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;
213 OBD_ALLOC(eco, sizeof(*eco));
214 eco->eco_device = obd;
219 oa->o_id = ++last_object_id;
222 oa->o_gr = FILTER_GROUP_ECHO;
223 oa->o_valid |= OBD_MD_FLGROUP;
224 rc = obd_create(ec->ec_exp, oa, &lsm, oti);
228 /* See what object ID we were given */
229 eco->eco_id = oa->o_id = lsm->lsm_object_id;
230 oa->o_valid |= OBD_MD_FLID;
232 LASSERT(eco->eco_lsm == NULL || eco->eco_lsm == lsm);
236 spin_lock (&ec->ec_lock);
238 eco2 = echo_find_object_locked (obd, oa->o_id);
239 if (eco2 != NULL) { /* conflict */
240 spin_unlock (&ec->ec_lock);
242 CERROR ("Can't create object id "LPX64": id already exists%s\n",
243 oa->o_id, on_target ? " (undoing create)" : "");
246 obd_destroy(ec->ec_exp, oa, lsm, oti);
252 list_add (&eco->eco_obj_chain, &ec->ec_objects);
253 spin_unlock (&ec->ec_lock);
255 "created %p: "LPX64"=%u#%u@%u refs %d del %d\n",
257 eco->eco_lsm->lsm_stripe_size,
258 eco->eco_lsm->lsm_stripe_count,
259 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
260 eco->eco_refcount, eco->eco_deleted);
264 echo_free_object (eco);
269 echo_get_object (struct ec_object **ecop, struct obd_device *obd,
272 struct echo_client_obd *ec = &obd->u.echocli;
273 struct ec_object *eco;
274 struct ec_object *eco2;
277 if ((oa->o_valid & OBD_MD_FLID) == 0 ||
278 oa->o_id == 0) /* disallow use of object id 0 */
280 CERROR ("No valid oid\n");
284 spin_lock (&ec->ec_lock);
285 eco = echo_find_object_locked (obd, oa->o_id);
287 if (eco->eco_deleted) /* being deleted */
288 return (-EAGAIN); /* (see comment in cleanup) */
291 spin_unlock (&ec->ec_lock);
294 "found %p: "LPX64"=%u#%u@%u refs %d del %d\n",
296 eco->eco_lsm->lsm_stripe_size,
297 eco->eco_lsm->lsm_stripe_count,
298 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
299 eco->eco_refcount, eco->eco_deleted);
302 spin_unlock (&ec->ec_lock);
304 if (ec->ec_nstripes != 0) /* striping required */
307 eco = echo_allocate_object (obd);
311 eco->eco_id = eco->eco_lsm->lsm_object_id = oa->o_id;
313 spin_lock (&ec->ec_lock);
315 eco2 = echo_find_object_locked (obd, oa->o_id);
316 if (eco2 == NULL) { /* didn't race */
317 list_add (&eco->eco_obj_chain, &ec->ec_objects);
318 spin_unlock (&ec->ec_lock);
319 eco->eco_refcount = 1;
322 "created %p: "LPX64"=%u#%u@%d refs %d del %d\n",
324 eco->eco_lsm->lsm_stripe_size,
325 eco->eco_lsm->lsm_stripe_count,
326 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
327 eco->eco_refcount, eco->eco_deleted);
331 if (eco2->eco_deleted)
332 rc = -EAGAIN; /* lose race */
334 eco2->eco_refcount++; /* take existing */
337 LASSERT (eco2->eco_id == eco2->eco_lsm->lsm_object_id);
339 "found(2) %p: "LPX64"=%u#%u@%d refs %d del %d\n",
341 eco2->eco_lsm->lsm_stripe_size,
342 eco2->eco_lsm->lsm_stripe_count,
343 eco2->eco_lsm->lsm_oinfo[0].loi_ost_idx,
344 eco2->eco_refcount, eco2->eco_deleted);
347 spin_unlock (&ec->ec_lock);
349 echo_free_object (eco);
354 echo_put_object (struct ec_object *eco)
356 struct obd_device *obd = eco->eco_device;
357 struct echo_client_obd *ec = &obd->u.echocli;
359 /* Release caller's ref on the object.
360 * delete => mark for deletion when last ref goes
363 spin_lock (&ec->ec_lock);
366 LASSERT (eco->eco_refcount >= 0);
368 CDEBUG(D_INFO, "put %p: "LPX64"=%u#%u@%d refs %d del %d\n",
370 eco->eco_lsm->lsm_stripe_size,
371 eco->eco_lsm->lsm_stripe_count,
372 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
373 eco->eco_refcount, eco->eco_deleted);
375 if (eco->eco_refcount != 0 || !eco->eco_deleted) {
376 spin_unlock (&ec->ec_lock);
380 spin_unlock (&ec->ec_lock);
382 /* NB leave obj in the object list. We must prevent anyone from
383 * attempting to enqueue on this object number until we can be
384 * sure there will be no more lock callbacks.
386 obd_cancel_unused(ec->ec_exp, eco->eco_lsm, 0, NULL);
388 /* now we can let it go */
389 spin_lock (&ec->ec_lock);
390 list_del (&eco->eco_obj_chain);
391 spin_unlock (&ec->ec_lock);
393 LASSERT (eco->eco_refcount == 0);
395 echo_free_object (eco);
399 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
401 unsigned long stripe_count;
402 unsigned long stripe_size;
404 unsigned long woffset;
408 if (lsm->lsm_stripe_count <= 1)
412 stripe_size = lsm->lsm_stripe_size;
413 stripe_count = lsm->lsm_stripe_count;
415 /* width = # bytes in all stripes */
416 width = stripe_size * stripe_count;
418 /* woffset = offset within a width; offset = whole number of widths */
419 woffset = do_div (offset, width);
421 stripe_index = woffset / stripe_size;
423 *idp = lsm->lsm_oinfo[stripe_index].loi_id;
424 *offp = offset * stripe_size + woffset % stripe_size;
428 echo_client_page_debug_setup(struct lov_stripe_md *lsm,
429 struct page *page, int rw, obd_id id,
430 obd_off offset, obd_off count)
437 /* no partial pages on the client */
438 LASSERT(count == PAGE_SIZE);
442 for (delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
443 if (rw == OBD_BRW_WRITE) {
444 stripe_off = offset + delta;
446 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
448 stripe_off = 0xdeadbeef00c0ffeeULL;
449 stripe_id = 0xdeadbeef00c0ffeeULL;
451 block_debug_setup(addr + delta, OBD_ECHO_BLOCK_SIZE,
452 stripe_off, stripe_id);
459 echo_client_page_debug_check(struct lov_stripe_md *lsm,
460 struct page *page, obd_id id,
461 obd_off offset, obd_off count)
470 /* no partial pages on the client */
471 LASSERT(count == PAGE_SIZE);
475 for (rc = delta = 0; delta < PAGE_SIZE; delta += OBD_ECHO_BLOCK_SIZE) {
476 stripe_off = offset + delta;
478 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
480 rc2 = block_debug_check("test_brw",
481 addr + delta, OBD_ECHO_BLOCK_SIZE,
482 stripe_off, stripe_id);
484 CERROR ("Error in echo object "LPX64"\n", id);
493 static int echo_client_kbrw(struct obd_device *obd, int rw, struct obdo *oa,
494 struct lov_stripe_md *lsm, obd_off offset,
495 obd_size count, struct obd_trans_info *oti)
497 struct echo_client_obd *ec = &obd->u.echocli;
499 struct brw_page *pga;
500 struct brw_page *pgp;
507 verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID &&
508 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
509 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
511 gfp_mask = ((oa->o_id & 2) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
513 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
516 (count & (PAGE_SIZE - 1)) != 0 ||
518 lsm->lsm_object_id != oa->o_id))
521 /* XXX think again with misaligned I/O */
522 npages = count >> PAGE_SHIFT;
524 OBD_ALLOC(pga, npages * sizeof(*pga));
528 for (i = 0, pgp = pga, off = offset;
530 i++, pgp++, off += PAGE_SIZE) {
532 LASSERT (pgp->pg == NULL); /* for cleanup */
535 pgp->pg = alloc_pages (gfp_mask, 0);
539 pgp->count = PAGE_SIZE;
540 pgp->disk_offset = pgp->page_offset = off;
544 echo_client_page_debug_setup(lsm, pgp->pg, rw,
545 oa->o_id, off, pgp->count);
548 rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti);
551 if (rc != 0 || rw != OBD_BRW_READ)
554 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
560 vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id,
563 if (vrc != 0 && rc == 0)
566 __free_pages(pgp->pg, 0);
568 OBD_FREE(pga, npages * sizeof(*pga));
573 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
574 static int echo_client_ubrw(struct obd_device *obd, int rw,
575 struct obdo *oa, struct lov_stripe_md *lsm,
576 obd_off offset, obd_size count, char *buffer,
577 struct obd_trans_info *oti)
579 struct echo_client_obd *ec = &obd->u.echocli;
581 struct brw_page *pga;
582 struct brw_page *pgp;
584 struct kiobuf *kiobuf;
588 LASSERT (rw == OBD_BRW_WRITE ||
591 /* NB: for now, only whole pages, page aligned */
594 ((long)buffer & (PAGE_SIZE - 1)) != 0 ||
595 (count & (PAGE_SIZE - 1)) != 0 ||
596 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
599 /* XXX think again with misaligned I/O */
600 npages = count >> PAGE_SHIFT;
602 OBD_ALLOC(pga, npages * sizeof(*pga));
606 rc = alloc_kiovec (1, &kiobuf);
610 rc = map_user_kiobuf ((rw == OBD_BRW_READ) ? READ : WRITE,
611 kiobuf, (unsigned long)buffer, count);
615 LASSERT (kiobuf->offset == 0);
616 LASSERT (kiobuf->nr_pages == npages);
618 for (i = 0, off = offset, pgp = pga;
620 i++, off += PAGE_SIZE, pgp++) {
621 pgp->disk_offset = pgp->page_offset = off;
622 pgp->pg = kiobuf->maplist[i];
623 pgp->count = PAGE_SIZE;
627 rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti);
629 // if (rw == OBD_BRW_READ)
630 // mark_dirty_kiobuf (kiobuf, count);
632 unmap_kiobuf (kiobuf);
634 free_kiovec (1, &kiobuf);
636 OBD_FREE(pga, npages * sizeof(*pga));
640 static int echo_client_ubrw(struct obd_device *obd, int rw,
641 struct obdo *oa, struct lov_stripe_md *lsm,
642 obd_off offset, obd_size count, char *buffer,
643 struct obd_trans_info *oti)
645 #warning "echo_client_ubrw() needs to be ported on 2.6 yet"
652 struct echo_async_state;
654 #define EAP_MAGIC 79277927
655 struct echo_async_page {
657 struct page *eap_page;
660 struct echo_async_state *eap_eas;
661 struct list_head eap_item;
664 #define EAP_FROM_COOKIE(c) \
665 (LASSERT(((struct echo_async_page *)(c))->eap_magic == EAP_MAGIC), \
666 (struct echo_async_page *)(c))
668 struct echo_async_state {
670 obd_off eas_next_offset;
671 obd_off eas_end_offset;
674 wait_queue_head_t eas_waitq;
675 struct list_head eas_avail;
677 struct lov_stripe_md *eas_lsm;
680 static int eas_should_wake(struct echo_async_state *eas)
684 spin_lock_irqsave(&eas->eas_lock, flags);
685 if (eas->eas_rc == 0 && !list_empty(&eas->eas_avail))
687 spin_unlock_irqrestore(&eas->eas_lock, flags);
691 static int ec_ap_make_ready(void *data, int cmd)
693 /* our pages are issued ready */
697 static int ec_ap_refresh_count(void *data, int cmd)
699 /* our pages are issued with a stable count */
703 static void ec_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
705 struct echo_async_page *eap = EAP_FROM_COOKIE(data);
707 memcpy(oa, &eap->eap_eas->eas_oa, sizeof(*oa));
710 static void ec_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
712 struct echo_async_page *eap = EAP_FROM_COOKIE(data);
713 struct echo_async_state *eas;
718 if (cmd == OBD_BRW_READ &&
719 eas->eas_oa.o_id != ECHO_PERSISTENT_OBJID &&
720 (eas->eas_oa.o_valid & OBD_MD_FLFLAGS) != 0 &&
721 (eas->eas_oa.o_flags & OBD_FL_DEBUG_CHECK) != 0)
722 echo_client_page_debug_check(eas->eas_lsm, eap->eap_page,
723 eas->eas_oa.o_id, eap->eap_off,
726 spin_lock_irqsave(&eas->eas_lock, flags);
727 if (rc && !eas->eas_rc)
729 eas->eas_in_flight--;
730 list_add(&eap->eap_item, &eas->eas_avail);
731 wake_up(&eas->eas_waitq);
732 spin_unlock_irqrestore(&eas->eas_lock, flags);
735 static struct obd_async_page_ops ec_async_page_ops = {
736 .ap_make_ready = ec_ap_make_ready,
737 .ap_refresh_count = ec_ap_refresh_count,
738 .ap_fill_obdo = ec_ap_fill_obdo,
739 .ap_completion = ec_ap_completion,
742 static int echo_client_async_page(struct obd_export *exp, int rw,
743 struct obdo *oa, struct lov_stripe_md *lsm,
744 obd_off offset, obd_size count,
748 struct echo_async_page *eap;
749 struct echo_async_state eas;
750 struct list_head *pos, *n;
757 /* oa_id == 0 => speed test (no verification) else...
758 * oa & 1 => use HIGHMEM
760 verify = (oa->o_id != 0);
761 gfp_mask = ((oa->o_id & 1) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
764 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
767 (count & (PAGE_SIZE - 1)) != 0 ||
769 lsm->lsm_object_id != oa->o_id))
772 /* XXX think again with misaligned I/O */
773 npages = batching >> PAGE_SHIFT;
775 memcpy(&eas.eas_oa, oa, sizeof(*oa));
776 eas.eas_next_offset = offset;
777 eas.eas_end_offset = offset + count;
778 spin_lock_init(&eas.eas_lock);
779 init_waitqueue_head(&eas.eas_waitq);
780 eas.eas_in_flight = 0;
783 INIT_LIST_HEAD(&eas.eas_avail);
785 /* prepare the group of pages that we're going to be keeping
787 for (i = 0; i < npages; i++) {
788 struct page *page = alloc_page(GFP_KERNEL);
790 GOTO(out, rc = -ENOMEM);
793 list_add_tail(&PAGE_LIST(page), &pages);
795 OBD_ALLOC(eap, sizeof(*eap));
797 GOTO(out, rc = -ENOMEM);
799 eap->eap_magic = EAP_MAGIC;
800 eap->eap_page = page;
802 page->private = (unsigned long)eap;
803 list_add_tail(&eap->eap_item, &eas.eas_avail);
806 /* first we spin queueing io and being woken by its completion */
807 spin_lock_irqsave(&eas.eas_lock, flags);
811 /* sleep until we have a page to send */
812 spin_unlock_irqrestore(&eas.eas_lock, flags);
813 rc = wait_event_interruptible(eas.eas_waitq,
814 eas_should_wake(&eas));
815 spin_lock_irqsave(&eas.eas_lock, flags);
816 if (rc && !eas.eas_rc)
820 if (list_empty(&eas.eas_avail))
822 eap = list_entry(eas.eas_avail.next, struct echo_async_page,
824 list_del(&eap->eap_item);
825 spin_unlock_irqrestore(&eas.eas_lock, flags);
827 /* unbind the eap from its old page offset */
828 if (eap->eap_cookie != NULL) {
829 obd_teardown_async_page(exp, lsm, NULL,
831 eap->eap_cookie = NULL;
834 eas.eas_next_offset += PAGE_SIZE;
835 eap->eap_off = eas.eas_next_offset;
837 rc = obd_prep_async_page(exp, lsm, NULL, eap->eap_page,
838 eap->eap_off, &ec_async_page_ops,
839 eap, &eap->eap_cookie);
841 spin_lock_irqsave(&eas.eas_lock, flags);
846 if (oa->o_id != ECHO_PERSISTENT_OBJID &&
847 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
848 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0)
849 echo_client_page_debug_setup(lsm, eap->eap_page, rw,
851 eap->eap_off, PAGE_SIZE);
853 /* always asserts urgent, which isn't quite right */
854 rc = obd_queue_async_io(exp, lsm, NULL, eap->eap_cookie,
856 ASYNC_READY | ASYNC_URGENT |
858 spin_lock_irqsave(&eas.eas_lock, flags);
859 if (rc && !eas.eas_rc) {
864 if (eas.eas_next_offset == eas.eas_end_offset)
868 /* still hold the eas_lock here.. */
870 /* now we just spin waiting for all the rpcs to complete */
871 while(eas.eas_in_flight) {
872 spin_unlock_irqrestore(&eas.eas_lock, flags);
873 wait_event_interruptible(eas.eas_waitq,
874 eas.eas_in_flight == 0);
875 spin_lock_irqsave(&eas.eas_lock, flags);
877 spin_unlock_irqrestore(&eas.eas_lock, flags);
880 list_for_each_safe(pos, n, &pages) {
881 struct page *page = list_entry(pos, struct page,
884 list_del(&PAGE_LIST(page));
885 if (page->private != 0) {
886 eap = (struct echo_async_page *)page->private;
887 if (eap->eap_cookie != NULL)
888 obd_teardown_async_page(exp, lsm, NULL,
890 OBD_FREE(eap, sizeof(*eap));
898 static int echo_client_prep_commit(struct obd_export *exp, int rw,
899 struct obdo *oa, struct lov_stripe_md *lsm,
900 obd_off offset, obd_size count,
901 obd_size batch, struct obd_trans_info *oti)
903 struct obd_ioobj ioo;
904 struct niobuf_local *lnb;
905 struct niobuf_remote *rnb;
907 obd_size npages, tot_pages;
911 if (count <= 0 || (count & (PAGE_SIZE - 1)) != 0 ||
912 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
915 npages = batch >> PAGE_SHIFT;
916 tot_pages = count >> PAGE_SHIFT;
918 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
919 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
921 if (lnb == NULL || rnb == NULL)
922 GOTO(out, ret = -ENOMEM);
924 obdo_to_ioobj(oa, &ioo);
928 for(; tot_pages; tot_pages -= npages) {
929 if (tot_pages < npages)
932 for (i = 0; i < npages; i++, off += PAGE_SIZE) {
934 rnb[i].len = PAGE_SIZE;
937 /* XXX this can't be the best.. */
938 memset(oti, 0, sizeof(*oti));
939 ioo.ioo_bufcnt = npages;
941 ret = obd_preprw(rw, exp, oa, 1, &ioo, npages, rnb, lnb, oti);
945 for (i = 0; i < npages; i++) {
946 struct page *page = lnb[i].page;
949 if (page == NULL && lnb[i].rc == 0)
952 if (oa->o_id == ECHO_PERSISTENT_OBJID ||
953 (oa->o_valid & OBD_MD_FLFLAGS) == 0 ||
954 (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0)
958 if (rw == OBD_BRW_WRITE)
959 echo_client_page_debug_setup(lsm, page, rw,
964 echo_client_page_debug_check(lsm, page,
970 ret = obd_commitrw(rw, exp, oa, 1, &ioo, npages, lnb, oti, ret);
977 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
979 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
983 int echo_client_brw_ioctl(int rw, struct obd_export *exp,
984 struct obd_ioctl_data *data)
986 struct obd_device *obd = class_exp2obd(exp);
987 struct echo_client_obd *ec = &obd->u.echocli;
988 struct obd_trans_info dummy_oti;
989 struct ec_object *eco;
993 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
997 memset(&dummy_oti, 0, sizeof(dummy_oti));
999 data->ioc_obdo1.o_valid &= ~OBD_MD_FLHANDLE;
1000 data->ioc_obdo1.o_valid |= OBD_MD_FLGROUP;
1001 data->ioc_obdo1.o_gr = FILTER_GROUP_ECHO;
1003 switch((long)data->ioc_pbuf1) {
1005 if (data->ioc_pbuf2 == NULL) { // NULL user data pointer
1006 rc = echo_client_kbrw(obd, rw, &data->ioc_obdo1,
1007 eco->eco_lsm, data->ioc_offset,
1008 data->ioc_count, &dummy_oti);
1011 rc = echo_client_ubrw(obd, rw, &data->ioc_obdo1,
1012 eco->eco_lsm, data->ioc_offset,
1013 data->ioc_count, data->ioc_pbuf2,
1019 rc = echo_client_async_page(ec->ec_exp, rw, &data->ioc_obdo1,
1020 eco->eco_lsm, data->ioc_offset,
1021 data->ioc_count, data->ioc_plen1);
1024 rc = echo_client_prep_commit(ec->ec_exp, rw, &data->ioc_obdo1,
1025 eco->eco_lsm, data->ioc_offset,
1026 data->ioc_count, data->ioc_plen1,
1032 echo_put_object(eco);
1037 echo_ldlm_callback (struct ldlm_lock *lock, struct ldlm_lock_desc *new,
1038 void *data, int flag)
1040 struct ec_object *eco = (struct ec_object *)data;
1041 struct echo_client_obd *ec = &(eco->eco_device->u.echocli);
1042 struct lustre_handle lockh;
1043 struct list_head *el;
1047 ldlm_lock2handle (lock, &lockh);
1049 /* #ifdef this out if we're not feeling paranoid */
1050 spin_lock (&ec->ec_lock);
1051 list_for_each (el, &ec->ec_objects) {
1052 found = (eco == list_entry(el, struct ec_object,
1057 spin_unlock (&ec->ec_lock);
1061 case LDLM_CB_BLOCKING:
1062 CDEBUG(D_INFO, "blocking callback on "LPX64", handle "LPX64"\n",
1063 eco->eco_id, lockh.cookie);
1064 rc = ldlm_cli_cancel (&lockh);
1066 CERROR ("ldlm_cli_cancel failed: %d\n", rc);
1069 case LDLM_CB_CANCELING:
1070 CDEBUG(D_INFO, "cancel callback on "LPX64", handle "LPX64"\n",
1071 eco->eco_id, lockh.cookie);
1082 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
1083 int mode, obd_off offset, obd_size nob)
1085 struct obd_device *obd = exp->exp_obd;
1086 struct echo_client_obd *ec = &obd->u.echocli;
1087 struct lustre_handle *ulh = obdo_handle (oa);
1088 struct ec_object *eco;
1089 struct ec_lock *ecl;
1093 if (!(mode == LCK_PR || mode == LCK_PW))
1096 if ((offset & (PAGE_SIZE - 1)) != 0 ||
1097 (nob & (PAGE_SIZE - 1)) != 0)
1100 rc = echo_get_object (&eco, obd, oa);
1105 OBD_ALLOC (ecl, sizeof (*ecl));
1109 ecl->ecl_mode = mode;
1110 ecl->ecl_object = eco;
1111 ecl->ecl_policy.l_extent.start = offset;
1112 ecl->ecl_policy.l_extent.end =
1113 (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
1116 rc = obd_enqueue(ec->ec_exp, eco->eco_lsm, LDLM_EXTENT,
1117 &ecl->ecl_policy, mode, &flags, echo_ldlm_callback,
1118 ldlm_completion_ast, NULL, eco, sizeof(struct ost_lvb),
1119 lustre_swab_ost_lvb, &ecl->ecl_lock_handle);
1123 CDEBUG(D_INFO, "enqueue handle "LPX64"\n", ecl->ecl_lock_handle.cookie);
1125 /* NB ecl takes object ref from echo_get_object() above */
1126 spin_lock(&ec->ec_lock);
1128 list_add(&ecl->ecl_exp_chain, &exp->exp_ec_data.eced_locks);
1129 ulh->cookie = ecl->ecl_cookie = ec->ec_unique++;
1131 spin_unlock(&ec->ec_lock);
1133 oa->o_valid |= OBD_MD_FLHANDLE;
1137 OBD_FREE (ecl, sizeof (*ecl));
1139 echo_put_object (eco);
1144 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
1146 struct obd_device *obd = exp->exp_obd;
1147 struct echo_client_obd *ec = &obd->u.echocli;
1148 struct lustre_handle *ulh = obdo_handle (oa);
1149 struct ec_lock *ecl = NULL;
1151 struct list_head *el;
1154 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
1157 spin_lock (&ec->ec_lock);
1159 list_for_each (el, &exp->exp_ec_data.eced_locks) {
1160 ecl = list_entry (el, struct ec_lock, ecl_exp_chain);
1161 found = (ecl->ecl_cookie == ulh->cookie);
1163 list_del (&ecl->ecl_exp_chain);
1168 spin_unlock (&ec->ec_lock);
1173 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm, ecl->ecl_mode,
1174 &ecl->ecl_lock_handle);
1176 echo_put_object (ecl->ecl_object);
1177 OBD_FREE (ecl, sizeof (*ecl));
1183 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp,
1184 int len, void *karg, void *uarg)
1186 struct obd_device *obd;
1187 struct echo_client_obd *ec;
1188 struct ec_object *eco;
1189 struct obd_ioctl_data *data = karg;
1190 struct obd_trans_info dummy_oti;
1191 struct oti_req_ack_lock *ack_lock;
1193 int rw = OBD_BRW_READ;
1200 memset(&dummy_oti, 0, sizeof(dummy_oti));
1203 ec = &obd->u.echocli;
1206 case OBD_IOC_CREATE: /* may create echo object */
1207 if (!capable (CAP_SYS_ADMIN))
1208 GOTO (out, rc = -EPERM);
1210 rc = echo_create_object (obd, 1, &data->ioc_obdo1,
1211 data->ioc_pbuf1, data->ioc_plen1,
1215 case OBD_IOC_DESTROY:
1216 if (!capable (CAP_SYS_ADMIN))
1217 GOTO (out, rc = -EPERM);
1219 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1221 oa = &data->ioc_obdo1;
1222 oa->o_gr = FILTER_GROUP_ECHO;
1223 oa->o_valid |= OBD_MD_FLGROUP;
1224 rc = obd_destroy(ec->ec_exp, oa, eco->eco_lsm,
1227 eco->eco_deleted = 1;
1228 echo_put_object(eco);
1232 case OBD_IOC_GETATTR:
1233 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1235 rc = obd_getattr(ec->ec_exp, &data->ioc_obdo1,
1237 echo_put_object(eco);
1241 case OBD_IOC_SETATTR:
1242 if (!capable (CAP_SYS_ADMIN))
1243 GOTO (out, rc = -EPERM);
1245 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1247 rc = obd_setattr(ec->ec_exp, &data->ioc_obdo1,
1248 eco->eco_lsm, NULL);
1249 echo_put_object(eco);
1253 case OBD_IOC_BRW_WRITE:
1254 if (!capable (CAP_SYS_ADMIN))
1255 GOTO (out, rc = -EPERM);
1259 case OBD_IOC_BRW_READ:
1260 rc = echo_client_brw_ioctl(rw, exp, data);
1263 case ECHO_IOC_GET_STRIPE:
1264 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1266 rc = echo_copyout_lsm(eco->eco_lsm, data->ioc_pbuf1,
1268 echo_put_object(eco);
1272 case ECHO_IOC_SET_STRIPE:
1273 if (!capable (CAP_SYS_ADMIN))
1274 GOTO (out, rc = -EPERM);
1276 if (data->ioc_pbuf1 == NULL) { /* unset */
1277 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1279 eco->eco_deleted = 1;
1280 echo_put_object(eco);
1283 rc = echo_create_object(obd, 0, &data->ioc_obdo1,
1285 data->ioc_plen1, &dummy_oti);
1289 case ECHO_IOC_ENQUEUE:
1290 if (!capable (CAP_SYS_ADMIN))
1291 GOTO (out, rc = -EPERM);
1293 rc = echo_client_enqueue(exp, &data->ioc_obdo1,
1294 data->ioc_conn1, /* lock mode */
1295 data->ioc_offset, data->ioc_count);/*extent*/
1298 case ECHO_IOC_CANCEL:
1299 rc = echo_client_cancel(exp, &data->ioc_obdo1);
1303 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
1304 GOTO (out, rc = -ENOTTY);
1310 /* XXX this should be in a helper also called by target_send_reply */
1311 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
1313 if (!ack_lock->mode)
1315 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
1324 echo_client_setup(struct obd_device *obddev, obd_count len, void *buf)
1326 struct lustre_cfg* lcfg = buf;
1327 struct echo_client_obd *ec = &obddev->u.echocli;
1328 struct obd_device *tgt;
1329 struct lustre_handle conn = {0, };
1330 struct obd_uuid echo_uuid = { "ECHO_UUID" };
1334 if (lcfg->lcfg_inllen1 < 1) {
1335 CERROR("requires a TARGET OBD name\n");
1339 tgt = class_name2obd(lcfg->lcfg_inlbuf1);
1340 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
1341 CERROR("device not attached or not set up (%s)\n",
1342 lcfg->lcfg_inlbuf1);
1346 spin_lock_init (&ec->ec_lock);
1347 INIT_LIST_HEAD (&ec->ec_objects);
1350 rc = obd_connect(&conn, tgt, &echo_uuid, NULL, FILTER_GROUP_ECHO);
1352 CERROR("fail to connect to device %s\n", lcfg->lcfg_inlbuf1);
1355 ec->ec_exp = class_conn2export(&conn);
1360 static int echo_client_cleanup(struct obd_device *obddev, int flags)
1362 struct list_head *el;
1363 struct ec_object *eco;
1364 struct echo_client_obd *ec = &obddev->u.echocli;
1368 if (!list_empty(&obddev->obd_exports)) {
1369 CERROR("still has clients!\n");
1373 /* XXX assuming sole access */
1374 while (!list_empty(&ec->ec_objects)) {
1375 el = ec->ec_objects.next;
1376 eco = list_entry(el, struct ec_object, eco_obj_chain);
1378 LASSERT(eco->eco_refcount == 0);
1379 eco->eco_refcount = 1;
1380 eco->eco_deleted = 1;
1381 echo_put_object(eco);
1384 rc = obd_disconnect(ec->ec_exp, 0);
1386 CERROR("fail to disconnect device: %d\n", rc);
1391 static int echo_client_connect(struct lustre_handle *conn,
1392 struct obd_device *src,
1393 struct obd_uuid *cluuid,
1394 struct obd_connect_data *data,
1395 unsigned long flags)
1397 struct obd_export *exp;
1400 rc = class_connect(conn, src, cluuid);
1402 exp = class_conn2export(conn);
1403 INIT_LIST_HEAD(&exp->exp_ec_data.eced_locks);
1404 class_export_put(exp);
1410 static int echo_client_disconnect(struct obd_export *exp,
1411 unsigned long flags)
1413 struct obd_device *obd;
1414 struct echo_client_obd *ec;
1415 struct ec_lock *ecl;
1420 GOTO(out, rc = -EINVAL);
1423 ec = &obd->u.echocli;
1425 /* no more contention on export's lock list */
1426 while (!list_empty (&exp->exp_ec_data.eced_locks)) {
1427 ecl = list_entry (exp->exp_ec_data.eced_locks.next,
1428 struct ec_lock, ecl_exp_chain);
1429 list_del (&ecl->ecl_exp_chain);
1431 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
1432 ecl->ecl_mode, &ecl->ecl_lock_handle);
1434 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
1435 "(%d)\n", ecl->ecl_object->eco_id, rc);
1437 echo_put_object (ecl->ecl_object);
1438 OBD_FREE (ecl, sizeof (*ecl));
1441 rc = class_disconnect(exp, 0);
1447 static struct obd_ops echo_obd_ops = {
1448 .o_owner = THIS_MODULE,
1449 .o_setup = echo_client_setup,
1450 .o_cleanup = echo_client_cleanup,
1451 .o_iocontrol = echo_client_iocontrol,
1452 .o_connect = echo_client_connect,
1453 .o_disconnect = echo_client_disconnect
1456 int echo_client_init(void)
1458 struct lprocfs_static_vars lvars;
1460 lprocfs_init_vars(echo, &lvars);
1461 return class_register_type(&echo_obd_ops, NULL, lvars.module_vars,
1462 OBD_ECHO_CLIENT_DEVICENAME);
1465 void echo_client_exit(void)
1467 class_unregister_type(OBD_ECHO_CLIENT_DEVICENAME);