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>
33 #include <liblustre.h>
36 #include <linux/obd.h>
37 #include <linux/obd_support.h>
38 #include <linux/obd_class.h>
39 #include <linux/obd_echo.h>
40 #include <linux/lustre_debug.h>
41 #include <linux/lprocfs_status.h>
43 static obd_id last_object_id;
47 echo_printk_object (char *msg, struct ec_object *eco)
49 struct lov_stripe_md *lsm = eco->eco_lsm;
52 printk (KERN_INFO "Lustre: %s: object %p: "LPX64", refs %d%s: "LPX64
53 "=%u!%u\n", msg, eco, eco->eco_id, eco->eco_refcount,
54 eco->eco_deleted ? "(deleted) " : "",
55 lsm->lsm_object_id, lsm->lsm_stripe_size,
56 lsm->lsm_stripe_count);
58 for (i = 0; i < lsm->lsm_stripe_count; i++)
59 printk (KERN_INFO "Lustre: @%2u:"LPX64"\n",
60 lsm->lsm_oinfo[i].loi_ost_idx,
61 lsm->lsm_oinfo[i].loi_id);
65 static struct ec_object *
66 echo_find_object_locked (struct obd_device *obd, obd_id id)
68 struct echo_client_obd *ec = &obd->u.echo_client;
69 struct ec_object *eco = NULL;
72 list_for_each (el, &ec->ec_objects) {
73 eco = list_entry (el, struct ec_object, eco_obj_chain);
75 if (eco->eco_id == id)
82 echo_copyout_lsm (struct lov_stripe_md *lsm, void *ulsm, int ulsm_nob)
86 nob = offsetof (struct lov_stripe_md, lsm_oinfo[lsm->lsm_stripe_count]);
90 if (copy_to_user (ulsm, lsm, nob))
97 echo_copyin_lsm (struct obd_device *obd, struct lov_stripe_md *lsm,
98 void *ulsm, int ulsm_nob)
100 struct echo_client_obd *ec = &obd->u.echo_client;
103 if (ulsm_nob < sizeof (*lsm))
106 if (copy_from_user (lsm, ulsm, sizeof (*lsm)))
109 nob = lsm->lsm_stripe_count * sizeof (lsm->lsm_oinfo[0]);
111 if (ulsm_nob < nob ||
112 lsm->lsm_stripe_count > ec->ec_nstripes ||
113 lsm->lsm_magic != LOV_MAGIC ||
114 (lsm->lsm_stripe_size & (PAGE_SIZE - 1)) != 0 ||
115 ((__u64)lsm->lsm_stripe_size * lsm->lsm_stripe_count > ~0UL))
118 if (copy_from_user(lsm->lsm_oinfo,
119 ((struct lov_stripe_md *)ulsm)->lsm_oinfo, nob))
125 static struct ec_object *
126 echo_allocate_object (struct obd_device *obd)
128 struct echo_client_obd *ec = &obd->u.echo_client;
129 struct ec_object *eco;
132 OBD_ALLOC(eco, sizeof (*eco));
136 rc = obd_alloc_memmd(ec->ec_exp, &eco->eco_lsm);
138 OBD_FREE(eco, sizeof (*eco));
142 eco->eco_device = obd;
143 eco->eco_deleted = 0;
144 eco->eco_refcount = 0;
145 eco->eco_lsm->lsm_magic = LOV_MAGIC;
146 /* leave stripe count 0 by default */
152 echo_free_object (struct ec_object *eco)
154 struct obd_device *obd = eco->eco_device;
155 struct echo_client_obd *ec = &obd->u.echo_client;
157 LASSERT (eco->eco_refcount == 0);
158 obd_free_memmd(ec->ec_exp, &eco->eco_lsm);
159 OBD_FREE (eco, sizeof (*eco));
162 static int echo_create_object(struct obd_device *obd, int on_target,
163 struct obdo *oa, void *ulsm, int ulsm_nob,
164 struct obd_trans_info *oti)
166 struct echo_client_obd *ec = &obd->u.echo_client;
167 struct ec_object *eco2;
168 struct ec_object *eco;
169 struct lov_stripe_md *lsm;
173 if ((oa->o_valid & OBD_MD_FLID) == 0 && /* no obj id */
174 (on_target || /* set_stripe */
175 ec->ec_nstripes != 0)) { /* LOV */
176 CERROR ("No valid oid\n");
181 eco = echo_allocate_object (obd);
187 rc = echo_copyin_lsm (obd, lsm, ulsm, ulsm_nob);
191 /* setup object ID here for !on_target and LOV hint */
192 if ((oa->o_valid & OBD_MD_FLID) != 0)
193 eco->eco_id = lsm->lsm_object_id = oa->o_id;
195 if (lsm->lsm_stripe_count == 0)
196 lsm->lsm_stripe_count = ec->ec_nstripes;
198 if (lsm->lsm_stripe_size == 0)
199 lsm->lsm_stripe_size = PAGE_SIZE;
201 idx = ll_insecure_random_int();
203 /* setup stripes: indices + default ids if required */
204 for (i = 0; i < lsm->lsm_stripe_count; i++) {
205 if (lsm->lsm_oinfo[i].loi_id == 0)
206 lsm->lsm_oinfo[i].loi_id = lsm->lsm_object_id;
208 lsm->lsm_oinfo[i].loi_ost_idx =
209 (idx + i) % ec->ec_nstripes;
212 OBD_ALLOC(eco, sizeof(*eco));
213 eco->eco_device = obd;
218 oa->o_id = ++last_object_id;
221 /* XXX get some filter group constants */
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.echo_client;
273 struct ec_object *eco;
274 struct ec_object *eco2;
277 if ((oa->o_valid & OBD_MD_FLID) == 0)
279 CERROR ("No valid oid\n");
283 spin_lock (&ec->ec_lock);
284 eco = echo_find_object_locked (obd, oa->o_id);
286 if (eco->eco_deleted) /* being deleted */
287 return (-EAGAIN); /* (see comment in cleanup) */
290 spin_unlock (&ec->ec_lock);
293 "found %p: "LPX64"=%u#%u@%u refs %d del %d\n",
295 eco->eco_lsm->lsm_stripe_size,
296 eco->eco_lsm->lsm_stripe_count,
297 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
298 eco->eco_refcount, eco->eco_deleted);
301 spin_unlock (&ec->ec_lock);
303 if (ec->ec_nstripes != 0) /* striping required */
306 eco = echo_allocate_object (obd);
310 eco->eco_id = eco->eco_lsm->lsm_object_id = oa->o_id;
312 spin_lock (&ec->ec_lock);
314 eco2 = echo_find_object_locked (obd, oa->o_id);
315 if (eco2 == NULL) { /* didn't race */
316 list_add (&eco->eco_obj_chain, &ec->ec_objects);
317 spin_unlock (&ec->ec_lock);
318 eco->eco_refcount = 1;
321 "created %p: "LPX64"=%u#%u@%d refs %d del %d\n",
323 eco->eco_lsm->lsm_stripe_size,
324 eco->eco_lsm->lsm_stripe_count,
325 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
326 eco->eco_refcount, eco->eco_deleted);
330 if (eco2->eco_deleted)
331 rc = -EAGAIN; /* lose race */
333 eco2->eco_refcount++; /* take existing */
336 LASSERT (eco2->eco_id == eco2->eco_lsm->lsm_object_id);
338 "found(2) %p: "LPX64"=%u#%u@%d refs %d del %d\n",
340 eco2->eco_lsm->lsm_stripe_size,
341 eco2->eco_lsm->lsm_stripe_count,
342 eco2->eco_lsm->lsm_oinfo[0].loi_ost_idx,
343 eco2->eco_refcount, eco2->eco_deleted);
346 spin_unlock (&ec->ec_lock);
348 echo_free_object (eco);
353 echo_put_object (struct ec_object *eco)
355 struct obd_device *obd = eco->eco_device;
356 struct echo_client_obd *ec = &obd->u.echo_client;
358 /* Release caller's ref on the object.
359 * delete => mark for deletion when last ref goes
362 spin_lock (&ec->ec_lock);
365 LASSERT (eco->eco_refcount >= 0);
367 CDEBUG(D_INFO, "put %p: "LPX64"=%u#%u@%d refs %d del %d\n",
369 eco->eco_lsm->lsm_stripe_size,
370 eco->eco_lsm->lsm_stripe_count,
371 eco->eco_lsm->lsm_oinfo[0].loi_ost_idx,
372 eco->eco_refcount, eco->eco_deleted);
374 if (eco->eco_refcount != 0 || !eco->eco_deleted) {
375 spin_unlock (&ec->ec_lock);
379 spin_unlock (&ec->ec_lock);
381 /* NB leave obj in the object list. We must prevent anyone from
382 * attempting to enqueue on this object number until we can be
383 * sure there will be no more lock callbacks.
385 obd_cancel_unused(ec->ec_exp, eco->eco_lsm, 0, NULL);
387 /* now we can let it go */
388 spin_lock (&ec->ec_lock);
389 list_del (&eco->eco_obj_chain);
390 spin_unlock (&ec->ec_lock);
392 LASSERT (eco->eco_refcount == 0);
394 echo_free_object (eco);
398 echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp)
400 unsigned long stripe_count;
401 unsigned long stripe_size;
403 unsigned long woffset;
407 if (lsm->lsm_stripe_count <= 1)
411 stripe_size = lsm->lsm_stripe_size;
412 stripe_count = lsm->lsm_stripe_count;
414 /* width = # bytes in all stripes */
415 width = stripe_size * stripe_count;
417 /* woffset = offset within a width; offset = whole number of widths */
418 woffset = do_div (offset, width);
420 stripe_index = woffset / stripe_size;
422 *idp = lsm->lsm_oinfo[stripe_index].loi_id;
423 *offp = offset * stripe_size + woffset % stripe_size;
426 static void echo_page_debug_setup(struct lov_stripe_md *lsm,
427 struct page *page, int rw, obd_id id,
428 obd_off offset, obd_off count)
439 if (rw == OBD_BRW_WRITE) {
442 echo_get_stripe_off_id(lsm, &stripe_off, &stripe_id);
444 stripe_off = 0xdeadbeef00c0ffeeULL;
445 stripe_id = 0xdeadbeef00c0ffeeULL;
447 page_debug_setup(addr, count, stripe_off, stripe_id);
452 static int echo_page_debug_check(struct lov_stripe_md *lsm,
453 struct page *page, obd_id id,
454 obd_off offset, obd_off count)
456 obd_off stripe_off = offset;
457 obd_id stripe_id = id;
465 echo_get_stripe_off_id (lsm, &stripe_off, &stripe_id);
466 rc = page_debug_check("test_brw", addr, count, stripe_off, stripe_id);
471 static int echo_client_kbrw(struct obd_device *obd, int rw, struct obdo *oa,
472 struct lov_stripe_md *lsm, obd_off offset,
473 obd_size count, struct obd_trans_info *oti)
475 struct echo_client_obd *ec = &obd->u.echo_client;
477 struct brw_page *pga;
478 struct brw_page *pgp;
485 /* oa_id == 0 => speed test (no verification) else...
486 * oa & 1 => use HIGHMEM
488 gfp_mask = ((oa->o_id & 1) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
490 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
493 (count & (PAGE_SIZE - 1)) != 0 ||
495 lsm->lsm_object_id != oa->o_id))
498 /* XXX think again with misaligned I/O */
499 npages = count >> PAGE_SHIFT;
501 OBD_ALLOC(pga, npages * sizeof(*pga));
505 for (i = 0, pgp = pga, off = offset;
507 i++, pgp++, off += PAGE_SIZE) {
509 LASSERT (pgp->pg == NULL); /* for cleanup */
512 pgp->pg = alloc_pages (gfp_mask, 0);
516 pgp->count = PAGE_SIZE;
520 echo_page_debug_setup(lsm, pgp->pg, rw, oa->o_id, off,
524 rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti);
527 if (rc == 0 && rw == OBD_BRW_READ)
530 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
536 vrc = echo_page_debug_check(lsm, pgp->pg, oa->o_id,
537 pgp->off, pgp->count);
538 if (vrc != 0 && rc == 0)
541 __free_pages(pgp->pg, 0);
543 OBD_FREE(pga, npages * sizeof(*pga));
548 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
549 static int echo_client_ubrw(struct obd_device *obd, int rw,
550 struct obdo *oa, struct lov_stripe_md *lsm,
551 obd_off offset, obd_size count, char *buffer,
552 struct obd_trans_info *oti)
554 struct echo_client_obd *ec = &obd->u.echo_client;
556 struct brw_page *pga;
557 struct brw_page *pgp;
559 struct kiobuf *kiobuf;
563 LASSERT (rw == OBD_BRW_WRITE ||
566 /* NB: for now, only whole pages, page aligned */
569 ((long)buffer & (PAGE_SIZE - 1)) != 0 ||
570 (count & (PAGE_SIZE - 1)) != 0 ||
571 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
574 /* XXX think again with misaligned I/O */
575 npages = count >> PAGE_SHIFT;
577 OBD_ALLOC(pga, npages * sizeof(*pga));
581 rc = alloc_kiovec (1, &kiobuf);
585 rc = map_user_kiobuf ((rw == OBD_BRW_READ) ? READ : WRITE,
586 kiobuf, (unsigned long)buffer, count);
590 LASSERT (kiobuf->offset == 0);
591 LASSERT (kiobuf->nr_pages == npages);
593 for (i = 0, off = offset, pgp = pga;
595 i++, off += PAGE_SIZE, pgp++) {
597 pgp->pg = kiobuf->maplist[i];
598 pgp->count = PAGE_SIZE;
602 rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti);
604 // if (rw == OBD_BRW_READ)
605 // mark_dirty_kiobuf (kiobuf, count);
607 unmap_kiobuf (kiobuf);
609 free_kiovec (1, &kiobuf);
611 OBD_FREE(pga, npages * sizeof(*pga));
615 static int echo_client_ubrw(struct obd_device *obd, int rw,
616 struct obdo *oa, struct lov_stripe_md *lsm,
617 obd_off offset, obd_size count, char *buffer,
618 struct obd_trans_info *oti)
620 #warning "echo_client_ubrw() needs to be ported on 2.6 yet"
627 struct echo_async_state;
629 #define EAP_MAGIC 79277927
630 struct echo_async_page {
632 struct page *eap_page;
635 struct echo_async_state *eap_eas;
636 struct list_head eap_item;
639 struct echo_async_state {
641 obd_off eas_next_offset;
642 obd_off eas_end_offset;
645 wait_queue_head_t eas_waitq;
646 struct list_head eas_avail;
648 struct lov_stripe_md *eas_lsm;
651 static int eas_should_wake(struct echo_async_state *eas)
655 spin_lock_irqsave(&eas->eas_lock, flags);
656 if (eas->eas_rc == 0 && !list_empty(&eas->eas_avail))
658 spin_unlock_irqrestore(&eas->eas_lock, flags);
662 struct echo_async_page *eap_from_cookie(void *cookie)
664 struct echo_async_page *eap = cookie;
665 if (eap->eap_magic != EAP_MAGIC)
666 return ERR_PTR(-EINVAL);
670 static int ec_ap_make_ready(void *data, int cmd)
672 /* our pages are issued ready */
676 static int ec_ap_refresh_count(void *data, int cmd)
678 /* our pages are issued with a stable count */
682 static void ec_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
684 struct echo_async_page *eap;
685 eap = eap_from_cookie(data);
689 memcpy(oa, &eap->eap_eas->eas_oa, sizeof(*oa));
691 static void ec_ap_completion(void *data, int cmd, int rc)
693 struct echo_async_page *eap = eap_from_cookie(data);
694 struct echo_async_state *eas;
701 if (cmd == OBD_BRW_READ)
702 echo_page_debug_check(eas->eas_lsm, eap->eap_page,
703 eas->eas_oa.o_id, eap->eap_off,
706 spin_lock_irqsave(&eas->eas_lock, flags);
707 if (rc && !eas->eas_rc)
709 eas->eas_in_flight--;
710 list_add(&eap->eap_item, &eas->eas_avail);
711 wake_up(&eas->eas_waitq);
712 spin_unlock_irqrestore(&eas->eas_lock, flags);
715 static struct obd_async_page_ops ec_async_page_ops = {
716 .ap_make_ready = ec_ap_make_ready,
717 .ap_refresh_count = ec_ap_refresh_count,
718 .ap_fill_obdo = ec_ap_fill_obdo,
719 .ap_completion = ec_ap_completion,
722 static int echo_client_async_page(struct obd_export *exp, int rw,
723 struct obdo *oa, struct lov_stripe_md *lsm,
724 obd_off offset, obd_size count,
728 struct echo_async_page *eap;
729 struct echo_async_state eas;
730 struct list_head *pos, *n;
737 /* oa_id == 0 => speed test (no verification) else...
738 * oa & 1 => use HIGHMEM
740 verify = (oa->o_id != 0);
741 gfp_mask = ((oa->o_id & 1) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
744 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
747 (count & (PAGE_SIZE - 1)) != 0 ||
749 lsm->lsm_object_id != oa->o_id))
752 /* XXX think again with misaligned I/O */
753 npages = batching >> PAGE_SHIFT;
755 memcpy(&eas.eas_oa, oa, sizeof(*oa));
756 eas.eas_next_offset = offset;
757 eas.eas_end_offset = offset + count;
758 spin_lock_init(&eas.eas_lock);
759 init_waitqueue_head(&eas.eas_waitq);
760 eas.eas_in_flight = 0;
763 INIT_LIST_HEAD(&eas.eas_avail);
765 /* prepare the group of pages that we're going to be keeping
767 for (i = 0; i < npages; i++) {
768 struct page *page = alloc_page(GFP_KERNEL);
770 GOTO(out, rc = -ENOMEM);
773 list_add_tail(&page->list, &pages);
775 OBD_ALLOC(eap, sizeof(*eap));
777 GOTO(out, rc = -ENOMEM);
779 eap->eap_magic = EAP_MAGIC;
780 eap->eap_page = page;
782 page->private = (unsigned long)eap;
783 list_add_tail(&eap->eap_item, &eas.eas_avail);
786 /* first we spin queueing io and being woken by its completion */
787 spin_lock_irqsave(&eas.eas_lock, flags);
791 /* sleep until we have a page to send */
792 spin_unlock_irqrestore(&eas.eas_lock, flags);
793 rc = wait_event_interruptible(eas.eas_waitq,
794 eas_should_wake(&eas));
795 spin_lock_irqsave(&eas.eas_lock, flags);
796 if (rc && !eas.eas_rc)
800 if (list_empty(&eas.eas_avail))
802 eap = list_entry(eas.eas_avail.next, struct echo_async_page,
804 list_del(&eap->eap_item);
805 spin_unlock_irqrestore(&eas.eas_lock, flags);
807 /* unbind the eap from its old page offset */
808 if (eap->eap_cookie != NULL) {
809 obd_teardown_async_page(exp, lsm, NULL,
811 eap->eap_cookie = NULL;
814 eas.eas_next_offset += PAGE_SIZE;
815 eap->eap_off = eas.eas_next_offset;
817 rc = obd_prep_async_page(exp, lsm, NULL, eap->eap_page,
818 eap->eap_off, &ec_async_page_ops,
819 eap, &eap->eap_cookie);
821 spin_lock_irqsave(&eas.eas_lock, flags);
826 if (rw == OBD_BRW_WRITE)
827 echo_page_debug_setup(lsm, eap->eap_page, rw, oa->o_id,
828 eap->eap_off, PAGE_SIZE);
830 /* always asserts urgent, which isn't quite right */
831 rc = obd_queue_async_io(exp, lsm, NULL, eap->eap_cookie,
833 ASYNC_READY | ASYNC_URGENT |
835 spin_lock_irqsave(&eas.eas_lock, flags);
836 if (rc && !eas.eas_rc) {
841 if (eas.eas_next_offset == eas.eas_end_offset)
845 /* still hold the eas_lock here.. */
847 /* now we just spin waiting for all the rpcs to complete */
848 while(eas.eas_in_flight) {
849 spin_unlock_irqrestore(&eas.eas_lock, flags);
850 wait_event_interruptible(eas.eas_waitq,
851 eas.eas_in_flight == 0);
852 spin_lock_irqsave(&eas.eas_lock, flags);
854 spin_unlock_irqrestore(&eas.eas_lock, flags);
857 list_for_each_safe(pos, n, &pages) {
858 struct page *page = list_entry(pos, struct page, list);
860 list_del(&page->list);
861 if (page->private != 0) {
862 eap = (struct echo_async_page *)page->private;
863 if (eap->eap_cookie != NULL)
864 obd_teardown_async_page(exp, lsm, NULL,
866 OBD_FREE(eap, sizeof(*eap));
874 static int echo_client_prep_commit(struct obd_export *exp, int rw,
875 struct obdo *oa, struct lov_stripe_md *lsm,
876 obd_off offset, obd_size count,
877 obd_size batch, struct obd_trans_info *oti)
879 struct obd_ioobj ioo;
880 struct niobuf_local *lnb;
881 struct niobuf_remote *rnb;
883 obd_size npages, tot_pages;
884 int i, ret = 0, err = 0;
887 if (count <= 0 || (count & (PAGE_SIZE - 1)) != 0 ||
888 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
891 npages = batch >> PAGE_SHIFT;
892 tot_pages = count >> PAGE_SHIFT;
894 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
895 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
897 if (lnb == NULL || rnb == NULL)
898 GOTO(out, ret = -ENOMEM);
900 obdo_to_ioobj(oa, &ioo);
904 for(; tot_pages; tot_pages -= npages) {
905 if (tot_pages < npages)
908 for (i = 0; i < npages; i++, off += PAGE_SIZE) {
910 rnb[i].len = PAGE_SIZE;
913 /* XXX this can't be the best.. */
914 memset(oti, 0, sizeof(*oti));
915 ioo.ioo_bufcnt = npages;
917 ret = obd_preprw(rw, exp, oa, 1, &ioo, npages, rnb, lnb, oti);
921 for (i = 0; i < npages; i++) {
922 struct page *page = lnb[i].page;
925 if (page == NULL && lnb[i].rc == 0)
928 if (rw == OBD_BRW_WRITE)
929 echo_page_debug_setup(lsm, page, rw, oa->o_id,
933 echo_page_debug_check(lsm, page, oa->o_id,
938 ret = obd_commitrw(rw, exp, oa, 1, &ioo, npages, lnb, oti);
942 GOTO(out, ret = err);
947 OBD_FREE(lnb, npages * sizeof(struct niobuf_local));
949 OBD_FREE(rnb, npages * sizeof(struct niobuf_remote));
953 int echo_client_brw_ioctl(int rw, struct obd_export *exp,
954 struct obd_ioctl_data *data)
956 struct obd_device *obd = class_exp2obd(exp);
957 struct echo_client_obd *ec = &obd->u.echo_client;
958 struct obd_trans_info dummy_oti;
959 struct ec_object *eco;
963 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
967 memset(&dummy_oti, 0, sizeof(dummy_oti));
969 data->ioc_obdo1.o_valid &= ~OBD_MD_FLHANDLE;
970 data->ioc_obdo1.o_valid |= OBD_MD_FLGROUP;
971 data->ioc_obdo1.o_gr = 2;
973 switch((long)data->ioc_pbuf1) {
975 if (data->ioc_pbuf2 == NULL) { // NULL user data pointer
976 rc = echo_client_kbrw(obd, rw, &data->ioc_obdo1,
977 eco->eco_lsm, data->ioc_offset,
978 data->ioc_count, &dummy_oti);
981 rc = echo_client_ubrw(obd, rw, &data->ioc_obdo1,
982 eco->eco_lsm, data->ioc_offset,
983 data->ioc_count, data->ioc_pbuf2,
989 rc = echo_client_async_page(ec->ec_exp, rw, &data->ioc_obdo1,
990 eco->eco_lsm, data->ioc_offset,
991 data->ioc_count, data->ioc_plen1);
994 rc = echo_client_prep_commit(ec->ec_exp, rw, &data->ioc_obdo1,
995 eco->eco_lsm, data->ioc_offset,
996 data->ioc_count, data->ioc_plen1,
1002 echo_put_object(eco);
1007 echo_ldlm_callback (struct ldlm_lock *lock, struct ldlm_lock_desc *new,
1008 void *data, int flag)
1010 struct ec_object *eco = (struct ec_object *)data;
1011 struct echo_client_obd *ec = &(eco->eco_device->u.echo_client);
1012 struct lustre_handle lockh;
1013 struct list_head *el;
1017 ldlm_lock2handle (lock, &lockh);
1019 /* #ifdef this out if we're not feeling paranoid */
1020 spin_lock (&ec->ec_lock);
1021 list_for_each (el, &ec->ec_objects) {
1022 found = (eco == list_entry(el, struct ec_object,
1027 spin_unlock (&ec->ec_lock);
1031 case LDLM_CB_BLOCKING:
1032 CDEBUG(D_INFO, "blocking callback on "LPX64", handle "LPX64"\n",
1033 eco->eco_id, lockh.cookie);
1034 rc = ldlm_cli_cancel (&lockh);
1036 CERROR ("ldlm_cli_cancel failed: %d\n", rc);
1039 case LDLM_CB_CANCELING:
1040 CDEBUG(D_INFO, "cancel callback on "LPX64", handle "LPX64"\n",
1041 eco->eco_id, lockh.cookie);
1052 echo_client_enqueue(struct obd_export *exp, struct obdo *oa,
1053 int mode, obd_off offset, obd_size nob)
1055 struct obd_device *obd = exp->exp_obd;
1056 struct echo_client_obd *ec = &obd->u.echo_client;
1057 struct lustre_handle *ulh = obdo_handle (oa);
1058 struct ec_object *eco;
1059 struct ec_lock *ecl;
1063 if (!(mode == LCK_PR || mode == LCK_PW))
1066 if ((offset & (PAGE_SIZE - 1)) != 0 ||
1067 (nob & (PAGE_SIZE - 1)) != 0)
1070 rc = echo_get_object (&eco, obd, oa);
1075 OBD_ALLOC (ecl, sizeof (*ecl));
1079 ecl->ecl_mode = mode;
1080 ecl->ecl_object = eco;
1081 ecl->ecl_extent.start = offset;
1082 ecl->ecl_extent.end = (nob == 0) ? ((obd_off) -1) : (offset + nob - 1);
1085 rc = obd_enqueue(ec->ec_exp, eco->eco_lsm, NULL, LDLM_EXTENT,
1086 &ecl->ecl_extent,sizeof(ecl->ecl_extent), mode,
1087 &flags, echo_ldlm_callback, eco,
1088 &ecl->ecl_lock_handle);
1092 CDEBUG(D_INFO, "enqueue handle "LPX64"\n", ecl->ecl_lock_handle.cookie);
1094 /* NB ecl takes object ref from echo_get_object() above */
1095 spin_lock(&ec->ec_lock);
1097 list_add(&ecl->ecl_exp_chain, &exp->exp_ec_data.eced_locks);
1098 ulh->cookie = ecl->ecl_cookie = ec->ec_unique++;
1100 spin_unlock(&ec->ec_lock);
1102 oa->o_valid |= OBD_MD_FLHANDLE;
1106 OBD_FREE (ecl, sizeof (*ecl));
1108 echo_put_object (eco);
1113 echo_client_cancel(struct obd_export *exp, struct obdo *oa)
1115 struct obd_device *obd = exp->exp_obd;
1116 struct echo_client_obd *ec = &obd->u.echo_client;
1117 struct lustre_handle *ulh = obdo_handle (oa);
1118 struct ec_lock *ecl = NULL;
1120 struct list_head *el;
1123 if ((oa->o_valid & OBD_MD_FLHANDLE) == 0)
1126 spin_lock (&ec->ec_lock);
1128 list_for_each (el, &exp->exp_ec_data.eced_locks) {
1129 ecl = list_entry (el, struct ec_lock, ecl_exp_chain);
1130 found = (ecl->ecl_cookie == ulh->cookie);
1132 list_del (&ecl->ecl_exp_chain);
1137 spin_unlock (&ec->ec_lock);
1142 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm, ecl->ecl_mode,
1143 &ecl->ecl_lock_handle);
1145 echo_put_object (ecl->ecl_object);
1146 OBD_FREE (ecl, sizeof (*ecl));
1152 echo_client_iocontrol(unsigned int cmd, struct obd_export *exp,
1153 int len, void *karg, void *uarg)
1155 struct obd_device *obd;
1156 struct echo_client_obd *ec;
1157 struct ec_object *eco;
1158 struct obd_ioctl_data *data = karg;
1159 struct obd_trans_info dummy_oti;
1160 struct oti_req_ack_lock *ack_lock;
1162 int rw = OBD_BRW_READ;
1167 memset(&dummy_oti, 0, sizeof(dummy_oti));
1170 ec = &obd->u.echo_client;
1173 case OBD_IOC_CREATE: /* may create echo object */
1174 if (!capable (CAP_SYS_ADMIN))
1175 GOTO (out, rc = -EPERM);
1177 rc = echo_create_object (obd, 1, &data->ioc_obdo1,
1178 data->ioc_pbuf1, data->ioc_plen1,
1182 case OBD_IOC_DESTROY:
1183 if (!capable (CAP_SYS_ADMIN))
1184 GOTO (out, rc = -EPERM);
1186 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1188 oa = &data->ioc_obdo1;
1190 oa->o_valid |= OBD_MD_FLGROUP;
1191 rc = obd_destroy(ec->ec_exp, oa, eco->eco_lsm,
1194 eco->eco_deleted = 1;
1195 echo_put_object(eco);
1199 case OBD_IOC_GETATTR:
1200 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1202 rc = obd_getattr(ec->ec_exp, &data->ioc_obdo1,
1204 echo_put_object(eco);
1208 case OBD_IOC_SETATTR:
1209 if (!capable (CAP_SYS_ADMIN))
1210 GOTO (out, rc = -EPERM);
1212 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1214 rc = obd_setattr(ec->ec_exp, &data->ioc_obdo1,
1215 eco->eco_lsm, NULL);
1216 echo_put_object(eco);
1220 case OBD_IOC_BRW_WRITE:
1221 if (!capable (CAP_SYS_ADMIN))
1222 GOTO (out, rc = -EPERM);
1226 case OBD_IOC_BRW_READ:
1227 rc = echo_client_brw_ioctl(rw, exp, data);
1230 case ECHO_IOC_GET_STRIPE:
1231 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1233 rc = echo_copyout_lsm(eco->eco_lsm, data->ioc_pbuf1,
1235 echo_put_object(eco);
1239 case ECHO_IOC_SET_STRIPE:
1240 if (!capable (CAP_SYS_ADMIN))
1241 GOTO (out, rc = -EPERM);
1243 if (data->ioc_pbuf1 == NULL) { /* unset */
1244 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1246 eco->eco_deleted = 1;
1247 echo_put_object(eco);
1250 rc = echo_create_object(obd, 0, &data->ioc_obdo1,
1252 data->ioc_plen1, &dummy_oti);
1256 case ECHO_IOC_ENQUEUE:
1257 if (!capable (CAP_SYS_ADMIN))
1258 GOTO (out, rc = -EPERM);
1260 rc = echo_client_enqueue(exp, &data->ioc_obdo1,
1261 data->ioc_conn1, /* lock mode */
1262 data->ioc_offset, data->ioc_count);/*extent*/
1265 case ECHO_IOC_CANCEL:
1266 rc = echo_client_cancel(exp, &data->ioc_obdo1);
1270 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
1271 GOTO (out, rc = -ENOTTY);
1277 /* XXX this should be in a helper also called by target_send_reply */
1278 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
1280 if (!ack_lock->mode)
1282 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
1289 echo_client_setup(struct obd_device *obddev, obd_count len, void *buf)
1291 struct lustre_cfg* lcfg = buf;
1292 struct echo_client_obd *ec = &obddev->u.echo_client;
1293 struct obd_device *tgt;
1294 struct lustre_handle conn = {0, };
1295 struct obd_uuid echo_uuid = { "ECHO_UUID" };
1299 if (lcfg->lcfg_inllen1 < 1) {
1300 CERROR("requires a TARGET OBD name\n");
1304 tgt = class_name2obd(lcfg->lcfg_inlbuf1);
1305 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
1306 CERROR("device not attached or not set up (%s)\n",
1307 lcfg->lcfg_inlbuf1);
1311 spin_lock_init (&ec->ec_lock);
1312 INIT_LIST_HEAD (&ec->ec_objects);
1315 rc = obd_connect(&conn, tgt, &echo_uuid);
1317 CERROR("fail to connect to device %s\n", lcfg->lcfg_inlbuf1);
1320 ec->ec_exp = class_conn2export(&conn);
1325 static int echo_client_cleanup(struct obd_device *obddev, int flags)
1327 struct list_head *el;
1328 struct ec_object *eco;
1329 struct echo_client_obd *ec = &obddev->u.echo_client;
1333 if (!list_empty(&obddev->obd_exports)) {
1334 CERROR("still has clients!\n");
1338 /* XXX assuming sole access */
1339 while (!list_empty(&ec->ec_objects)) {
1340 el = ec->ec_objects.next;
1341 eco = list_entry(el, struct ec_object, eco_obj_chain);
1343 LASSERT(eco->eco_refcount == 0);
1344 eco->eco_refcount = 1;
1345 eco->eco_deleted = 1;
1346 echo_put_object(eco);
1349 rc = obd_disconnect(ec->ec_exp, 0);
1351 CERROR("fail to disconnect device: %d\n", rc);
1356 static int echo_client_connect(struct lustre_handle *conn,
1357 struct obd_device *src, struct obd_uuid *cluuid)
1359 struct obd_export *exp;
1362 rc = class_connect(conn, src, cluuid);
1364 exp = class_conn2export(conn);
1365 INIT_LIST_HEAD(&exp->exp_ec_data.eced_locks);
1366 class_export_put(exp);
1372 static int echo_client_disconnect(struct obd_export *exp, int flags)
1374 struct obd_device *obd;
1375 struct echo_client_obd *ec;
1376 struct ec_lock *ecl;
1381 GOTO(out, rc = -EINVAL);
1384 ec = &obd->u.echo_client;
1386 /* no more contention on export's lock list */
1387 while (!list_empty (&exp->exp_ec_data.eced_locks)) {
1388 ecl = list_entry (exp->exp_ec_data.eced_locks.next,
1389 struct ec_lock, ecl_exp_chain);
1390 list_del (&ecl->ecl_exp_chain);
1392 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
1393 ecl->ecl_mode, &ecl->ecl_lock_handle);
1395 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
1396 "(%d)\n", ecl->ecl_object->eco_id, rc);
1398 echo_put_object (ecl->ecl_object);
1399 OBD_FREE (ecl, sizeof (*ecl));
1402 rc = class_disconnect(exp, 0);
1408 static struct obd_ops echo_obd_ops = {
1409 o_owner: THIS_MODULE,
1410 o_setup: echo_client_setup,
1411 o_cleanup: echo_client_cleanup,
1412 o_iocontrol: echo_client_iocontrol,
1413 o_connect: echo_client_connect,
1414 o_disconnect: echo_client_disconnect
1417 int echo_client_init(void)
1419 struct lprocfs_static_vars lvars;
1421 lprocfs_init_vars(echo, &lvars);
1422 return class_register_type(&echo_obd_ops, lvars.module_vars,
1423 OBD_ECHO_CLIENT_DEVICENAME);
1426 void echo_client_exit(void)
1428 class_unregister_type(OBD_ECHO_CLIENT_DEVICENAME);