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>
34 #include <sys/capability.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.echo_client;
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.echo_client;
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.echo_client;
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.echo_client;
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.echo_client;
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 /* XXX get some filter group constants */
224 oa->o_valid |= OBD_MD_FLGROUP;
225 rc = obd_create(ec->ec_exp, oa, &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.echo_client;
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.echo_client;
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);
492 static int echo_client_kbrw(struct obd_device *obd, int rw, struct obdo *oa,
493 struct lov_stripe_md *lsm, obd_off offset,
494 obd_size count, struct obd_trans_info *oti)
496 struct echo_client_obd *ec = &obd->u.echo_client;
498 struct brw_page *pga;
499 struct brw_page *pgp;
506 /* oa_id == ECHO_PERSISTENT_OBJID => speed test (no verification).
507 * oa & 1 => use HIGHMEM */
509 verify = (oa->o_id) != ECHO_PERSISTENT_OBJID;
510 gfp_mask = ((oa->o_id & 2) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
512 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
515 (count & (PAGE_SIZE - 1)) != 0 ||
517 lsm->lsm_object_id != oa->o_id))
520 /* XXX think again with misaligned I/O */
521 npages = count >> PAGE_SHIFT;
523 OBD_ALLOC(pga, npages * sizeof(*pga));
527 for (i = 0, pgp = pga, off = offset;
529 i++, pgp++, off += PAGE_SIZE) {
531 LASSERT (pgp->pg == NULL); /* for cleanup */
534 pgp->pg = alloc_pages (gfp_mask, 0);
538 pgp->count = PAGE_SIZE;
543 echo_client_page_debug_setup(lsm, pgp->pg, rw,
544 oa->o_id, off, pgp->count);
547 rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti);
550 if (rc != 0 || rw != OBD_BRW_READ)
553 for (i = 0, pgp = pga; i < npages; i++, pgp++) {
559 vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id,
560 pgp->off, pgp->count);
561 if (vrc != 0 && rc == 0)
564 __free_pages(pgp->pg, 0);
566 OBD_FREE(pga, npages * sizeof(*pga));
571 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
572 static int echo_client_ubrw(struct obd_device *obd, int rw,
573 struct obdo *oa, struct lov_stripe_md *lsm,
574 obd_off offset, obd_size count, char *buffer,
575 struct obd_trans_info *oti)
577 struct echo_client_obd *ec = &obd->u.echo_client;
579 struct brw_page *pga;
580 struct brw_page *pgp;
582 struct kiobuf *kiobuf;
586 LASSERT (rw == OBD_BRW_WRITE ||
589 /* NB: for now, only whole pages, page aligned */
592 ((long)buffer & (PAGE_SIZE - 1)) != 0 ||
593 (count & (PAGE_SIZE - 1)) != 0 ||
594 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
597 /* XXX think again with misaligned I/O */
598 npages = count >> PAGE_SHIFT;
600 OBD_ALLOC(pga, npages * sizeof(*pga));
604 rc = alloc_kiovec (1, &kiobuf);
608 rc = map_user_kiobuf ((rw == OBD_BRW_READ) ? READ : WRITE,
609 kiobuf, (unsigned long)buffer, count);
613 LASSERT (kiobuf->offset == 0);
614 LASSERT (kiobuf->nr_pages == npages);
616 for (i = 0, off = offset, pgp = pga;
618 i++, off += PAGE_SIZE, pgp++) {
620 pgp->pg = kiobuf->maplist[i];
621 pgp->count = PAGE_SIZE;
625 rc = obd_brw(rw, ec->ec_exp, oa, lsm, npages, pga, oti);
627 // if (rw == OBD_BRW_READ)
628 // mark_dirty_kiobuf (kiobuf, count);
630 unmap_kiobuf (kiobuf);
632 free_kiovec (1, &kiobuf);
634 OBD_FREE(pga, npages * sizeof(*pga));
638 static int echo_client_ubrw(struct obd_device *obd, int rw,
639 struct obdo *oa, struct lov_stripe_md *lsm,
640 obd_off offset, obd_size count, char *buffer,
641 struct obd_trans_info *oti)
643 #warning "echo_client_ubrw() needs to be ported on 2.6 yet"
650 struct echo_async_state;
652 #define EAP_MAGIC 79277927
653 struct echo_async_page {
655 struct page *eap_page;
658 struct echo_async_state *eap_eas;
659 struct list_head eap_item;
662 struct echo_async_state {
664 obd_off eas_next_offset;
665 obd_off eas_end_offset;
668 wait_queue_head_t eas_waitq;
669 struct list_head eas_avail;
671 struct lov_stripe_md *eas_lsm;
674 static int eas_should_wake(struct echo_async_state *eas)
678 spin_lock_irqsave(&eas->eas_lock, flags);
679 if (eas->eas_rc == 0 && !list_empty(&eas->eas_avail))
681 spin_unlock_irqrestore(&eas->eas_lock, flags);
685 struct echo_async_page *eap_from_cookie(void *cookie)
687 struct echo_async_page *eap = cookie;
688 if (eap->eap_magic != EAP_MAGIC)
689 return ERR_PTR(-EINVAL);
693 static int ec_ap_make_ready(void *data, int cmd)
695 /* our pages are issued ready */
699 static int ec_ap_refresh_count(void *data, int cmd)
701 /* our pages are issued with a stable count */
705 static void ec_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
707 struct echo_async_page *eap;
708 eap = eap_from_cookie(data);
712 memcpy(oa, &eap->eap_eas->eas_oa, sizeof(*oa));
715 static void ec_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
717 struct echo_async_page *eap = eap_from_cookie(data);
718 struct echo_async_state *eas;
725 if (cmd == OBD_BRW_READ &&
726 eas->eas_oa.o_id != ECHO_PERSISTENT_OBJID)
727 echo_client_page_debug_check(eas->eas_lsm, eap->eap_page,
728 eas->eas_oa.o_id, eap->eap_off,
731 spin_lock_irqsave(&eas->eas_lock, flags);
732 if (rc && !eas->eas_rc)
734 eas->eas_in_flight--;
735 list_add(&eap->eap_item, &eas->eas_avail);
736 wake_up(&eas->eas_waitq);
737 spin_unlock_irqrestore(&eas->eas_lock, flags);
740 static struct obd_async_page_ops ec_async_page_ops = {
741 .ap_make_ready = ec_ap_make_ready,
742 .ap_refresh_count = ec_ap_refresh_count,
743 .ap_fill_obdo = ec_ap_fill_obdo,
744 .ap_completion = ec_ap_completion,
747 static int echo_client_async_page(struct obd_export *exp, int rw,
748 struct obdo *oa, struct lov_stripe_md *lsm,
749 obd_off offset, obd_size count,
753 struct echo_async_page *eap;
754 struct echo_async_state eas;
755 struct list_head *pos, *n;
762 /* oa_id == 0 => speed test (no verification) else...
763 * oa & 1 => use HIGHMEM
765 verify = (oa->o_id != 0);
766 gfp_mask = ((oa->o_id & 1) == 0) ? GFP_KERNEL : GFP_HIGHUSER;
769 LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ);
772 (count & (PAGE_SIZE - 1)) != 0 ||
774 lsm->lsm_object_id != oa->o_id))
777 /* XXX think again with misaligned I/O */
778 npages = batching >> PAGE_SHIFT;
780 memcpy(&eas.eas_oa, oa, sizeof(*oa));
781 eas.eas_next_offset = offset;
782 eas.eas_end_offset = offset + count;
783 spin_lock_init(&eas.eas_lock);
784 init_waitqueue_head(&eas.eas_waitq);
785 eas.eas_in_flight = 0;
788 INIT_LIST_HEAD(&eas.eas_avail);
790 /* prepare the group of pages that we're going to be keeping
792 for (i = 0; i < npages; i++) {
793 struct page *page = alloc_page(GFP_KERNEL);
795 GOTO(out, rc = -ENOMEM);
798 list_add_tail(&PAGE_LIST(page), &pages);
800 OBD_ALLOC(eap, sizeof(*eap));
802 GOTO(out, rc = -ENOMEM);
804 eap->eap_magic = EAP_MAGIC;
805 eap->eap_page = page;
807 page->private = (unsigned long)eap;
808 list_add_tail(&eap->eap_item, &eas.eas_avail);
811 /* first we spin queueing io and being woken by its completion */
812 spin_lock_irqsave(&eas.eas_lock, flags);
816 /* sleep until we have a page to send */
817 spin_unlock_irqrestore(&eas.eas_lock, flags);
818 rc = wait_event_interruptible(eas.eas_waitq,
819 eas_should_wake(&eas));
820 spin_lock_irqsave(&eas.eas_lock, flags);
821 if (rc && !eas.eas_rc)
825 if (list_empty(&eas.eas_avail))
827 eap = list_entry(eas.eas_avail.next, struct echo_async_page,
829 list_del(&eap->eap_item);
830 spin_unlock_irqrestore(&eas.eas_lock, flags);
832 /* unbind the eap from its old page offset */
833 if (eap->eap_cookie != NULL) {
834 obd_teardown_async_page(exp, lsm, NULL,
836 eap->eap_cookie = NULL;
839 eas.eas_next_offset += PAGE_SIZE;
840 eap->eap_off = eas.eas_next_offset;
842 rc = obd_prep_async_page(exp, lsm, NULL, eap->eap_page,
843 eap->eap_off, &ec_async_page_ops,
844 eap, &eap->eap_cookie);
846 spin_lock_irqsave(&eas.eas_lock, flags);
851 if (oa->o_id != ECHO_PERSISTENT_OBJID)
852 echo_client_page_debug_setup(lsm, eap->eap_page, rw,
854 eap->eap_off, PAGE_SIZE);
856 /* always asserts urgent, which isn't quite right */
857 rc = obd_queue_async_io(exp, lsm, NULL, eap->eap_cookie,
859 ASYNC_READY | ASYNC_URGENT |
861 spin_lock_irqsave(&eas.eas_lock, flags);
862 if (rc && !eas.eas_rc) {
867 if (eas.eas_next_offset == eas.eas_end_offset)
871 /* still hold the eas_lock here.. */
873 /* now we just spin waiting for all the rpcs to complete */
874 while(eas.eas_in_flight) {
875 spin_unlock_irqrestore(&eas.eas_lock, flags);
876 wait_event_interruptible(eas.eas_waitq,
877 eas.eas_in_flight == 0);
878 spin_lock_irqsave(&eas.eas_lock, flags);
880 spin_unlock_irqrestore(&eas.eas_lock, flags);
883 list_for_each_safe(pos, n, &pages) {
884 struct page *page = list_entry(pos, struct page,
887 list_del(&PAGE_LIST(page));
888 if (page->private != 0) {
889 eap = (struct echo_async_page *)page->private;
890 if (eap->eap_cookie != NULL)
891 obd_teardown_async_page(exp, lsm, NULL,
893 OBD_FREE(eap, sizeof(*eap));
901 static int echo_client_prep_commit(struct obd_export *exp, int rw,
902 struct obdo *oa, struct lov_stripe_md *lsm,
903 obd_off offset, obd_size count,
904 obd_size batch, struct obd_trans_info *oti)
906 struct obd_ioobj ioo;
907 struct niobuf_local *lnb;
908 struct niobuf_remote *rnb;
910 obd_size npages, tot_pages;
914 if (count <= 0 || (count & (PAGE_SIZE - 1)) != 0 ||
915 (lsm != NULL && lsm->lsm_object_id != oa->o_id))
918 npages = batch >> PAGE_SHIFT;
919 tot_pages = count >> PAGE_SHIFT;
921 OBD_ALLOC(lnb, npages * sizeof(struct niobuf_local));
922 OBD_ALLOC(rnb, npages * sizeof(struct niobuf_remote));
924 if (lnb == NULL || rnb == NULL)
925 GOTO(out, ret = -ENOMEM);
927 obdo_to_ioobj(oa, &ioo);
931 for(; tot_pages; tot_pages -= npages) {
932 if (tot_pages < npages)
935 for (i = 0; i < npages; i++, off += PAGE_SIZE) {
937 rnb[i].len = PAGE_SIZE;
940 /* XXX this can't be the best.. */
941 memset(oti, 0, sizeof(*oti));
942 ioo.ioo_bufcnt = npages;
944 ret = obd_preprw(rw, exp, oa, 1, &ioo, npages, rnb, lnb, oti);
948 for (i = 0; i < npages; i++) {
949 struct page *page = lnb[i].page;
952 if (page == NULL && lnb[i].rc == 0)
955 if (oa->o_id == ECHO_PERSISTENT_OBJID)
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.echo_client;
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 = 2;
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.echo_client);
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.echo_client;
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.echo_client;
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;
1198 memset(&dummy_oti, 0, sizeof(dummy_oti));
1201 ec = &obd->u.echo_client;
1204 case OBD_IOC_CREATE: /* may create echo object */
1205 if (!capable (CAP_SYS_ADMIN))
1206 GOTO (out, rc = -EPERM);
1208 rc = echo_create_object (obd, 1, &data->ioc_obdo1,
1209 data->ioc_pbuf1, data->ioc_plen1,
1213 case OBD_IOC_DESTROY:
1214 if (!capable (CAP_SYS_ADMIN))
1215 GOTO (out, rc = -EPERM);
1217 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1219 oa = &data->ioc_obdo1;
1221 oa->o_valid |= OBD_MD_FLGROUP;
1222 rc = obd_destroy(ec->ec_exp, oa, eco->eco_lsm,
1225 eco->eco_deleted = 1;
1226 echo_put_object(eco);
1230 case OBD_IOC_GETATTR:
1231 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1233 rc = obd_getattr(ec->ec_exp, &data->ioc_obdo1,
1235 echo_put_object(eco);
1239 case OBD_IOC_SETATTR:
1240 if (!capable (CAP_SYS_ADMIN))
1241 GOTO (out, rc = -EPERM);
1243 rc = echo_get_object (&eco, obd, &data->ioc_obdo1);
1245 rc = obd_setattr(ec->ec_exp, &data->ioc_obdo1,
1246 eco->eco_lsm, NULL);
1247 echo_put_object(eco);
1251 case OBD_IOC_BRW_WRITE:
1252 if (!capable (CAP_SYS_ADMIN))
1253 GOTO (out, rc = -EPERM);
1257 case OBD_IOC_BRW_READ:
1258 rc = echo_client_brw_ioctl(rw, exp, data);
1261 case ECHO_IOC_GET_STRIPE:
1262 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1264 rc = echo_copyout_lsm(eco->eco_lsm, data->ioc_pbuf1,
1266 echo_put_object(eco);
1270 case ECHO_IOC_SET_STRIPE:
1271 if (!capable (CAP_SYS_ADMIN))
1272 GOTO (out, rc = -EPERM);
1274 if (data->ioc_pbuf1 == NULL) { /* unset */
1275 rc = echo_get_object(&eco, obd, &data->ioc_obdo1);
1277 eco->eco_deleted = 1;
1278 echo_put_object(eco);
1281 rc = echo_create_object(obd, 0, &data->ioc_obdo1,
1283 data->ioc_plen1, &dummy_oti);
1287 case ECHO_IOC_ENQUEUE:
1288 if (!capable (CAP_SYS_ADMIN))
1289 GOTO (out, rc = -EPERM);
1291 rc = echo_client_enqueue(exp, &data->ioc_obdo1,
1292 data->ioc_conn1, /* lock mode */
1293 data->ioc_offset, data->ioc_count);/*extent*/
1296 case ECHO_IOC_CANCEL:
1297 rc = echo_client_cancel(exp, &data->ioc_obdo1);
1301 CERROR ("echo_ioctl(): unrecognised ioctl %#x\n", cmd);
1302 GOTO (out, rc = -ENOTTY);
1308 /* XXX this should be in a helper also called by target_send_reply */
1309 for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4;
1311 if (!ack_lock->mode)
1313 ldlm_lock_decref(&ack_lock->lock, ack_lock->mode);
1320 echo_client_setup(struct obd_device *obddev, obd_count len, void *buf)
1322 struct lustre_cfg* lcfg = buf;
1323 struct echo_client_obd *ec = &obddev->u.echo_client;
1324 struct obd_device *tgt;
1325 struct lustre_handle conn = {0, };
1326 struct obd_uuid echo_uuid = { "ECHO_UUID" };
1330 if (lcfg->lcfg_inllen1 < 1) {
1331 CERROR("requires a TARGET OBD name\n");
1335 tgt = class_name2obd(lcfg->lcfg_inlbuf1);
1336 if (!tgt || !tgt->obd_attached || !tgt->obd_set_up) {
1337 CERROR("device not attached or not set up (%s)\n",
1338 lcfg->lcfg_inlbuf1);
1342 spin_lock_init (&ec->ec_lock);
1343 INIT_LIST_HEAD (&ec->ec_objects);
1346 rc = obd_connect(&conn, tgt, &echo_uuid);
1348 CERROR("fail to connect to device %s\n", lcfg->lcfg_inlbuf1);
1351 ec->ec_exp = class_conn2export(&conn);
1356 static int echo_client_cleanup(struct obd_device *obddev, int flags)
1358 struct list_head *el;
1359 struct ec_object *eco;
1360 struct echo_client_obd *ec = &obddev->u.echo_client;
1364 if (!list_empty(&obddev->obd_exports)) {
1365 CERROR("still has clients!\n");
1369 /* XXX assuming sole access */
1370 while (!list_empty(&ec->ec_objects)) {
1371 el = ec->ec_objects.next;
1372 eco = list_entry(el, struct ec_object, eco_obj_chain);
1374 LASSERT(eco->eco_refcount == 0);
1375 eco->eco_refcount = 1;
1376 eco->eco_deleted = 1;
1377 echo_put_object(eco);
1380 rc = obd_disconnect(ec->ec_exp, 0);
1382 CERROR("fail to disconnect device: %d\n", rc);
1387 static int echo_client_connect(struct lustre_handle *conn,
1388 struct obd_device *src, struct obd_uuid *cluuid)
1390 struct obd_export *exp;
1393 rc = class_connect(conn, src, cluuid);
1395 exp = class_conn2export(conn);
1396 INIT_LIST_HEAD(&exp->exp_ec_data.eced_locks);
1397 class_export_put(exp);
1403 static int echo_client_disconnect(struct obd_export *exp, int flags)
1405 struct obd_device *obd;
1406 struct echo_client_obd *ec;
1407 struct ec_lock *ecl;
1412 GOTO(out, rc = -EINVAL);
1415 ec = &obd->u.echo_client;
1417 /* no more contention on export's lock list */
1418 while (!list_empty (&exp->exp_ec_data.eced_locks)) {
1419 ecl = list_entry (exp->exp_ec_data.eced_locks.next,
1420 struct ec_lock, ecl_exp_chain);
1421 list_del (&ecl->ecl_exp_chain);
1423 rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm,
1424 ecl->ecl_mode, &ecl->ecl_lock_handle);
1426 CDEBUG (D_INFO, "Cancel lock on object "LPX64" on disconnect "
1427 "(%d)\n", ecl->ecl_object->eco_id, rc);
1429 echo_put_object (ecl->ecl_object);
1430 OBD_FREE (ecl, sizeof (*ecl));
1433 rc = class_disconnect(exp, 0);
1439 static struct obd_ops echo_obd_ops = {
1440 .o_owner = THIS_MODULE,
1441 .o_setup = echo_client_setup,
1442 .o_cleanup = echo_client_cleanup,
1443 .o_iocontrol = echo_client_iocontrol,
1444 .o_connect = echo_client_connect,
1445 .o_disconnect = echo_client_disconnect
1448 int echo_client_init(void)
1450 struct lprocfs_static_vars lvars;
1452 lprocfs_init_vars(echo, &lvars);
1453 return class_register_type(&echo_obd_ops, lvars.module_vars,
1454 OBD_ECHO_CLIENT_DEVICENAME);
1457 void echo_client_exit(void)
1459 class_unregister_type(OBD_ECHO_CLIENT_DEVICENAME);