1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2001-2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 #define DEBUG_SUBSYSTEM S_ECHO
24 #include <linux/version.h>
25 #include <linux/module.h>
27 #include <linux/completion.h>
28 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
29 #include <linux/iobuf.h>
31 #include <asm/div64.h>
32 #include <linux/smp_lock.h>
34 #include <liblustre.h>
37 #include <linux/obd.h>
38 #include <linux/obd_support.h>
39 #include <linux/obd_class.h>
40 #include <linux/obd_echo.h>
41 #include <linux/lustre_debug.h>
42 #include <linux/lprocfs_status.h>
44 static obd_id last_object_id;
48 echo_printk_object (char *msg, struct ec_object *eco)
50 struct lov_stripe_md *lsm = eco->eco_lsm;
53 printk (KERN_INFO "Lustre: %s: object %p: "LPX64", refs %d%s: "LPX64
54 "=%u!%u\n", msg, eco, eco->eco_id, eco->eco_refcount,
55 eco->eco_deleted ? "(deleted) " : "",
56 lsm->lsm_object_id, lsm->lsm_stripe_size,
57 lsm->lsm_stripe_count);
59 for (i = 0; i < lsm->lsm_stripe_count; i++)
60 printk (KERN_INFO "Lustre: @%2u:"LPX64"\n",
61 lsm->lsm_oinfo[i].loi_ost_idx,
62 lsm->lsm_oinfo[i].loi_id);
66 static struct ec_object *
67 echo_find_object_locked (struct obd_device *obd, obd_id id)
69 struct echo_client_obd *ec = &obd->u.echocli;
70 struct ec_object *eco = NULL;
73 list_for_each (el, &ec->ec_objects) {
74 eco = list_entry (el, struct ec_object, eco_obj_chain);
76 if (eco->eco_id == id)
83 echo_copyout_lsm (struct lov_stripe_md *lsm, void *ulsm, int ulsm_nob)
87 nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
91 if (copy_to_user (ulsm, lsm, nob))
98 echo_copyin_lsm (struct obd_device *obd, struct lov_stripe_md *lsm,
99 void *ulsm, int ulsm_nob)
101 struct echo_client_obd *ec = &obd->u.echocli;
104 if (ulsm_nob < sizeof (*lsm))
107 if (copy_from_user (lsm, ulsm, sizeof (*lsm)))
110 nob = lsm->lsm_stripe_count * sizeof (lsm->lsm_oinfo[0]);
112 if (ulsm_nob < nob ||
113 lsm->lsm_stripe_count > ec->ec_nstripes ||
114 lsm->lsm_magic != LOV_MAGIC ||
115 (lsm->lsm_stripe_size & (PAGE_SIZE - 1)) != 0 ||
116 ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
119 if (copy_from_user(lsm->lsm_oinfo,
120 ((struct lov_stripe_md *)ulsm)->lsm_oinfo, nob))
126 static struct ec_object *
127 echo_allocate_object (struct obd_device *obd)
129 struct echo_client_obd *ec = &obd->u.echocli;
130 struct ec_object *eco;
133 OBD_ALLOC(eco, sizeof (*eco));
137 rc = obd_alloc_memmd(ec->ec_exp, &eco->eco_lsm);
139 OBD_FREE(eco, sizeof (*eco));
143 eco->eco_device = obd;
144 eco->eco_deleted = 0;
145 eco->eco_refcount = 0;
146 eco->eco_lsm->lsm_magic = LOV_MAGIC;
147 /* leave stripe count 0 by default */
153 echo_free_object (struct ec_object *eco)
155 struct obd_device *obd = eco->eco_device;
156 struct echo_client_obd *ec = &obd->u.echocli;
158 LASSERT (eco->eco_refcount == 0);
159 obd_free_memmd(ec->ec_exp, &eco->eco_lsm);
160 OBD_FREE (eco, sizeof (*eco));
163 static int echo_create_object(struct obd_device *obd, int on_target,
164 struct obdo *oa, void *ulsm, int ulsm_nob,
165 struct obd_trans_info *oti)
167 struct echo_client_obd *ec = &obd->u.echocli;
168 struct ec_object *eco2;
169 struct ec_object *eco;
170 struct lov_stripe_md *lsm;
174 if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
175 (on_target || /* set_stripe */
176 ec->ec_nstripes != 0)) { /* LOV */
177 CERROR ("No valid oid\n");
182 eco = echo_allocate_object (obd);
188 rc = echo_copyin_lsm (obd, lsm, ulsm, ulsm_nob);
192 /* setup object ID here for !on_target and LOV hint */
193 if ((oa->o_valid & OBD_MD_FLID) != 0)
194 eco->eco_id = lsm->lsm_object_id = oa->o_id;
196 if (lsm->lsm_stripe_count == 0)
197 lsm->lsm_stripe_count = ec->ec_nstripes;
199 if (lsm->lsm_stripe_size == 0)
200 lsm->lsm_stripe_size = PAGE_SIZE;
202 idx = ll_insecure_random_int();
204 /* setup stripes: indices + default ids if required */
205 for (i = 0; i < lsm->lsm_stripe_count; i++) {
206 if (lsm->lsm_oinfo[i].loi_id == 0)
207 lsm->lsm_oinfo[i].loi_id = lsm->lsm_object_id;
209 lsm->lsm_oinfo[i].loi_ost_idx =
210 (idx + i) % ec->ec_nstripes;
211 lsm->lsm_oinfo[i].loi_ost_gen = 1;
214 OBD_ALLOC(eco, sizeof(*eco));
215 eco->eco_device = obd;
220 oa->o_id = ++last_object_id;
223 oa->o_gr = FILTER_GROUP_ECHO;
224 oa->o_valid |= OBD_MD_FLGROUP;
225 rc = obd_create(ec->ec_exp, oa, NULL, 0, &lsm, oti);
229 /* See what object ID we were given */
230 eco->eco_id = oa->o_id = lsm->lsm_object_id;
231 oa->o_valid |= OBD_MD_FLID;
233 LASSERT(eco->eco_lsm == NULL || eco->eco_lsm == lsm);
237 spin_lock (&ec->ec_lock);
239 eco2 = echo_find_object_locked (obd, oa->o_id);
240 if (eco2 != NULL) { /* conflict */
241 spin_unlock (&ec->ec_lock);
243 CERROR ("Can't create object id "LPX64": id already exists%s\n",
244 oa->o_id, on_target ? " (undoing create)" : "");
247 obd_destroy(ec->ec_exp, oa, lsm, oti);
253 list_add (&eco->eco_obj_chain, &ec->ec_objects);
254 spin_unlock (&ec->ec_lock);
256 "created %p: "LPX64"=%u#%u@%u refs %d del %d\n",
258 eco->eco_lsm->lsm_stripe_size,
259 eco->eco_lsm->lsm_stripe_count,
260 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
261 eco->eco_refcount, eco->eco_deleted);
265 echo_free_object (eco);
270 echo_get_object (struct ec_object **ecop, struct obd_device *obd,
273 struct echo_client_obd *ec = &obd->u.echocli;
274 struct ec_object *eco;
275 struct ec_object *eco2;
278 if ((oa->o_valid & OBD_MD_FLID) == 0 ||
279 oa->o_id == 0) /* disallow use of object id 0 */
281 CERROR ("No valid oid\n");
285 spin_lock (&ec->ec_lock);
286 eco = echo_find_object_locked (obd, oa->o_id);
288 if (eco->eco_deleted) /* being deleted */
289 return (-EAGAIN); /* (see comment in cleanup) */
292 spin_unlock (&ec->ec_lock);
295 "found %p: "LPX64"=%u#%u@%u refs %d del %d\n",
297 eco->eco_lsm->lsm_stripe_size,
298 eco->eco_lsm->lsm_stripe_count,
299 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
300 eco->eco_refcount, eco->eco_deleted);
303 spin_unlock (&ec->ec_lock);
305 if (ec->ec_nstripes != 0) /* striping required */
308 eco = echo_allocate_object (obd);
312 eco->eco_id = eco->eco_lsm->lsm_object_id = oa->o_id;
314 spin_lock (&ec->ec_lock);
316 eco2 = echo_find_object_locked (obd, oa->o_id);
317 if (eco2 == NULL) { /* didn't race */
318 list_add (&eco->eco_obj_chain, &ec->ec_objects);
319 spin_unlock (&ec->ec_lock);
320 eco->eco_refcount = 1;
323 "created %p: "LPX64"=%u#%u@%d refs %d del %d\n",
325 eco->eco_lsm->lsm_stripe_size,
326 eco->eco_lsm->lsm_stripe_count,
327 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
328 eco->eco_refcount, eco->eco_deleted);
332 if (eco2->eco_deleted)
333 rc = -EAGAIN; /* lose race */
335 eco2->eco_refcount++; /* take existing */
338 LASSERT (eco2->eco_id == eco2->eco_lsm->lsm_object_id);
340 "found(2) %p: "LPX64"=%u#%u@%d refs %d del %d\n",
342 eco2->eco_lsm->lsm_stripe_size,
343 eco2->eco_lsm->lsm_stripe_count,
344 eco2->eco_lsm->lsm_oinfo[0].loi_ost_idx,
345 eco2->eco_refcount, eco2->eco_deleted);
348 spin_unlock (&ec->ec_lock);
350 echo_free_object (eco);
355 echo_put_object (struct ec_object *eco)
357 struct obd_device *obd = eco->eco_device;
358 struct echo_client_obd *ec = &obd->u.echocli;
360 /* Release caller's ref on the object.
361 * delete => mark for deletion when last ref goes
364 spin_lock (&ec->ec_lock);
367 LASSERT (eco->eco_refcount >= 0);
369 CDEBUG(D_INFO, "put %p: "LPX64"=%u#%u@%d refs %d del %d\n",
371 eco->eco_lsm->lsm_stripe_size,
372 eco->eco_lsm->lsm_stripe_count,
373 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
374 eco->eco_refcount, eco->eco_deleted);
376 if (eco->eco_refcount != 0 || !eco->eco_deleted) {
377 spin_unlock (&ec->ec_lock);
381 spin_unlock (&ec->ec_lock);
383 /* NB leave obj in the object list. We must prevent anyone from
384 * attempting to enqueue on this object number until we can be
385 * sure there will be no more lock callbacks.
387 obd_cancel_unused(ec->ec_exp, eco->eco_lsm, 0, NULL);
389 /* now we can let it go */
390 spin_lock (&ec->ec_lock);
391 list_del (&eco->eco_obj_chain);
392 spin_unlock (&ec->ec_lock);
394 LASSERT (eco->eco_refcount == 0);
396 echo_free_object (eco);
400 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
402 unsigned long stripe_count;
403 unsigned long stripe_size;
405 unsigned long woffset;
409 if (lsm->lsm_stripe_count <= 1)
413 stripe_size = lsm->lsm_stripe_size;
414 stripe_count = lsm->lsm_stripe_count;
416 /* width = # bytes in all stripes */
417 width = stripe_size * stripe_count;
419 /* woffset = offset within a width; offset = whole number of widths */
420 woffset = do_div (offset, width);
422 stripe_index = woffset / stripe_size;
424 *idp = lsm->lsm_oinfo[stripe_index].loi_id;
425 *offp = offset * stripe_size + woffset % stripe_size;
429 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
430 struct page *page, int rw, obd_id id,
431 obd_off offset, obd_off count)
438 /* no partial pages on the client */
439 LASSERT(count == PAGE_SIZE);
443 for (delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
444 if (rw == OBD_BRW_WRITE) {
445 stripe_off = offset + delta;
447 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
449 stripe_off = 0xdeadbeef00c0ffeeULL;
450 stripe_id = 0xdeadbeef00c0ffeeULL;
452 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
453 stripe_off, stripe_id);
460 echo_client_page_debug_check(struct lov_stripe_md *lsm,
461 struct page *page, obd_id id,
462 obd_off offset, obd_off count)
471 /* no partial pages on the client */
472 LASSERT(count == PAGE_SIZE);
476 for (rc = delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
477 stripe_off = offset + delta;
479 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
481 rc2 = block_debug_check("test_brw",
482 addr + delta, OBD_ECHO_BLOCK_SIZE,
483 stripe_off, stripe_id);
485 CERROR ("Error in echo object "LPX64"\n", id);
494 static int echo_client_kbrw(struct obd_device *obd, int rw, struct obdo *oa,
495 struct lov_stripe_md *lsm, obd_off offset,
496 obd_size count, struct obd_trans_info *oti)
498 struct echo_client_obd *ec = &obd->u.echocli;
500 struct brw_page *pga;
501 struct brw_page *pgp;
508 verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID &&
509 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
510 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
512 gfp_mask = ((oa->o_id & 2) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
514 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
517 (count & (PAGE_SIZE - 1)) != 0 ||
519 lsm->lsm_object_id != oa->o_id))
522 /* XXX think again with misaligned I/O */
523 npages = count >> PAGE_SHIFT;
525 OBD_ALLOC(pga, npages * sizeof(*pga));
529 for (i = 0, pgp = pga, off = offset;
531 i++, pgp++, off += PAGE_SIZE) {
533 LASSERT (pgp->pg == NULL); /* for cleanup */
536 pgp->pg = alloc_pages (gfp_mask, 0);
540 pgp->count = PAGE_SIZE;
541 pgp->disk_offset = pgp->page_offset = off;
545 echo_client_page_debug_setup(lsm, pgp->pg, rw,
546 oa->o_id, off, pgp->count);
549 rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti);
552 if (rc != 0 || rw != OBD_BRW_READ)
555 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
561 vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id,
564 if (vrc != 0 && rc == 0)
567 __free_pages(pgp->pg, 0);
569 OBD_FREE(pga, npages * sizeof(*pga));
574 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
575 static int echo_client_ubrw(struct obd_device *obd, int rw,
576 struct obdo *oa, struct lov_stripe_md *lsm,
577 obd_off offset, obd_size count, char *buffer,
578 struct obd_trans_info *oti)
580 struct echo_client_obd *ec = &obd->u.echocli;
582 struct brw_page *pga;
583 struct brw_page *pgp;
585 struct kiobuf *kiobuf;
589 LASSERT (rw == OBD_BRW_WRITE ||
592 /* NB: for now, only whole pages, page aligned */
595 ((long)buffer & (PAGE_SIZE - 1)) != 0 ||
596 (count & (PAGE_SIZE - 1)) != 0 ||
597 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
600 /* XXX think again with misaligned I/O */
601 npages = count >> PAGE_SHIFT;
603 OBD_ALLOC(pga, npages * sizeof(*pga));
607 rc = alloc_kiovec (1, &kiobuf);
611 rc = map_user_kiobuf ((rw == OBD_BRW_READ) ? READ : WRITE,
612 kiobuf, (unsigned long)buffer, count);
616 LASSERT (kiobuf->offset == 0);
617 LASSERT (kiobuf->nr_pages == npages);
619 for (i = 0, off = offset, pgp = pga;
621 i++, off += PAGE_SIZE, pgp++) {
622 pgp->disk_offset = pgp->page_offset = off;
623 pgp->pg = kiobuf->maplist[i];
624 pgp->count = PAGE_SIZE;
628 rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti);
630 // if (rw == OBD_BRW_READ)
631 // mark_dirty_kiobuf (kiobuf, count);
633 unmap_kiobuf (kiobuf);
635 free_kiovec (1, &kiobuf);
637 OBD_FREE(pga, npages * sizeof(*pga));
641 static int echo_client_ubrw(struct obd_device *obd, int rw,
642 struct obdo *oa, struct lov_stripe_md *lsm,
643 obd_off offset, obd_size count, char *buffer,
644 struct obd_trans_info *oti)
646 #warning "echo_client_ubrw() needs to be ported on 2.6 yet"
653 struct echo_async_state;
655 #define EAP_MAGIC 79277927
656 struct echo_async_page {
658 struct page *eap_page;
661 struct echo_async_state *eap_eas;
662 struct list_head eap_item;
665 #define EAP_FROM_COOKIE(c) \
666 (LASSERT(((struct echo_async_page *)(c))->eap_magic == EAP_MAGIC), \
667 (struct echo_async_page *)(c))
669 struct echo_async_state {
671 obd_off eas_next_offset;
672 obd_off eas_end_offset;
675 wait_queue_head_t eas_waitq;
676 struct list_head eas_avail;
678 struct lov_stripe_md *eas_lsm;
681 static int eas_should_wake(struct echo_async_state *eas)
685 spin_lock_irqsave(&eas->eas_lock, flags);
686 if (eas->eas_rc == 0 && !list_empty(&eas->eas_avail))
688 spin_unlock_irqrestore(&eas->eas_lock, flags);
692 static int ec_ap_make_ready(void *data, int cmd)
694 /* our pages are issued ready */
698 static int ec_ap_refresh_count(void *data, int cmd)
700 /* our pages are issued with a stable count */
704 static void ec_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
706 struct echo_async_page *eap = EAP_FROM_COOKIE(data);
708 memcpy(oa, &eap->eap_eas->eas_oa, sizeof(*oa));
711 static void ec_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
713 struct echo_async_page *eap = EAP_FROM_COOKIE(data);
714 struct echo_async_state *eas;
719 if (cmd == OBD_BRW_READ &&
720 eas->eas_oa.o_id != ECHO_PERSISTENT_OBJID &&
721 (eas->eas_oa.o_valid & OBD_MD_FLFLAGS) != 0 &&
722 (eas->eas_oa.o_flags & OBD_FL_DEBUG_CHECK) != 0)
723 echo_client_page_debug_check(eas->eas_lsm, eap->eap_page,
724 eas->eas_oa.o_id, eap->eap_off,
727 spin_lock_irqsave(&eas->eas_lock, flags);
728 if (rc && !eas->eas_rc)
730 eas->eas_in_flight--;
731 list_add(&eap->eap_item, &eas->eas_avail);
732 wake_up(&eas->eas_waitq);
733 spin_unlock_irqrestore(&eas->eas_lock, flags);
736 static struct obd_async_page_ops ec_async_page_ops = {
737 .ap_make_ready = ec_ap_make_ready,
738 .ap_refresh_count = ec_ap_refresh_count,
739 .ap_fill_obdo = ec_ap_fill_obdo,
740 .ap_completion = ec_ap_completion,
743 static int echo_client_async_page(struct obd_export *exp, int rw,
744 struct obdo *oa, struct lov_stripe_md *lsm,
745 obd_off offset, obd_size count,
749 struct echo_async_page *eap;
750 struct echo_async_state eas;
751 struct list_head *pos, *n;
758 /* oa_id == 0 => speed test (no verification) else...
759 * oa & 1 => use HIGHMEM
761 verify = (oa->o_id != 0);
762 gfp_mask = ((oa->o_id & 1) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
765 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
768 (count & (PAGE_SIZE - 1)) != 0 ||
770 lsm->lsm_object_id != oa->o_id))
773 /* XXX think again with misaligned I/O */
774 npages = batching >> PAGE_SHIFT;
776 memcpy(&eas.eas_oa, oa, sizeof(*oa));
777 eas.eas_next_offset = offset;
778 eas.eas_end_offset = offset + count;
779 spin_lock_init(&eas.eas_lock);
780 init_waitqueue_head(&eas.eas_waitq);
781 eas.eas_in_flight = 0;
784 INIT_LIST_HEAD(&eas.eas_avail);
786 /* prepare the group of pages that we're going to be keeping
788 for (i = 0; i < npages; i++) {
789 struct page *page = alloc_page(GFP_KERNEL);
791 GOTO(out, rc = -ENOMEM);
794 list_add_tail(&PAGE_LIST(page), &pages);
796 OBD_ALLOC(eap, sizeof(*eap));
798 GOTO(out, rc = -ENOMEM);
800 eap->eap_magic = EAP_MAGIC;
801 eap->eap_page = page;
803 page->private = (unsigned long)eap;
804 list_add_tail(&eap->eap_item, &eas.eas_avail);
807 /* first we spin queueing io and being woken by its completion */
808 spin_lock_irqsave(&eas.eas_lock, flags);
812 /* sleep until we have a page to send */
813 spin_unlock_irqrestore(&eas.eas_lock, flags);
814 rc = wait_event_interruptible(eas.eas_waitq,
815 eas_should_wake(&eas));
816 spin_lock_irqsave(&eas.eas_lock, flags);
817 if (rc && !eas.eas_rc)
821 if (list_empty(&eas.eas_avail))
823 eap = list_entry(eas.eas_avail.next, struct echo_async_page,
825 list_del(&eap->eap_item);
826 spin_unlock_irqrestore(&eas.eas_lock, flags);
828 /* unbind the eap from its old page offset */
829 if (eap->eap_cookie != NULL) {
830 obd_teardown_async_page(exp, lsm, NULL,
832 eap->eap_cookie = NULL;
835 eas.eas_next_offset += PAGE_SIZE;
836 eap->eap_off = eas.eas_next_offset;
838 rc = obd_prep_async_page(exp, lsm, NULL, eap->eap_page,
839 eap->eap_off, &ec_async_page_ops,
840 eap, &eap->eap_cookie);
842 spin_lock_irqsave(&eas.eas_lock, flags);
847 if (oa->o_id != ECHO_PERSISTENT_OBJID &&
848 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
849 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0)
850 echo_client_page_debug_setup(lsm, eap->eap_page, rw,
852 eap->eap_off, PAGE_SIZE);
854 /* always asserts urgent, which isn't quite right */
855 rc = obd_queue_async_io(exp, lsm, NULL, eap->eap_cookie,
857 ASYNC_READY | ASYNC_URGENT |
859 spin_lock_irqsave(&eas.eas_lock, flags);
860 if (rc && !eas.eas_rc) {
865 if (eas.eas_next_offset == eas.eas_end_offset)
869 /* still hold the eas_lock here.. */
871 /* now we just spin waiting for all the rpcs to complete */
872 while(eas.eas_in_flight) {
873 spin_unlock_irqrestore(&eas.eas_lock, flags);
874 wait_event_interruptible(eas.eas_waitq,
875 eas.eas_in_flight == 0);
876 spin_lock_irqsave(&eas.eas_lock, flags);
878 spin_unlock_irqrestore(&eas.eas_lock, flags);
881 list_for_each_safe(pos, n, &pages) {
882 struct page *page = list_entry(pos, struct page,
885 list_del(&PAGE_LIST(page));
886 if (page->private != 0) {
887 eap = (struct echo_async_page *)page->private;
888 if (eap->eap_cookie != NULL)
889 obd_teardown_async_page(exp, lsm, NULL,
891 OBD_FREE(eap, sizeof(*eap));
899 static int echo_client_prep_commit(struct obd_export *exp, int rw,
900 struct obdo *oa, struct lov_stripe_md *lsm,
901 obd_off offset, obd_size count,
902 obd_size batch, struct obd_trans_info *oti)
904 struct obd_ioobj ioo;
905 struct niobuf_local *lnb;
906 struct niobuf_remote *rnb;
908 obd_size npages, tot_pages;
912 if (count <= 0 || (count & (PAGE_SIZE - 1)) != 0 ||
913 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
916 npages = batch >> PAGE_SHIFT;
917 tot_pages = count >> PAGE_SHIFT;
919 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
920 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
922 if (lnb == NULL || rnb == NULL)
923 GOTO(out, ret = -ENOMEM);
925 obdo_to_ioobj(oa, &ioo);
929 for(; tot_pages; tot_pages -= npages) {
930 if (tot_pages < npages)
933 for (i = 0; i < npages; i++, off += PAGE_SIZE) {
935 rnb[i].len = PAGE_SIZE;
938 /* XXX this can't be the best.. */
939 memset(oti, 0, sizeof(*oti));
940 ioo.ioo_bufcnt = npages;
942 ret = obd_preprw(rw, exp, oa, 1, &ioo, npages, rnb, lnb,
947 for (i = 0; i < npages; i++) {
948 struct page *page = lnb[i].page;
951 if (page == NULL && lnb[i].rc == 0)
954 if (oa->o_id == ECHO_PERSISTENT_OBJID ||
955 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
956 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
960 if (rw == OBD_BRW_WRITE)
961 echo_client_page_debug_setup(lsm, page, rw,
966 echo_client_page_debug_check(lsm, page,
972 ret = obd_commitrw(rw, exp, oa, 1, &ioo, npages, lnb, oti, ret);
979 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
981 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
985 int echo_client_brw_ioctl(int rw, struct obd_export *exp,
986 struct obd_ioctl_data *data)
988 struct obd_device *obd = class_exp2obd(exp);
989 struct echo_client_obd *ec = &obd->u.echocli;
990 struct obd_trans_info dummy_oti;
991 struct ec_object *eco;
995 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
999 memset(&dummy_oti, 0, sizeof(dummy_oti));
1001 data->ioc_obdo1.o_valid &= ~OBD_MD_FLHANDLE;
1002 data->ioc_obdo1.o_valid |= OBD_MD_FLGROUP;
1003 data->ioc_obdo1.o_gr = FILTER_GROUP_ECHO;
1005 switch((long)data->ioc_pbuf1) {
1007 if (data->ioc_pbuf2 == NULL) { // NULL user data pointer
1008 rc = echo_client_kbrw(obd, rw, &data->ioc_obdo1,
1009 eco->eco_lsm, data->ioc_offset,
1010 data->ioc_count, &dummy_oti);
1013 rc = echo_client_ubrw(obd, rw, &data->ioc_obdo1,
1014 eco->eco_lsm, data->ioc_offset,
1015 data->ioc_count, data->ioc_pbuf2,
1021 rc = echo_client_async_page(ec->ec_exp, rw, &data->ioc_obdo1,
1022 eco->eco_lsm, data->ioc_offset,
1023 data->ioc_count, data->ioc_plen1);
1026 rc = echo_client_prep_commit(ec->ec_exp, rw, &data->ioc_obdo1,
1027 eco->eco_lsm, data->ioc_offset,
1028 data->ioc_count, data->ioc_plen1,
1034 echo_put_object(eco);
1039 echo_ldlm_callback (struct ldlm_lock *lock, struct ldlm_lock_desc *new,
1040 void *data, int flag)
1042 struct ec_object *eco = (struct ec_object *)data;
1043 struct echo_client_obd *ec = &(eco->eco_device->u.echocli);
1044 struct lustre_handle lockh;
1045 struct list_head *el;
1049 ldlm_lock2handle (lock, &lockh);
1051 /* #ifdef this out if we're not feeling paranoid */
1052 spin_lock (&ec->ec_lock);
1053 list_for_each (el, &ec->ec_objects) {
1054 found = (eco == list_entry(el, struct ec_object,
1059 spin_unlock (&ec->ec_lock);
1063 case LDLM_CB_BLOCKING:
1064 CDEBUG(D_INFO, "blocking callback on "LPX64", handle "LPX64"\n",
1065 eco->eco_id, lockh.cookie);
1066 rc = ldlm_cli_cancel (&lockh);
1068 CERROR ("ldlm_cli_cancel failed: %d\n", rc);
1071 case LDLM_CB_CANCELING:
1072 CDEBUG(D_INFO, "cancel callback on "LPX64", handle "LPX64"\n",
1073 eco->eco_id, lockh.cookie);
1084 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
1085 int mode, obd_off offset, obd_size nob)
1087 struct obd_device *obd = exp->exp_obd;
1088 struct echo_client_obd *ec = &obd->u.echocli;
1089 struct lustre_handle *ulh = obdo_handle (oa);
1090 struct ec_object *eco;
1091 struct ec_lock *ecl;
1095 if (!(mode == LCK_PR || mode == LCK_PW))
1098 if ((offset & (PAGE_SIZE - 1)) != 0 ||
1099 (nob & (PAGE_SIZE - 1)) != 0)
1102 rc = echo_get_object (&eco, obd, oa);
1107 OBD_ALLOC (ecl, sizeof (*ecl));
1111 ecl->ecl_mode = mode;
1112 ecl->ecl_object = eco;
1113 ecl->ecl_policy.l_extent.start = offset;
1114 ecl->ecl_policy.l_extent.end =
1115 (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
1118 rc = obd_enqueue(ec->ec_exp, eco->eco_lsm, LDLM_EXTENT,
1119 &ecl->ecl_policy, mode, &flags, echo_ldlm_callback,
1120 ldlm_completion_ast, NULL, eco, sizeof(struct ost_lvb),
1121 lustre_swab_ost_lvb, &ecl->ecl_lock_handle);
1125 CDEBUG(D_INFO, "enqueue handle "LPX64"\n", ecl->ecl_lock_handle.cookie);
1127 /* NB ecl takes object ref from echo_get_object() above */
1128 spin_lock(&ec->ec_lock);
1130 list_add(&ecl->ecl_exp_chain, &exp->exp_ec_data.eced_locks);
1131 ulh->cookie = ecl->ecl_cookie = ec->ec_unique++;
1133 spin_unlock(&ec->ec_lock);
1135 oa->o_valid |= OBD_MD_FLHANDLE;
1139 OBD_FREE (ecl, sizeof (*ecl));
1141 echo_put_object (eco);
1146 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
1148 struct obd_device *obd = exp->exp_obd;
1149 struct echo_client_obd *ec = &obd->u.echocli;
1150 struct lustre_handle *ulh = obdo_handle (oa);
1151 struct ec_lock *ecl = NULL;
1153 struct list_head *el;
1156 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
1159 spin_lock (&ec->ec_lock);
1161 list_for_each (el, &exp->exp_ec_data.eced_locks) {
1162 ecl = list_entry (el, struct ec_lock, ecl_exp_chain);
1163 found = (ecl->ecl_cookie == ulh->cookie);
1165 list_del (&ecl->ecl_exp_chain);
1170 spin_unlock (&ec->ec_lock);
1175 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm, ecl->ecl_mode,
1176 &ecl->ecl_lock_handle);
1178 echo_put_object (ecl->ecl_object);
1179 OBD_FREE (ecl, sizeof (*ecl));
1185 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp,
1186 int len, void *karg, void *uarg)
1188 struct obd_device *obd;
1189 struct echo_client_obd *ec;
1190 struct ec_object *eco;
1191 struct obd_ioctl_data *data = karg;
1192 struct obd_trans_info dummy_oti;
1193 struct oti_req_ack_lock *ack_lock;
1195 int rw = OBD_BRW_READ;
1202 memset(&dummy_oti, 0, sizeof(dummy_oti));
1205 ec = &obd->u.echocli;
1208 case OBD_IOC_CREATE: /* may create echo object */
1209 if (!capable (CAP_SYS_ADMIN))
1210 GOTO (out, rc = -EPERM);
1212 rc = echo_create_object (obd, 1, &data->ioc_obdo1,
1213 data->ioc_pbuf1, data->ioc_plen1,
1217 case OBD_IOC_DESTROY:
1218 if (!capable (CAP_SYS_ADMIN))
1219 GOTO (out, rc = -EPERM);
1221 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1223 oa = &data->ioc_obdo1;
1224 oa->o_gr = FILTER_GROUP_ECHO;
1225 oa->o_valid |= OBD_MD_FLGROUP;
1226 rc = obd_destroy(ec->ec_exp, oa, eco->eco_lsm,
1229 eco->eco_deleted = 1;
1230 echo_put_object(eco);
1234 case OBD_IOC_GETATTR:
1235 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1237 rc = obd_getattr(ec->ec_exp, &data->ioc_obdo1,
1239 echo_put_object(eco);
1243 case OBD_IOC_SETATTR:
1244 if (!capable (CAP_SYS_ADMIN))
1245 GOTO (out, rc = -EPERM);
1247 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1249 rc = obd_setattr(ec->ec_exp, &data->ioc_obdo1,
1250 eco->eco_lsm, NULL);
1251 echo_put_object(eco);
1255 case OBD_IOC_BRW_WRITE:
1256 if (!capable (CAP_SYS_ADMIN))
1257 GOTO (out, rc = -EPERM);
1261 case OBD_IOC_BRW_READ:
1262 rc = echo_client_brw_ioctl(rw, exp, data);
1265 case ECHO_IOC_GET_STRIPE:
1266 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1268 rc = echo_copyout_lsm(eco->eco_lsm, data->ioc_pbuf1,
1270 echo_put_object(eco);
1274 case ECHO_IOC_SET_STRIPE:
1275 if (!capable (CAP_SYS_ADMIN))
1276 GOTO (out, rc = -EPERM);
1278 if (data->ioc_pbuf1 == NULL) { /* unset */
1279 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1281 eco->eco_deleted = 1;
1282 echo_put_object(eco);
1285 rc = echo_create_object(obd, 0, &data->ioc_obdo1,
1287 data->ioc_plen1, &dummy_oti);
1291 case ECHO_IOC_ENQUEUE:
1292 if (!capable (CAP_SYS_ADMIN))
1293 GOTO (out, rc = -EPERM);
1295 rc = echo_client_enqueue(exp, &data->ioc_obdo1,
1296 data->ioc_conn1, /* lock mode */
1297 data->ioc_offset, data->ioc_count);/*extent*/
1300 case ECHO_IOC_CANCEL:
1301 rc = echo_client_cancel(exp, &data->ioc_obdo1);
1305 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
1306 GOTO (out, rc = -ENOTTY);
1312 /* XXX this should be in a helper also called by target_send_reply */
1313 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
1315 if (!ack_lock->mode)
1317 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
1326 echo_client_setup(struct obd_device *obddev, obd_count len, void *buf)
1328 struct lustre_cfg* lcfg = buf;
1329 struct echo_client_obd *ec = &obddev->u.echocli;
1330 struct obd_device *tgt;
1331 struct lustre_handle conn = {0, };
1332 struct obd_uuid echo_uuid = { "ECHO_UUID" };
1336 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1337 CERROR("requires a TARGET OBD name\n");
1341 tgt = class_name2obd(lustre_cfg_string(lcfg, 1));
1342 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
1343 CERROR("device not attached or not set up (%s)\n",
1344 lustre_cfg_string(lcfg, 1));
1348 spin_lock_init (&ec->ec_lock);
1349 INIT_LIST_HEAD (&ec->ec_objects);
1352 rc = obd_connect(&conn, tgt, &echo_uuid, NULL, FILTER_GROUP_ECHO);
1354 CERROR("fail to connect to device %s\n", lustre_cfg_string(lcfg, 1));
1357 ec->ec_exp = class_conn2export(&conn);
1362 static int echo_client_cleanup(struct obd_device *obddev, int flags)
1364 struct list_head *el;
1365 struct ec_object *eco;
1366 struct echo_client_obd *ec = &obddev->u.echocli;
1370 if (!list_empty(&obddev->obd_exports)) {
1371 CERROR("still has clients!\n");
1375 /* XXX assuming sole access */
1376 while (!list_empty(&ec->ec_objects)) {
1377 el = ec->ec_objects.next;
1378 eco = list_entry(el, struct ec_object, eco_obj_chain);
1380 LASSERT(eco->eco_refcount == 0);
1381 eco->eco_refcount = 1;
1382 eco->eco_deleted = 1;
1383 echo_put_object(eco);
1386 rc = obd_disconnect(ec->ec_exp, 0);
1388 CERROR("fail to disconnect device: %d\n", rc);
1393 static int echo_client_connect(struct lustre_handle *conn,
1394 struct obd_device *src,
1395 struct obd_uuid *cluuid,
1396 struct obd_connect_data *data,
1397 unsigned long flags)
1399 struct obd_export *exp;
1402 rc = class_connect(conn, src, cluuid);
1404 exp = class_conn2export(conn);
1405 INIT_LIST_HEAD(&exp->exp_ec_data.eced_locks);
1406 class_export_put(exp);
1412 static int echo_client_disconnect(struct obd_export *exp,
1413 unsigned long flags)
1415 struct obd_device *obd;
1416 struct echo_client_obd *ec;
1417 struct ec_lock *ecl;
1422 GOTO(out, rc = -EINVAL);
1425 ec = &obd->u.echocli;
1427 /* no more contention on export's lock list */
1428 while (!list_empty (&exp->exp_ec_data.eced_locks)) {
1429 ecl = list_entry (exp->exp_ec_data.eced_locks.next,
1430 struct ec_lock, ecl_exp_chain);
1431 list_del (&ecl->ecl_exp_chain);
1433 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
1434 ecl->ecl_mode, &ecl->ecl_lock_handle);
1436 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
1437 "(%d)\n", ecl->ecl_object->eco_id, rc);
1439 echo_put_object (ecl->ecl_object);
1440 OBD_FREE (ecl, sizeof (*ecl));
1443 rc = class_disconnect(exp, 0);
1449 static struct obd_ops echo_obd_ops = {
1450 .o_owner = THIS_MODULE,
1451 .o_setup = echo_client_setup,
1452 .o_cleanup = echo_client_cleanup,
1453 .o_iocontrol = echo_client_iocontrol,
1454 .o_connect = echo_client_connect,
1455 .o_disconnect = echo_client_disconnect
1458 int echo_client_init(void)
1460 struct lprocfs_static_vars lvars;
1462 lprocfs_init_vars(echo, &lvars);
1463 return class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
1464 OBD_ECHO_CLIENT_DEVICENAME);
1467 void echo_client_exit(void)
1469 class_unregister_type(OBD_ECHO_CLIENT_DEVICENAME);