass/cl_object.c b/drivers/staging/lustre/lustre/obdclass/cl_object.c index 43e299d4d416..5940f30318ec 100644 --- a/drivers/staging/lustre/lustre/obdclass/cl_object.c +++ b/drivers/staging/lustre/lustre/obdclass/cl_object.c @@ -36,6 +36,7 @@ * Client Lustre Object. * * Author: Nikita Danilov <nikita.danilov@sun.com> + * Author: Jinshan Xiong <jinshan.xiong@intel.com> */ /* @@ -43,8 +44,6 @@ * * i_mutex * PG_locked - * ->coh_page_guard - * ->coh_lock_guard * ->coh_attr_guard * ->ls_guard */ @@ -63,10 +62,6 @@ static struct kmem_cache *cl_env_kmem; -/** Lock class of cl_object_header::coh_page_guard */ -static struct lock_class_key cl_page_guard_class; -/** Lock class of cl_object_header::coh_lock_guard */ -static struct lock_class_key cl_lock_guard_class; /** Lock class of cl_object_header::coh_attr_guard */ static struct lock_class_key cl_attr_guard_class; @@ -81,17 +76,9 @@ int cl_object_header_init(struct cl_object_header *h) result = lu_object_header_init(&h->coh_lu); if (result == 0) { - spin_lock_init(&h->coh_page_guard); - spin_lock_init(&h->coh_lock_guard); spin_lock_init(&h->coh_attr_guard); - lockdep_set_class(&h->coh_page_guard, &cl_page_guard_class); - lockdep_set_class(&h->coh_lock_guard, &cl_lock_guard_class); lockdep_set_class(&h->coh_attr_guard, &cl_attr_guard_class); - h->coh_pages = 0; - /* XXX hard coded GFP_* mask. */ - INIT_RADIX_TREE(&h->coh_tree, GFP_ATOMIC); - INIT_LIST_HEAD(&h->coh_locks); - h->coh_page_bufsize = ALIGN(sizeof(struct cl_page), 8); + h->coh_page_bufsize = 0; } return result; } @@ -145,7 +132,7 @@ EXPORT_SYMBOL(cl_object_get); /** * Returns the top-object for a given \a o. * - * \see cl_page_top(), cl_io_top() + * \see cl_io_top() */ struct cl_object *cl_object_top(struct cl_object *o) { @@ -315,6 +302,29 @@ int cl_conf_set(const struct lu_env *env, struct cl_object *obj, EXPORT_SYMBOL(cl_conf_set); /** + * Prunes caches of pages and locks for this object. + */ +int cl_object_prune(const struct lu_env *env, struct cl_object *obj) +{ + struct lu_object_header *top; + struct cl_object *o; + int result; + + top = obj->co_lu.lo_header; + result = 0; + list_for_each_entry(o, &top->loh_layers, co_lu.lo_linkage) { + if (o->co_ops->coo_prune) { + result = o->co_ops->coo_prune(env, o); + if (result != 0) + break; + } + } + + return result; +} +EXPORT_SYMBOL(cl_object_prune); + +/** * Helper function removing all object locks, and marking object for * deletion. All object pages must have been deleted at this point. * @@ -323,34 +333,12 @@ EXPORT_SYMBOL(cl_conf_set); */ void cl_object_kill(const struct lu_env *env, struct cl_object *obj) { - struct cl_object_header *hdr; - - hdr = cl_object_header(obj); - LASSERT(!hdr->coh_tree.rnode); - LASSERT(hdr->coh_pages == 0); + struct cl_object_header *hdr = cl_object_header(obj); set_bit(LU_OBJECT_HEARD_BANSHEE, &hdr->coh_lu.loh_flags); - /* - * Destroy all locks. Object destruction (including cl_inode_fini()) - * cannot cancel the locks, because in the case of a local client, - * where client and server share the same thread running - * prune_icache(), this can dead-lock with ldlm_cancel_handler() - * waiting on __wait_on_freeing_inode(). - */ - cl_locks_prune(env, obj, 0); } EXPORT_SYMBOL(cl_object_kill); -/** - * Prunes caches of pages and locks for this object. - */ -void cl_object_prune(const struct lu_env *env, struct cl_object *obj) -{ - cl_pages_prune(env, obj); - cl_locks_prune(env, obj, 1); -} -EXPORT_SYMBOL(cl_object_prune); - void cache_stats_init(struct cache_stats *cs, const char *name) { int i; @@ -383,6 +371,8 @@ static int cache_stats_print(const struct cache_stats *cs, return 0; } +static void cl_env_percpu_refill(void); + /** * Initialize client site. * @@ -397,11 +387,9 @@ int cl_site_init(struct cl_site *s, struct cl_device *d) result = lu_site_init(&s->cs_lu, &d->cd_lu_dev); if (result == 0) { cache_stats_init(&s->cs_pages, "pages"); - cache_stats_init(&s->cs_locks, "locks"); for (i = 0; i < ARRAY_SIZE(s->cs_pages_state); ++i) atomic_set(&s->cs_pages_state[0], 0); - for (i = 0; i < ARRAY_SIZE(s->cs_locks_state); ++i) - atomic_set(&s->cs_locks_state[i], 0); + cl_env_percpu_refill(); } return result; } @@ -435,15 +423,6 @@ int cl_site_stats_print(const struct cl_site *site, struct seq_file *m) [CPS_PAGEIN] = "r", [CPS_FREEING] = "f" }; - static const char *lstate[] = { - [CLS_NEW] = "n", - [CLS_QUEUING] = "q", - [CLS_ENQUEUED] = "e", - [CLS_HELD] = "h", - [CLS_INTRANSIT] = "t", - [CLS_CACHED] = "c", - [CLS_FREEING] = "f" - }; /* lookup hit total busy create pages: ...... ...... ...... ...... ...... [...... ...... ...... ......] @@ -457,12 +436,6 @@ locks: ...... ...... ...... ...... ...... [...... ...... ...... ...... ......] seq_printf(m, "%s: %u ", pstate[i], atomic_read(&site->cs_pages_state[i])); seq_printf(m, "]\n"); - cache_stats_print(&site->cs_locks, m, 0); - seq_printf(m, " ["); - for (i = 0; i < ARRAY_SIZE(site->cs_locks_state); ++i) - seq_printf(m, "%s: %u ", lstate[i], - atomic_read(&site->cs_locks_state[i])); - seq_printf(m, "]\n"); cache_stats_print(&cl_env_stats, m, 0); seq_printf(m, "\n"); return 0; @@ -492,6 +465,13 @@ EXPORT_SYMBOL(cl_site_stats_print); * bz20044, bz22683. */ +static LIST_HEAD(cl_envs); +static unsigned int cl_envs_cached_nr; +static unsigned int cl_envs_cached_max = 128; /* XXX: prototype: arbitrary limit + * for now. + */ +static DEFINE_SPINLOCK(cl_envs_guard); + struct cl_env { void *ce_magic; struct lu_env ce_lu; @@ -674,8 +654,9 @@ static struct lu_env *cl_env_new(__u32 ctx_tags, __u32 ses_tags, void *debug) lu_context_enter(&cle->ce_ses); env->le_ses = &cle->ce_ses; cl_env_init0(cle, debug); - } else + } else { lu_env_fini(env); + } } if (rc != 0) { kmem_cache_free(cl_env_kmem, cle); @@ -684,8 +665,9 @@ static struct lu_env *cl_env_new(__u32 ctx_tags, __u32 ses_tags, void *debug) CL_ENV_INC(create); CL_ENV_INC(total); } - } else + } else { env = ERR_PTR(-ENOMEM); + } return env; } @@ -697,6 +679,39 @@ static void cl_env_fini(struct cl_env *cle) kmem_cache_free(cl_env_kmem, cle); } +static struct lu_env *cl_env_obtain(void *debug) +{ + struct cl_env *cle; + struct lu_env *env; + + spin_lock(&cl_envs_guard); + LASSERT(equi(cl_envs_cached_nr == 0, list_empty(&cl_envs))); + if (cl_envs_cached_nr > 0) { + int rc; + + cle = container_of(cl_envs.next, struct cl_env, ce_linkage); + list_del_init(&cle->ce_linkage); + cl_envs_cached_nr--; + spin_unlock(&cl_envs_guard); + + env = &cle->ce_lu; + rc = lu_env_refill(env); + if (rc == 0) { + cl_env_init0(cle, debug); + lu_context_enter(&env->le_ctx); + lu_context_enter(&cle->ce_ses); + } else { + cl_env_fini(cle); + env = ERR_PTR(rc); + } + } else { + spin_unlock(&cl_envs_guard); + env = cl_env_new(lu_context_tags_default, + lu_session_tags_default, debug); + } + return env; +} + static inline struct cl_env *cl_env_container(struct lu_env *env) { return container_of(env, struct cl_env, ce_lu); @@ -727,6 +742,8 @@ static struct lu_env *cl_env_peek(int *refcheck) * Returns lu_env: if there already is an environment associated with the * current thread, it is returned, otherwise, new environment is allocated. * + * Allocations are amortized through the global cache of environments. + * * \param refcheck pointer to a counter used to detect environment leaks. In * the usual case cl_env_get() and cl_env_put() are called in the same lexical * scope and pointer to the same integer is passed as \a refcheck. This is @@ -740,10 +757,7 @@ struct lu_env *cl_env_get(int *refcheck) env = cl_env_peek(refcheck); if (!env) { - env = cl_env_new(lu_context_tags_default, - lu_session_tags_default, - __builtin_return_address(0)); - + env = cl_env_obtain(__builtin_return_address(0)); if (!IS_ERR(env)) { struct cl_env *cle; @@ -787,6 +801,32 @@ static void cl_env_exit(struct cl_env *cle) } /** + * Finalizes and frees a given number of cached environments. This is done to + * (1) free some memory (not currently hooked into VM), or (2) release + * references to modules. + */ +unsigned int cl_env_cache_purge(unsigned int nr) +{ + struct cl_env *cle; + + spin_lock(&cl_envs_guard); + for (; !list_empty(&cl_envs) && nr > 0; --nr) { + cle = container_of(cl_envs.next, struct cl_env, ce_linkage); + list_del_init(&cle->ce_linkage); + LASSERT(cl_envs_cached_nr > 0); + cl_envs_cached_nr--; + spin_unlock(&cl_envs_guard); + + cl_env_fini(cle); + spin_lock(&cl_envs_guard); + } + LASSERT(equi(cl_envs_cached_nr == 0, list_empty(&cl_envs))); + spin_unlock(&cl_envs_guard); + return nr; +} +EXPORT_SYMBOL(cl_env_cache_purge); + +/** * Release an environment. * * Decrement \a env reference counter. When counter drops to 0, nothing in @@ -808,7 +848,22 @@ void cl_env_put(struct lu_env *env, int *refcheck) cl_env_detach(cle); cle->ce_debug = NULL; cl_env_exit(cle); - cl_env_fini(cle); + /* + * Don't bother to take a lock here. + * + * Return environment to the cache only when it was allocated + * with the standard tags. + */ + if (cl_envs_cached_nr < cl_envs_cached_max && + (env->le_ctx.lc_tags & ~LCT_HAS_EXIT) == LCT_CL_THREAD && + (env->le_ses->lc_tags & ~LCT_HAS_EXIT) == LCT_SESSION) { + spin_lock(&cl_envs_guard); + list_add(&cle->ce_linkage, &cl_envs); + cl_envs_cached_nr++; + spin_unlock(&cl_envs_guard); + } else { + cl_env_fini(cle); + } } } EXPORT_SYMBOL(cl_env_put); @@ -914,6 +969,104 @@ void cl_lvb2attr(struct cl_attr *attr, const struct ost_lvb *lvb) } EXPORT_SYMBOL(cl_lvb2attr); +static struct cl_env cl_env_percpu[NR_CPUS]; + +static int cl_env_percpu_init(void) +{ + struct cl_env *cle; + int tags = LCT_REMEMBER | LCT_NOREF; + int i, j; + int rc = 0; + + for_each_possible_cpu(i) { + struct lu_env *env; + + cle = &cl_env_percpu[i]; + env = &cle->ce_lu; + + INIT_LIST_HEAD(&cle->ce_linkage); + cle->ce_magic = &cl_env_init0; + rc = lu_env_init(env, LCT_CL_THREAD | tags); + if (rc == 0) { + rc = lu_context_init(&cle->ce_ses, LCT_SESSION | tags); + if (rc == 0) { + lu_context_enter(&cle->ce_ses); + env->le_ses = &cle->ce_ses; + } else { + lu_env_fini(env); + } + } + if (rc != 0) + break; + } + if (rc != 0) { + /* Indices 0 to i (excluding i) were correctly initialized, + * thus we must uninitialize up to i, the rest are undefined. + */ + for (j = 0; j < i; j++) { + cle = &cl_env_percpu[i]; + lu_context_exit(&cle->ce_ses); + lu_context_fini(&cle->ce_ses); + lu_env_fini(&cle->ce_lu); + } + } + + return rc; +} + +static void cl_env_percpu_fini(void) +{ + int i; + + for_each_possible_cpu(i) { + struct cl_env *cle = &cl_env_percpu[i]; + + lu_context_exit(&cle->ce_ses); + lu_context_fini(&cle->ce_ses); + lu_env_fini(&cle->ce_lu); + } +} + +static void cl_env_percpu_refill(void) +{ + int i; + + for_each_possible_cpu(i) + lu_env_refill(&cl_env_percpu[i].ce_lu); +} + +void cl_env_percpu_put(struct lu_env *env) +{ + struct cl_env *cle; + int cpu; + + cpu = smp_processor_id(); + cle = cl_env_container(env); + LASSERT(cle == &cl_env_percpu[cpu]); + + cle->ce_ref--; + LASSERT(cle->ce_ref == 0); + + CL_ENV_DEC(busy); + cl_env_detach(cle); + cle->ce_debug = NULL; + + put_cpu(); +} +EXPORT_SYMBOL(cl_env_percpu_put); + +struct lu_env *cl_env_percpu_get(void) +{ + struct cl_env *cle; + + cle = &cl_env_percpu[get_cpu()]; + cl_env_init0(cle, __builtin_return_address(0)); + + cl_env_attach(cle); + return &cle->ce_lu; +} +EXPORT_SYMBOL(cl_env_percpu_get); + /***************************************************************************** * * Temporary prototype thing: mirror obd-devices into cl devices. @@ -944,8 +1097,9 @@ struct cl_device *cl_type_setup(const struct lu_env *env, struct lu_site *site, CERROR("can't init device '%s', %d\n", typename, rc); d = ERR_PTR(rc); } - } else + } else { CERROR("Cannot allocate device: '%s'\n", typename); + } return lu2cl_dev(d); } EXPORT_SYMBOL(cl_type_setup); @@ -959,12 +1113,6 @@ void cl_stack_fini(const struct lu_env *env, struct cl_device *cl) } EXPORT_SYMBOL(cl_stack_fini); -int cl_lock_init(void); -void cl_lock_fini(void); - -int cl_page_init(void); -void cl_page_fini(void); - static struct lu_context_key cl_key; struct cl_thread_info *cl_env_info(const struct lu_env *env) @@ -1059,17 +1207,13 @@ int cl_global_init(void) if (result) goto out_kmem; - result = cl_lock_init(); + result = cl_env_percpu_init(); if (result) + /* no cl_env_percpu_fini on error */ goto out_context; - result = cl_page_init(); - if (result) - goto out_lock; - return 0; -out_lock: - cl_lock_fini(); + out_context: lu_context_key_degister(&cl_key); out_kmem: @@ -1084,8 +1228,7 @@ out_store: */ void cl_global_fini(void) { - cl_lock_fini(); - cl_page_fini(); + cl_env_percpu_fini(); lu_context_key_degister(&cl_key); lu_kmem_fini(cl_object_caches); cl_env_store_fini(); diff --git a/drivers/staging/lustre/lustre/obdclass/cl_page.c b/drivers/staging/lustre/lustre/obdclass/cl_page.c index 394580016638..b754f516e557 100644 --- a/drivers/staging/lustre/lustre/obdclass/cl_page.c +++ b/drivers/staging/lustre/lustre/obdclass/cl_page.c @@ -36,6 +36,7 @@ * Client Lustre Page. * * Author: Nikita Danilov <nikita.danilov@sun.com> + * Author: Jinshan Xiong <jinshan.xiong@intel.com> */ #define DEBUG_SUBSYSTEM S_CLASS @@ -48,8 +49,7 @@ #include "../include/cl_object.h" #include "cl_internal.h" -static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, - int radix); +static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg); # define PASSERT(env, page, expr) \ do { \ @@ -63,24 +63,11 @@ static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp)) /** - * Internal version of cl_page_top, it should be called if the page is - * known to be not freed, says with page referenced, or radix tree lock held, - * or page owned. - */ -static struct cl_page *cl_page_top_trusted(struct cl_page *page) -{ - while (page->cp_parent) - page = page->cp_parent; - return page; -} - -/** * Internal version of cl_page_get(). * * This function can be used to obtain initial reference to previously * unreferenced cached object. It can be called only if concurrent page - * reclamation is somehow prevented, e.g., by locking page radix-tree - * (cl_object_header::hdr->coh_page_guard), or by keeping a lock on a VM page, + * reclamation is somehow prevented, e.g., by keeping a lock on a VM page, * associated with \a page. * * Use with care! Not exported. @@ -103,142 +90,12 @@ cl_page_at_trusted(const struct cl_page *page, { const struct cl_page_slice *slice; - page = cl_page_top_trusted((struct cl_page *)page); - do { - list_for_each_entry(slice, &page->cp_layers, cpl_linkage) { - if (slice->cpl_obj->co_lu.lo_dev->ld_type == dtype) - return slice; - } - page = page->cp_child; - } while (page); - return NULL; -} - -/** - * Returns a page with given index in the given object, or NULL if no page is - * found. Acquires a reference on \a page. - * - * Locking: called under cl_object_header::coh_page_guard spin-lock. - */ -struct cl_page *cl_page_lookup(struct cl_object_header *hdr, pgoff_t index) -{ - struct cl_page *page; - - assert_spin_locked(&hdr->coh_page_guard); - - page = radix_tree_lookup(&hdr->coh_tree, index); - if (page) - cl_page_get_trust(page); - return page; -} -EXPORT_SYMBOL(cl_page_lookup); - -/** - * Returns a list of pages by a given [start, end] of \a obj. - * - * \param resched If not NULL, then we give up before hogging CPU for too - * long and set *resched = 1, in that case caller should implement a retry - * logic. - * - * Gang tree lookup (radix_tree_gang_lookup()) optimization is absolutely - * crucial in the face of [offset, EOF] locks. - * - * Return at least one page in @queue unless there is no covered page. - */ -int cl_page_gang_lookup(const struct lu_env *env, struct cl_object *obj, - struct cl_io *io, pgoff_t start, pgoff_t end, - cl_page_gang_cb_t cb, void *cbdata) -{ - struct cl_object_header *hdr; - struct cl_page *page; - struct cl_page **pvec; - const struct cl_page_slice *slice; - const struct lu_device_type *dtype; - pgoff_t idx; - unsigned int nr; - unsigned int i; - unsigned int j; - int res = CLP_GANG_OKAY; - int tree_lock = 1; - - idx = start; - hdr = cl_object_header(obj); - pvec = cl_env_info(env)->clt_pvec; - dtype = cl_object_top(obj)->co_lu.lo_dev->ld_type; - spin_lock(&hdr->coh_page_guard); - while ((nr = radix_tree_gang_lookup(&hdr->coh_tree, (void **)pvec, - idx, CLT_PVEC_SIZE)) > 0) { - int end_of_region = 0; - - idx = pvec[nr - 1]->cp_index + 1; - for (i = 0, j = 0; i < nr; ++i) { - page = pvec[i]; - pvec[i] = NULL; - - LASSERT(page->cp_type == CPT_CACHEABLE); - if (page->cp_index > end) { - end_of_region = 1; - break; - } - if (page->cp_state == CPS_FREEING) - continue; - - slice = cl_page_at_trusted(page, dtype); - /* - * Pages for lsm-less file has no underneath sub-page - * for osc, in case of ... - */ - PASSERT(env, page, slice); - - page = slice->cpl_page; - /* - * Can safely call cl_page_get_trust() under - * radix-tree spin-lock. - * - * XXX not true, because @page is from object another - * than @hdr and protected by different tree lock. - */ - cl_page_get_trust(page); - lu_ref_add_atomic(&page->cp_reference, - "gang_lookup", current); - pvec[j++] = page; - } - - /* - * Here a delicate locking dance is performed. Current thread - * holds a reference to a page, but has to own it before it - * can be placed into queue. Owning implies waiting, so - * radix-tree lock is to be released. After a wait one has to - * check that pages weren't truncated (cl_page_own() returns - * error in the latter case). - */ - spin_unlock(&hdr->coh_page_guard); - tree_lock = 0; - - for (i = 0; i < j; ++i) { - page = pvec[i]; - if (res == CLP_GANG_OKAY) - res = (*cb)(env, io, page, cbdata); - lu_ref_del(&page->cp_reference, - "gang_lookup", current); - cl_page_put(env, page); - } - if (nr < CLT_PVEC_SIZE || end_of_region) - break; - - if (res == CLP_GANG_OKAY && need_resched()) - res = CLP_GANG_RESCHED; - if (res != CLP_GANG_OKAY) - break; - - spin_lock(&hdr->coh_page_guard); - tree_lock = 1; + list_for_each_entry(slice, &page->cp_layers, cpl_linkage) { + if (slice->cpl_obj->co_lu.lo_dev->ld_type == dtype) + return slice; } - if (tree_lock) - spin_unlock(&hdr->coh_page_guard); - return res; + return NULL; } -EXPORT_SYMBOL(cl_page_gang_lookup); static void cl_page_free(const struct lu_env *env, struct cl_page *page) { @@ -247,17 +104,16 @@ static void cl_page_free(const struct lu_env *env, struct cl_page *page) PASSERT(env, page, list_empty(&page->cp_batch)); PASSERT(env, page, !page->cp_owner); PASSERT(env, page, !page->cp_req); - PASSERT(env, page, !page->cp_parent); PASSERT(env, page, page->cp_state == CPS_FREEING); - might_sleep(); while (!list_empty(&page->cp_layers)) { struct cl_page_slice *slice; slice = list_entry(page->cp_layers.next, struct cl_page_slice, cpl_linkage); list_del_init(page->cp_layers.next); - slice->cpl_ops->cpo_fini(env, slice); + if (unlikely(slice->cpl_ops->cpo_fini)) + slice->cpl_ops->cpo_fini(env, slice); } lu_object_ref_del_at(&obj->co_lu, &page->cp_obj_ref, "cl_page", page); cl_object_put(env, obj); @@ -276,10 +132,10 @@ static inline void cl_page_state_set_trust(struct cl_page *page, *(enum cl_page_state *)&page->cp_state = state; } -static struct cl_page *cl_page_alloc(const struct lu_env *env, - struct cl_object *o, pgoff_t ind, - struct page *vmpage, - enum cl_page_type type) +struct cl_page *cl_page_alloc(const struct lu_env *env, + struct cl_object *o, pgoff_t ind, + struct page *vmpage, + enum cl_page_type type) { struct cl_page *page; struct lu_object_header *head; @@ -289,13 +145,11 @@ static struct cl_page *cl_page_alloc(const struct lu_env *env, int result = 0; atomic_set(&page->cp_ref, 1); - if (type == CPT_CACHEABLE) /* for radix tree */ - atomic_inc(&page->cp_ref); page->cp_obj = o; cl_object_get(o); lu_object_ref_add_at(&o->co_lu, &page->cp_obj_ref, "cl_page", page); - page->cp_index = ind; + page->cp_vmpage = vmpage; cl_page_state_set_trust(page, CPS_CACHED); page->cp_type = type; INIT_LIST_HEAD(&page->cp_layers); @@ -306,10 +160,10 @@ static struct cl_page *cl_page_alloc(const struct lu_env *env, head = o->co_lu.lo_header; list_for_each_entry(o, &head->loh_layers, co_lu.lo_linkage) { if (o->co_ops->coo_page_init) { - result = o->co_ops->coo_page_init(env, o, - page, vmpage); + result = o->co_ops->coo_page_init(env, o, page, + ind); if (result != 0) { - cl_page_delete0(env, page, 0); + cl_page_delete0(env, page); cl_page_free(env, page); page = ERR_PTR(result); break; @@ -321,6 +175,7 @@ static struct cl_page *cl_page_alloc(const struct lu_env *env, } return page; } +EXPORT_SYMBOL(cl_page_alloc); /** * Returns a cl_page with index \a idx at the object \a o, and associated with @@ -333,16 +188,13 @@ static struct cl_page *cl_page_alloc(const struct lu_env *env, * * \see cl_object_find(), cl_lock_find() */ -static struct cl_page *cl_page_find0(const struct lu_env *env, - struct cl_object *o, - pgoff_t idx, struct page *vmpage, - enum cl_page_type type, - struct cl_page *parent) +struct cl_page *cl_page_find(const struct lu_env *env, + struct cl_object *o, + pgoff_t idx, struct page *vmpage, + enum cl_page_type type) { struct cl_page *page = NULL; - struct cl_page *ghost = NULL; struct cl_object_header *hdr; - int err; LASSERT(type == CPT_CACHEABLE || type == CPT_TRANSIENT); might_sleep(); @@ -368,120 +220,25 @@ static struct cl_page *cl_page_find0(const struct lu_env *env, * reference on it. */ page = cl_vmpage_page(vmpage, o); - PINVRNT(env, page, - ergo(page, - cl_page_vmpage(env, page) == vmpage && - (void *)radix_tree_lookup(&hdr->coh_tree, - idx) == page)); - } - if (page) - return page; + if (page) + return page; + } /* allocate and initialize cl_page */ page = cl_page_alloc(env, o, idx, vmpage, type); - if (IS_ERR(page)) - return page; - - if (type == CPT_TRANSIENT) { - if (parent) { - LASSERT(!page->cp_parent); - page->cp_parent = parent; - parent->cp_child = page; - } - return page; - } - - /* - * XXX optimization: use radix_tree_preload() here, and change tree - * gfp mask to GFP_KERNEL in cl_object_header_init(). - */ - spin_lock(&hdr->coh_page_guard); - err = radix_tree_insert(&hdr->coh_tree, idx, page); - if (err != 0) { - ghost = page; - /* - * Noted by Jay: a lock on \a vmpage protects cl_page_find() - * from this race, but - * - * 0. it's better to have cl_page interface "locally - * consistent" so that its correctness can be reasoned - * about without appealing to the (obscure world of) VM - * locking. - * - * 1. handling this race allows ->coh_tree to remain - * consistent even when VM locking is somehow busted, - * which is very useful during diagnosing and debugging. - */ - page = ERR_PTR(err); - CL_PAGE_DEBUG(D_ERROR, env, ghost, - "fail to insert into radix tree: %d\n", err); - } else { - if (parent) { - LASSERT(!page->cp_parent); - page->cp_parent = parent; - parent->cp_child = page; - } - hdr->coh_pages++; - } - spin_unlock(&hdr->coh_page_guard); - - if (unlikely(ghost)) { - cl_page_delete0(env, ghost, 0); - cl_page_free(env, ghost); - } return page; } - -struct cl_page *cl_page_find(const struct lu_env *env, struct cl_object *o, - pgoff_t idx, struct page *vmpage, - enum cl_page_type type) -{ - return cl_page_find0(env, o, idx, vmpage, type, NULL); -} EXPORT_SYMBOL(cl_page_find); -struct cl_page *cl_page_find_sub(const struct lu_env *env, struct cl_object *o, - pgoff_t idx, struct page *vmpage, - struct cl_page *parent) -{ - return cl_page_find0(env, o, idx, vmpage, parent->cp_type, parent); -} -EXPORT_SYMBOL(cl_page_find_sub); - static inline int cl_page_invariant(const struct cl_page *pg) { - struct cl_object_header *header; - struct cl_page *parent; - struct cl_page *child; - struct cl_io *owner; - /* * Page invariant is protected by a VM lock. */ LINVRNT(cl_page_is_vmlocked(NULL, pg)); - header = cl_object_header(pg->cp_obj); - parent = pg->cp_parent; - child = pg->cp_child; - owner = pg->cp_owner; - - return cl_page_in_use(pg) && - ergo(parent, parent->cp_child == pg) && - ergo(child, child->cp_parent == pg) && - ergo(child, pg->cp_obj != child->cp_obj) && - ergo(parent, pg->cp_obj != parent->cp_obj) && - ergo(owner && parent, - parent->cp_owner == pg->cp_owner->ci_parent) && - ergo(owner && child, child->cp_owner->ci_parent == owner) && - /* - * Either page is early in initialization (has neither child - * nor parent yet), or it is in the object radix tree. - */ - ergo(pg->cp_state < CPS_FREEING && pg->cp_type == CPT_CACHEABLE, - (void *)radix_tree_lookup(&header->coh_tree, - pg->cp_index) == pg || - (!child && !parent)); + return cl_page_in_use_noref(pg); } static void cl_page_state_set0(const struct lu_env *env, @@ -534,13 +291,9 @@ static void cl_page_state_set0(const struct lu_env *env, old = page->cp_state; PASSERT(env, page, allowed_transitions[old][state]); CL_PAGE_HEADER(D_TRACE, env, page, "%d -> %d\n", old, state); - for (; page; page = page->cp_child) { - PASSERT(env, page, page->cp_state == old); - PASSERT(env, page, - equi(state == CPS_OWNED, page->cp_owner)); - - cl_page_state_set_trust(page, state); - } + PASSERT(env, page, page->cp_state == old); + PASSERT(env, page, equi(state == CPS_OWNED, page->cp_owner)); + cl_page_state_set_trust(page, state); } static void cl_page_state_set(const struct lu_env *env, @@ -574,8 +327,6 @@ EXPORT_SYMBOL(cl_page_get); */ void cl_page_put(const struct lu_env *env, struct cl_page *page) { - PASSERT(env, page, atomic_read(&page->cp_ref) > !!page->cp_parent); - CL_PAGE_HEADER(D_TRACE, env, page, "%d\n", atomic_read(&page->cp_ref)); @@ -595,34 +346,10 @@ void cl_page_put(const struct lu_env *env, struct cl_page *page) EXPORT_SYMBOL(cl_page_put); /** - * Returns a VM page associated with a given cl_page. - */ -struct page *cl_page_vmpage(const struct lu_env *env, struct cl_page *page) -{ - const struct cl_page_slice *slice; - - /* - * Find uppermost layer with ->cpo_vmpage() method, and return its - * result. - */ - page = cl_page_top(page); - do { - list_for_each_entry(slice, &page->cp_layers, cpl_linkage) { - if (slice->cpl_ops->cpo_vmpage) - return slice->cpl_ops->cpo_vmpage(env, slice); - } - page = page->cp_child; - } while (page); - LBUG(); /* ->cpo_vmpage() has to be defined somewhere in the stack */ -} -EXPORT_SYMBOL(cl_page_vmpage); - -/** * Returns a cl_page associated with a VM page, and given cl_object. */ struct cl_page *cl_vmpage_page(struct page *vmpage, struct cl_object *obj) { - struct cl_page *top; struct cl_page *page; KLASSERT(PageLocked(vmpage)); @@ -633,36 +360,15 @@ struct cl_page *cl_vmpage_page(struct page *vmpage, struct cl_object *obj) * bottom-to-top pass. */ - /* - * This loop assumes that ->private points to the top-most page. This - * can be rectified easily. - */ - top = (struct cl_page *)vmpage->private; - if (!top) - return NULL; - - for (page = top; page; page = page->cp_child) { - if (cl_object_same(page->cp_obj, obj)) { - cl_page_get_trust(page); - break; - } + page = (struct cl_page *)vmpage->private; + if (page) { + cl_page_get_trust(page); + LASSERT(page->cp_type == CPT_CACHEABLE); } - LASSERT(ergo(page, page->cp_type == CPT_CACHEABLE)); return page; } EXPORT_SYMBOL(cl_vmpage_page); -/** - * Returns the top-page for a given page. - * - * \see cl_object_top(), cl_io_top() - */ -struct cl_page *cl_page_top(struct cl_page *page) -{ - return cl_page_top_trusted(page); -} -EXPORT_SYMBOL(cl_page_top); - const struct cl_page_slice *cl_page_at(const struct cl_page *page, const struct lu_device_type *dtype) { @@ -682,26 +388,43 @@ EXPORT_SYMBOL(cl_page_at); int (*__method)_proto; \ \ __result = 0; \ - __page = cl_page_top(__page); \ - do { \ - list_for_each_entry(__scan, &__page->cp_layers, \ - cpl_linkage) { \ - __method = *(void **)((char *)__scan->cpl_ops + \ - __op); \ - if (__method) { \ - __result = (*__method)(__env, __scan, \ - ## __VA_ARGS__); \ - if (__result != 0) \ - break; \ - } \ - } \ - __page = __page->cp_child; \ - } while (__page && __result == 0); \ + list_for_each_entry(__scan, &__page->cp_layers, cpl_linkage) { \ + __method = *(void **)((char *)__scan->cpl_ops + __op); \ + if (__method) { \ + __result = (*__method)(__env, __scan, ## __VA_ARGS__); \ + if (__result != 0) \ + break; \ + } \ + } \ if (__result > 0) \ __result = 0; \ __result; \ }) +#define CL_PAGE_INVOKE_REVERSE(_env, _page, _op, _proto, ...) \ +({ \ + const struct lu_env *__env = (_env); \ + struct cl_page *__page = (_page); \ + const struct cl_page_slice *__scan; \ + int __result; \ + ptrdiff_t __op = (_op); \ + int (*__method)_proto; \ + \ + __result = 0; \ + list_for_each_entry_reverse(__scan, &__page->cp_layers, \ + cpl_linkage) { \ + __method = *(void **)((char *)__scan->cpl_ops + __op); \ + if (__method) { \ + __result = (*__method)(__env, __scan, ## __VA_ARGS__); \ + if (__result != 0) \ + break; \ + } \ + } \ + if (__result > 0) \ + __result = 0; \ + __result; \ +}) + #define CL_PAGE_INVOID(_env, _page, _op, _proto, ...) \ do { \ const struct lu_env *__env = (_env); \ @@ -710,18 +433,11 @@ do { \ ptrdiff_t __op = (_op); \ void (*__method)_proto; \ \ - __page = cl_page_top(__page); \ - do { \ - list_for_each_entry(__scan, &__page->cp_layers, \ - cpl_linkage) { \ - __method = *(void **)((char *)__scan->cpl_ops + \ - __op); \ - if (__method) \ - (*__method)(__env, __scan, \ - ## __VA_ARGS__); \ - } \ - __page = __page->cp_child; \ - } while (__page); \ + list_for_each_entry(__scan, &__page->cp_layers, cpl_linkage) { \ + __method = *(void **)((char *)__scan->cpl_ops + __op); \ + if (__method) \ + (*__method)(__env, __scan, ## __VA_ARGS__); \ + } \ } while (0) #define CL_PAGE_INVOID_REVERSE(_env, _page, _op, _proto, ...) \ @@ -732,20 +448,11 @@ do { \ ptrdiff_t __op = (_op); \ void (*__method)_proto; \ \ - /* get to the bottom page. */ \ - while (__page->cp_child) \ - __page = __page->cp_child; \ - do { \ - list_for_each_entry_reverse(__scan, &__page->cp_layers, \ - cpl_linkage) { \ - __method = *(void **)((char *)__scan->cpl_ops + \ - __op); \ - if (__method) \ - (*__method)(__env, __scan, \ - ## __VA_ARGS__); \ - } \ - __page = __page->cp_parent; \ - } while (__page); \ + list_for_each_entry_reverse(__scan, &__page->cp_layers, cpl_linkage) { \ + __method = *(void **)((char *)__scan->cpl_ops + __op); \ + if (__method) \ + (*__method)(__env, __scan, ## __VA_ARGS__); \ + } \ } while (0) static int cl_page_invoke(const struct lu_env *env, @@ -771,20 +478,17 @@ static void cl_page_invoid(const struct lu_env *env, static void cl_page_owner_clear(struct cl_page *page) { - for (page = cl_page_top(page); page; page = page->cp_child) { - if (page->cp_owner) { - LASSERT(page->cp_owner->ci_owned_nr > 0); - page->cp_owner->ci_owned_nr--; - page->cp_owner = NULL; - page->cp_task = NULL; - } + if (page->cp_owner) { + LASSERT(page->cp_owner->ci_owned_nr > 0); + page->cp_owner->ci_owned_nr--; + page->cp_owner = NULL; + page->cp_task = NULL; } } static void cl_page_owner_set(struct cl_page *page) { - for (page = cl_page_top(page); page; page = page->cp_child) - page->cp_owner->ci_owned_nr++; + page->cp_owner->ci_owned_nr++; } void cl_page_disown0(const struct lu_env *env, @@ -794,7 +498,7 @@ void cl_page_disown0(const struct lu_env *env, state = pg->cp_state; PINVRNT(env, pg, state == CPS_OWNED || state == CPS_FREEING); - PINVRNT(env, pg, cl_page_invariant(pg)); + PINVRNT(env, pg, cl_page_invariant(pg) || state == CPS_FREEING); cl_page_owner_clear(pg); if (state == CPS_OWNED) @@ -815,8 +519,9 @@ void cl_page_disown0(const struct lu_env *env, */ int cl_page_is_owned(const struct cl_page *pg, const struct cl_io *io) { + struct cl_io *top = cl_io_top((struct cl_io *)io); LINVRNT(cl_object_same(pg->cp_obj, io->ci_obj)); - return pg->cp_state == CPS_OWNED && pg->cp_owner == io; + return pg->cp_state == CPS_OWNED && pg->cp_owner == top; } EXPORT_SYMBOL(cl_page_is_owned); @@ -847,7 +552,6 @@ static int cl_page_own0(const struct lu_env *env, struct cl_io *io, PINVRNT(env, pg, !cl_page_is_owned(pg, io)); - pg = cl_page_top(pg); io = cl_io_top(io); if (pg->cp_state == CPS_FREEING) { @@ -861,7 +565,7 @@ static int cl_page_own0(const struct lu_env *env, struct cl_io *io, if (result == 0) { PASSERT(env, pg, !pg->cp_owner); PASSERT(env, pg, !pg->cp_req); - pg->cp_owner = io; + pg->cp_owner = cl_io_top(io); pg->cp_task = current; cl_page_owner_set(pg); if (pg->cp_state != CPS_FREEING) { @@ -914,12 +618,11 @@ void cl_page_assume(const struct lu_env *env, { PINVRNT(env, pg, cl_object_same(pg->cp_obj, io->ci_obj)); - pg = cl_page_top(pg); io = cl_io_top(io); cl_page_invoid(env, io, pg, CL_PAGE_OP(cpo_assume)); PASSERT(env, pg, !pg->cp_owner); - pg->cp_owner = io; + pg->cp_owner = cl_io_top(io); pg->cp_task = current; cl_page_owner_set(pg); cl_page_state_set(env, pg, CPS_OWNED); @@ -943,7 +646,6 @@ void cl_page_unassume(const struct lu_env *env, PINVRNT(env, pg, cl_page_is_owned(pg, io)); PINVRNT(env, pg, cl_page_invariant(pg)); - pg = cl_page_top(pg); io = cl_io_top(io); cl_page_owner_clear(pg); cl_page_state_set(env, pg, CPS_CACHED); @@ -968,9 +670,9 @@ EXPORT_SYMBOL(cl_page_unassume); void cl_page_disown(const struct lu_env *env, struct cl_io *io, struct cl_page *pg) { - PINVRNT(env, pg, cl_page_is_owned(pg, io)); + PINVRNT(env, pg, cl_page_is_owned(pg, io) || + pg->cp_state == CPS_FREEING); - pg = cl_page_top(pg); io = cl_io_top(io); cl_page_disown0(env, io, pg); } @@ -1001,12 +703,8 @@ EXPORT_SYMBOL(cl_page_discard); * pages, e.g,. in a error handling cl_page_find()->cl_page_delete0() * path. Doesn't check page invariant. */ -static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, - int radix) +static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg) { - struct cl_page *tmp = pg; - - PASSERT(env, pg, pg == cl_page_top(pg)); PASSERT(env, pg, pg->cp_state != CPS_FREEING); /* @@ -1014,41 +712,11 @@ static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, */ cl_page_owner_clear(pg); - /* - * unexport the page firstly before freeing it so that - * the page content is considered to be invalid. - * We have to do this because a CPS_FREEING cl_page may - * be NOT under the protection of a cl_lock. - * Afterwards, if this page is found by other threads, then this - * page will be forced to reread. - */ - cl_page_export(env, pg, 0); cl_page_state_set0(env, pg, CPS_FREEING); - CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_delete), - (const struct lu_env *, const struct cl_page_slice *)); - - if (tmp->cp_type == CPT_CACHEABLE) { - if (!radix) - /* !radix means that @pg is not yet in the radix tree, - * skip removing it. - */ - tmp = pg->cp_child; - for (; tmp; tmp = tmp->cp_child) { - void *value; - struct cl_object_header *hdr; - - hdr = cl_object_header(tmp->cp_obj); - spin_lock(&hdr->coh_page_guard); - value = radix_tree_delete(&hdr->coh_tree, - tmp->cp_index); - PASSERT(env, tmp, value == tmp); - PASSERT(env, tmp, hdr->coh_pages > 0); - hdr->coh_pages--; - spin_unlock(&hdr->coh_page_guard); - cl_page_put(env, tmp); - } - } + CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_delete), + (const struct lu_env *, + const struct cl_page_slice *)); } /** @@ -1070,7 +738,6 @@ static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, * Once page reaches cl_page_state::CPS_FREEING, all remaining references will * drain after some time, at which point page will be recycled. * - * \pre pg == cl_page_top(pg) * \pre VM page is locked * \post pg->cp_state == CPS_FREEING * @@ -1079,30 +746,11 @@ static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg, void cl_page_delete(const struct lu_env *env, struct cl_page *pg) { PINVRNT(env, pg, cl_page_invariant(pg)); - cl_page_delete0(env, pg, 1); + cl_page_delete0(env, pg); } EXPORT_SYMBOL(cl_page_delete); /** - * Unmaps page from user virtual memory. - * - * Calls cl_page_operations::cpo_unmap() through all layers top-to-bottom. The - * layer responsible for VM interaction has to unmap page from user space - * virtual memory. - * - * \see cl_page_operations::cpo_unmap() - */ -int cl_page_unmap(const struct lu_env *env, - struct cl_io *io, struct cl_page *pg) -{ - PINVRNT(env, pg, cl_page_is_owned(pg, io)); - PINVRNT(env, pg, cl_page_invariant(pg)); - - return cl_page_invoke(env, io, pg, CL_PAGE_OP(cpo_unmap)); -} -EXPORT_SYMBOL(cl_page_unmap); - -/** * Marks page up-to-date. * * Call cl_page_operations::cpo_export() through all layers top-to-bottom. The @@ -1129,7 +777,6 @@ int cl_page_is_vmlocked(const struct lu_env *env, const struct cl_page *pg) int result; const struct cl_page_slice *slice; - pg = cl_page_top_trusted((struct cl_page *)pg); slice = container_of(pg->cp_layers.next, const struct cl_page_slice, cpl_linkage); PASSERT(env, pg, slice->cpl_ops->cpo_is_vmlocked); @@ -1241,7 +888,7 @@ void cl_page_completion(const struct lu_env *env, cl_page_put(env, pg); if (anchor) - cl_sync_io_note(anchor, ioret); + cl_sync_io_note(env, anchor, ioret); } EXPORT_SYMBOL(cl_page_completion); @@ -1276,44 +923,6 @@ int cl_page_make_ready(const struct lu_env *env, struct cl_page *pg, EXPORT_SYMBOL(cl_page_make_ready); /** - * Notify layers that high level io decided to place this page into a cache - * for future transfer. - * - * The layer implementing transfer engine (osc) has to register this page in - * its queues. - * - * \pre cl_page_is_owned(pg, io) - * \post cl_page_is_owned(pg, io) - * - * \see cl_page_operations::cpo_cache_add() - */ -int cl_page_cache_add(const struct lu_env *env, struct cl_io *io, - struct cl_page *pg, enum cl_req_type crt) -{ - const struct cl_page_slice *scan; - int result = 0; - - PINVRNT(env, pg, crt < CRT_NR); - PINVRNT(env, pg, cl_page_is_owned(pg, io)); - PINVRNT(env, pg, cl_page_invariant(pg)); - - if (crt >= CRT_NR) - return -EINVAL; - - list_for_each_entry(scan, &pg->cp_layers, cpl_linkage) { - if (!scan->cpl_ops->io[crt].cpo_cache_add) - continue; - - result = scan->cpl_ops->io[crt].cpo_cache_add(env, scan, io); - if (result != 0) - break; - } - CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result); - return result; -} -EXPORT_SYMBOL(cl_page_cache_add); - -/** * Called if a pge is being written back by kernel's intention. * * \pre cl_page_is_owned(pg, io) @@ -1344,68 +953,21 @@ EXPORT_SYMBOL(cl_page_flush); * \see cl_page_operations::cpo_is_under_lock() */ int cl_page_is_under_lock(const struct lu_env *env, struct cl_io *io, - struct cl_page *page) + struct cl_page *page, pgoff_t *max_index) { int rc; PINVRNT(env, page, cl_page_invariant(page)); - rc = CL_PAGE_INVOKE(env, page, CL_PAGE_OP(cpo_is_under_lock), - (const struct lu_env *, - const struct cl_page_slice *, struct cl_io *), - io); - PASSERT(env, page, rc != 0); + rc = CL_PAGE_INVOKE_REVERSE(env, page, CL_PAGE_OP(cpo_is_under_lock), + (const struct lu_env *, + const struct cl_page_slice *, + struct cl_io *, pgoff_t *), + io, max_index); return rc; } EXPORT_SYMBOL(cl_page_is_under_lock); -static int page_prune_cb(const struct lu_env *env, struct cl_io *io, - struct cl_page *page, void *cbdata) -{ - cl_page_own(env, io, page); - cl_page_unmap(env, io, page); - cl_page_discard(env, io, page); - cl_page_disown(env, io, page); - return CLP_GANG_OKAY; -} - -/** - * Purges all cached pages belonging to the object \a obj. - */ -int cl_pages_prune(const struct lu_env *env, struct cl_object *clobj) -{ - struct cl_thread_info *info; - struct cl_object *obj = cl_object_top(clobj); - struct cl_io *io; - int result; - - info = cl_env_info(env); - io = &info->clt_io; - - /* - * initialize the io. This is ugly since we never do IO in this - * function, we just make cl_page_list functions happy. -jay - */ - io->ci_obj = obj; - io->ci_ignore_layout = 1; - result = cl_io_init(env, io, CIT_MISC, obj); - if (result != 0) { - cl_io_fini(env, io); - return io->ci_result; - } - - do { - result = cl_page_gang_lookup(env, obj, io, 0, CL_PAGE_EOF, - page_prune_cb, NULL); - if (result == CLP_GANG_RESCHED) - cond_resched(); - } while (result != CLP_GANG_OKAY); - - cl_io_fini(env, io); - return result; -} -EXPORT_SYMBOL(cl_pages_prune); - /** * Tells transfer engine that only part of a page is to be transmitted. * @@ -1431,9 +993,8 @@ void cl_page_header_print(const struct lu_env *env, void *cookie, lu_printer_t printer, const struct cl_page *pg) { (*printer)(env, cookie, - "page@%p[%d %p:%lu ^%p_%p %d %d %d %p %p %#x]\n", + "page@%p[%d %p %d %d %d %p %p %#x]\n", pg, atomic_read(&pg->cp_ref), pg->cp_obj, - pg->cp_index, pg->cp_parent, pg->cp_child, pg->cp_state, pg->cp_error, pg->cp_type, pg->cp_owner, pg->cp_req, pg->cp_flags); } @@ -1445,11 +1006,7 @@ EXPORT_SYMBOL(cl_page_header_print); void cl_page_print(const struct lu_env *env, void *cookie, lu_printer_t printer, const struct cl_page *pg) { - struct cl_page *scan; - - for (scan = cl_page_top((struct cl_page *)pg); scan; - scan = scan->cp_child) - cl_page_header_print(env, cookie, printer, scan); + cl_page_header_print(env, cookie, printer, pg); CL_PAGE_INVOKE(env, (struct cl_page *)pg, CL_PAGE_OP(cpo_print), (const struct lu_env *env, const struct cl_page_slice *slice, @@ -1509,21 +1066,13 @@ EXPORT_SYMBOL(cl_page_size); * \see cl_lock_slice_add(), cl_req_slice_add(), cl_io_slice_add() */ void cl_page_slice_add(struct cl_page *page, struct cl_page_slice *slice, - struct cl_object *obj, + struct cl_object *obj, pgoff_t index, const struct cl_page_operations *ops) { list_add_tail(&slice->cpl_linkage, &page->cp_layers); slice->cpl_obj = obj; + slice->cpl_index = index; slice->cpl_ops = ops; slice->cpl_page = page; } EXPORT_SYMBOL(cl_page_slice_add); - -int cl_page_init(void) -{ - return 0; -} - -void cl_page_fini(void) -{ -} diff --git a/drivers/staging/lustre/lustre/obdclass/class_obd.c b/drivers/staging/lustre/lustre/obdclass/class_obd.c index c2cf015962dd..f48816af8be7 100644 --- a/drivers/staging/lustre/lustre/obdclass/class_obd.c +++ b/drivers/staging/lustre/lustre/obdclass/class_obd.c @@ -60,6 +60,8 @@ unsigned int obd_dump_on_eviction; EXPORT_SYMBOL(obd_dump_on_eviction); unsigned int obd_max_dirty_pages = 256; EXPORT_SYMBOL(obd_max_dirty_pages); +atomic_t obd_unstable_pages; +EXPORT_SYMBOL(obd_unstable_pages); atomic_t obd_dirty_pages; EXPORT_SYMBOL(obd_dirty_pages); unsigned int obd_timeout = OBD_TIMEOUT_DEFAULT; /* seconds */ @@ -335,7 +337,6 @@ int class_handle_ioctl(unsigned int cmd, unsigned long arg) err = 0; goto out; } - } if (data->ioc_dev == OBD_DEV_BY_DEVNAME) { @@ -461,7 +462,7 @@ static int obd_init_checks(void) CWARN("LPD64 wrong length! strlen(%s)=%d != 2\n", buf, len); ret = -EINVAL; } - if ((u64val & ~CFS_PAGE_MASK) >= PAGE_SIZE) { + if ((u64val & ~PAGE_MASK) >= PAGE_SIZE) { CWARN("mask failed: u64val %llu >= %llu\n", u64val, (__u64)PAGE_SIZE); ret = -EINVAL; diff --git a/drivers/staging/lustre/lustre/obdclass/debug.c b/drivers/staging/lustre/lustre/obdclass/debug.c index 43a7f7a79b35..e4edfb2c0a20 100644 --- a/drivers/staging/lustre/lustre/obdclass/debug.c +++ b/drivers/staging/lustre/lustre/obdclass/debug.c @@ -68,8 +68,8 @@ int block_debug_check(char *who, void *addr, int end, __u64 off, __u64 id) LASSERT(addr); - ne_off = le64_to_cpu (off); - id = le64_to_cpu (id); + ne_off = le64_to_cpu(off); + id = le64_to_cpu(id); if (memcmp(addr, (char *)&ne_off, LPDS)) { CDEBUG(D_ERROR, "%s: id %#llx offset %llu off: %#llx != %#llx\n", who, id, off, *(__u64 *)addr, ne_off); diff --git a/drivers/staging/lustre/lustre/obdclass/genops.c b/drivers/staging/lustre/lustre/obdclass/genops.c index cf97b8f06764..d95f11d62a32 100644 --- a/drivers/staging/lustre/lustre/obdclass/genops.c +++ b/drivers/staging/lustre/lustre/obdclass/genops.c @@ -604,7 +604,6 @@ int obd_init_caches(void) out: obd_cleanup_caches(); return -ENOMEM; - } /* map connection to client */ diff --git a/drivers/staging/lustre/lustre/obdclass/linux/linux-module.c b/drivers/staging/lustre/lustre/obdclass/linux/linux-module.c index 8eddf206f1ed..2cd4522462d9 100644 --- a/drivers/staging/lustre/lustre/obdclass/linux/linux-module.c +++ b/drivers/staging/lustre/lustre/obdclass/linux/linux-module.c @@ -158,9 +158,7 @@ int obd_ioctl_popdata(void __user *arg, void *data, int len) { int err; - err = copy_to_user(arg, data, len); - if (err) - err = -EFAULT; + err = copy_to_user(arg, data, len) ? -EFAULT : 0; return err; } EXPORT_SYMBOL(obd_ioctl_popdata); diff --git a/drivers/staging/lustre/lustre/obdclass/llog.c b/drivers/staging/lustre/lustre/obdclass/llog.c index 992573eae1b1..79194d8cb587 100644 --- a/drivers/staging/lustre/lustre/obdclass/llog.c +++ b/drivers/staging/lustre/lustre/obdclass/llog.c @@ -265,7 +265,6 @@ repeat: for (rec = (struct llog_rec_hdr *)buf; (char *)rec < buf + LLOG_CHUNK_SIZE; rec = (struct llog_rec_hdr *)((char *)rec + rec->lrh_len)) { - CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n", rec, rec->lrh_type); diff --git a/drivers/staging/lustre/lustre/obdclass/lprocfs_status.c b/drivers/staging/lustre/lustre/obdclass/lprocfs_status.c index d93f42fee420..5a1eae1de2ec 100644 --- a/drivers/staging/lustre/lustre/obdclass/lprocfs_status.c +++ b/drivers/staging/lustre/lustre/obdclass/lprocfs_status.c @@ -49,7 +49,7 @@ static const char * const obd_connect_names[] = { "read_only", "lov_index", - "unused", + "connect_from_mds", "write_grant", "server_lock", "version", @@ -122,6 +122,56 @@ int obd_connect_flags2str(char *page, int count, __u64 flags, char *sep) } EXPORT_SYMBOL(obd_connect_flags2str); +static void obd_connect_data_seqprint(struct seq_file *m, + struct obd_connect_data *ocd) +{ + int flags; + + LASSERT(ocd); + flags = ocd->ocd_connect_flags; + + seq_printf(m, " connect_data:\n" + " flags: %llx\n" + " instance: %u\n", + ocd->ocd_connect_flags, + ocd->ocd_instance); + if (flags & OBD_CONNECT_VERSION) + seq_printf(m, " target_version: %u.%u.%u.%u\n", + OBD_OCD_VERSION_MAJOR(ocd->ocd_version), + OBD_OCD_VERSION_MINOR(ocd->ocd_version), + OBD_OCD_VERSION_PATCH(ocd->ocd_version), + OBD_OCD_VERSION_FIX(ocd->ocd_version)); + if (flags & OBD_CONNECT_MDS) + seq_printf(m, " mdt_index: %d\n", ocd->ocd_group); + if (flags & OBD_CONNECT_GRANT) + seq_printf(m, " initial_grant: %d\n", ocd->ocd_grant); + if (flags & OBD_CONNECT_INDEX) + seq_printf(m, " target_index: %u\n", ocd->ocd_index); + if (flags & OBD_CONNECT_BRW_SIZE) + seq_printf(m, " max_brw_size: %d\n", ocd->ocd_brw_size); + if (flags & OBD_CONNECT_IBITS) + seq_printf(m, " ibits_known: %llx\n", + ocd->ocd_ibits_known); + if (flags & OBD_CONNECT_GRANT_PARAM) + seq_printf(m, " grant_block_size: %d\n" + " grant_inode_size: %d\n" + " grant_extent_overhead: %d\n", + ocd->ocd_blocksize, + ocd->ocd_inodespace, + ocd->ocd_grant_extent); + if (flags & OBD_CONNECT_TRANSNO) + seq_printf(m, " first_transno: %llx\n", + ocd->ocd_transno); + if (flags & OBD_CONNECT_CKSUM) + seq_printf(m, " cksum_types: %#x\n", + ocd->ocd_cksum_types); + if (flags & OBD_CONNECT_MAX_EASIZE) + seq_printf(m, " max_easize: %d\n", ocd->ocd_max_easize); + if (flags & OBD_CONNECT_MAXBYTES) + seq_printf(m, " max_object_bytes: %llx\n", + ocd->ocd_maxbytes); +} + int lprocfs_read_frac_helper(char *buffer, unsigned long count, long val, int mult) { @@ -624,6 +674,7 @@ int lprocfs_rd_import(struct seq_file *m, void *data) struct obd_device *obd = data; struct obd_import *imp; struct obd_import_conn *conn; + struct obd_connect_data *ocd; int j; int k; int rw = 0; @@ -635,9 +686,9 @@ int lprocfs_rd_import(struct seq_file *m, void *data) return rc; imp = obd->u.cli.cl_import; + ocd = &imp->imp_connect_data; - seq_printf(m, - "import:\n" + seq_printf(m, "import:\n" " name: %s\n" " target: %s\n" " state: %s\n" @@ -649,9 +700,9 @@ int lprocfs_rd_import(struct seq_file *m, void *data) imp->imp_connect_data.ocd_instance); obd_connect_seq_flags2str(m, imp->imp_connect_data.ocd_connect_flags, ", "); - seq_printf(m, - " ]\n" - " import_flags: [ "); + seq_printf(m, " ]\n"); + obd_connect_data_seqprint(m, ocd); + seq_printf(m, " import_flags: [ "); obd_import_flags2str(imp, m); seq_printf(m, @@ -694,8 +745,9 @@ int lprocfs_rd_import(struct seq_file *m, void *data) do_div(sum, ret.lc_count); ret.lc_sum = sum; - } else + } else { ret.lc_sum = 0; + } seq_printf(m, " rpcs:\n" " inflight: %u\n" @@ -1471,10 +1523,10 @@ EXPORT_SYMBOL(lprocfs_oh_tally); void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value) { - unsigned int val; + unsigned int val = 0; - for (val = 0; ((1 << val) < value) && (val <= OBD_HIST_MAX); val++) - ; + if (likely(value != 0)) + val = min(fls(value - 1), OBD_HIST_MAX); lprocfs_oh_tally(oh, val); } diff --git a/drivers/staging/lustre/lustre/obdclass/lu_object.c b/drivers/staging/lustre/lustre/obdclass/lu_object.c index 978568ada8e9..e04385760f21 100644 --- a/drivers/staging/lustre/lustre/obdclass/lu_object.c +++ b/drivers/staging/lustre/lustre/obdclass/lu_object.c @@ -55,6 +55,7 @@ #include "../include/lustre_disk.h" #include "../include/lustre_fid.h" #include "../include/lu_object.h" +#include "../include/cl_object.h" #include "../include/lu_ref.h" #include <linux/list.h> @@ -103,7 +104,6 @@ void lu_object_put(const struct lu_env *env, struct lu_object *o) if (!cfs_hash_bd_dec_and_lock(site->ls_obj_hash, &bd, &top->loh_ref)) { if (lu_object_is_dying(top)) { - /* * somebody may be waiting for this, currently only * used for cl_object, see cl_object_put_last(). @@ -357,7 +357,6 @@ int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr) if (count > 0 && --count == 0) break; - } cfs_hash_bd_unlock(s->ls_obj_hash, &bd, 1); cond_resched(); @@ -715,8 +714,9 @@ struct lu_object *lu_object_find_slice(const struct lu_env *env, obj = lu_object_locate(top->lo_header, dev->ld_type); if (!obj) lu_object_put(env, top); - } else + } else { obj = top; + } return obj; } EXPORT_SYMBOL(lu_object_find_slice); @@ -935,7 +935,7 @@ static void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d) * Initialize site \a s, with \a d as the top level device. */ #define LU_SITE_BITS_MIN 12 -#define LU_SITE_BITS_MAX 24 +#define LU_SITE_BITS_MAX 19 /** * total 256 buckets, we don't want too many buckets because: * - consume too much memory @@ -1468,6 +1468,7 @@ void lu_context_key_quiesce(struct lu_context_key *key) /* * XXX layering violation. */ + cl_env_cache_purge(~0); key->lct_tags |= LCT_QUIESCENT; /* * XXX memory barrier has to go here. diff --git a/drivers/staging/lustre/lustre/obdclass/lustre_peer.c b/drivers/staging/lustre/lustre/obdclass/lustre_peer.c index 5f812460b3ea..b1abe023bb35 100644 --- a/drivers/staging/lustre/lustre/obdclass/lustre_peer.c +++ b/drivers/staging/lustre/lustre/obdclass/lustre_peer.c @@ -163,8 +163,9 @@ int class_del_uuid(const char *uuid) break; } } - } else + } else { list_splice_init(&g_uuid_list, &deathrow); + } spin_unlock(&g_uuid_lock); if (uuid && list_empty(&deathrow)) { diff --git a/drivers/staging/lustre/lustre/obdclass/obd_config.c b/drivers/staging/lustre/lustre/obdclass/obd_config.c index 5395e994deab..cb1d65c3d95d 100644 --- a/drivers/staging/lustre/lustre/obdclass/obd_config.c +++ b/drivers/staging/lustre/lustre/obdclass/obd_config.c @@ -606,7 +606,7 @@ static int class_del_conn(struct obd_device *obd, struct lustre_cfg *lcfg) return rc; } -LIST_HEAD(lustre_profile_list); +static LIST_HEAD(lustre_profile_list); struct lustre_profile *class_get_profile(const char *prof) { @@ -961,7 +961,6 @@ int class_process_config(struct lustre_cfg *lcfg) default: { err = obd_process_config(obd, sizeof(*lcfg), lcfg); goto out; - } } out: @@ -1001,7 +1000,13 @@ int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars, for (i = 1; i < lcfg->lcfg_bufcount; i++) { key = lustre_cfg_buf(lcfg, i); /* Strip off prefix */ - class_match_param(key, prefix, &key); + if (class_match_param(key, prefix, &key)) { + /* + * If the prefix doesn't match, return error so we + * can pass it down the stack + */ + return -ENOSYS; + } sval = strchr(key, '='); if (!sval || (*(sval + 1) == 0)) { CERROR("Can't parse param %s (missing '=')\n", key); @@ -1034,18 +1039,14 @@ int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars, j++; } if (!matched) { - /* If the prefix doesn't match, return error so we - * can pass it down the stack - */ - if (strnchr(key, keylen, '.')) - return -ENOSYS; - CERROR("%s: unknown param %s\n", + CERROR("%.*s: %s unknown param %s\n", + (int)strlen(prefix) - 1, prefix, (char *)lustre_cfg_string(lcfg, 0), key); /* rc = -EINVAL; continue parsing other params */ skip++; } else if (rc < 0) { - CERROR("writing proc entry %s err %d\n", - var->name, rc); + CERROR("%s: error writing proc entry '%s': rc = %d\n", + prefix, var->name, rc); rc = 0; } else { CDEBUG(D_CONFIG, "%s.%.*s: Set parameter %.*s=%s\n", @@ -1350,6 +1351,7 @@ static int class_config_parse_rec(struct llog_rec_hdr *rec, char *buf, lustre_cfg_string(lcfg, i)); } } + ptr += snprintf(ptr, end - ptr, "\n"); /* return consumed bytes */ rc = ptr - buf; return rc; @@ -1368,7 +1370,7 @@ int class_config_dump_handler(const struct lu_env *env, if (rec->lrh_type == OBD_CFG_REC) { class_config_parse_rec(rec, outstr, 256); - LCONSOLE(D_WARNING, " %s\n", outstr); + LCONSOLE(D_WARNING, " %s", outstr); } else { LCONSOLE(D_WARNING, "unhandled lrh_type: %#x\n", rec->lrh_type); rc = -EINVAL; diff --git a/drivers/staging/lustre/lustre/obdclass/obd_mount.c b/drivers/staging/lustre/lustre/obdclass/obd_mount.c index d3e28a389ac1..e0c90adc72a7 100644 --- a/drivers/staging/lustre/lustre/obdclass/obd_mount.c +++ b/drivers/staging/lustre/lustre/obdclass/obd_mount.c @@ -102,7 +102,7 @@ int lustre_process_log(struct super_block *sb, char *logname, LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s' failed from the MGS (%d). Make sure this client and the MGS are running compatible versions of Lustre.\n", mgc->obd_name, logname, rc); - if (rc) + else if (rc) LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' failed (%d). This may be the result of communication errors between this node and the MGS, a bad configuration, or other errors. See the syslog for more information.\n", mgc->obd_name, logname, rc); @@ -307,7 +307,8 @@ int lustre_start_mgc(struct super_block *sb) while (class_parse_nid(ptr, &nid, &ptr) == 0) { rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID, niduuid, NULL, NULL, NULL); - i++; + if (!rc) + i++; /* Stop at the first failover nid */ if (*ptr == ':') break; @@ -345,16 +346,18 @@ int lustre_start_mgc(struct super_block *sb) sprintf(niduuid, "%s_%x", mgcname, i); j = 0; while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) { - j++; - rc = do_lcfg(mgcname, nid, - LCFG_ADD_UUID, niduuid, NULL, NULL, NULL); + rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID, niduuid, + NULL, NULL, NULL); + if (!rc) + ++j; if (*ptr == ':') break; } if (j > 0) { rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN, niduuid, NULL, NULL, NULL); - i++; + if (!rc) + i++; } else { /* at ":/fsname" */ break; diff --git a/drivers/staging/lustre/lustre/obdclass/obdo.c b/drivers/staging/lustre/lustre/obdclass/obdo.c index e6436cb4ac62..748e33f017d5 100644 --- a/drivers/staging/lustre/lustre/obdclass/obdo.c +++ b/drivers/staging/lustre/lustre/obdclass/obdo.c @@ -185,8 +185,7 @@ void md_from_obdo(struct md_op_data *op_data, struct obdo *oa, u32 valid) op_data->op_attr.ia_valid |= ATTR_BLOCKS; } if (valid & OBD_MD_FLFLAGS) { - ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = - oa->o_flags; + op_data->op_attr_flags = oa->o_flags; op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG; } } |