4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2010, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdecho/echo.c
38 * Author: Peter Braam <braam@clusterfs.com>
39 * Author: Andreas Dilger <adilger@clusterfs.com>
42 #define DEBUG_SUBSYSTEM S_ECHO
44 #include <obd_support.h>
45 #include <obd_class.h>
46 #include <lustre_debug.h>
47 #include <lustre_dlm.h>
48 #include <lprocfs_status.h>
50 #include "echo_internal.h"
52 /* The echo objid needs to be below 2^32, because regular FID numbers are
53 * limited to 2^32 objects in f_oid for the FID_SEQ_ECHO range. b=23335 */
54 #define ECHO_INIT_OID 0x10000000ULL
55 #define ECHO_HANDLE_MAGIC 0xabcd0123fedc9876ULL
57 #define ECHO_PERSISTENT_PAGES (ECHO_PERSISTENT_SIZE >> PAGE_SHIFT)
58 static struct page *echo_persistent_pages[ECHO_PERSISTENT_PAGES];
61 LPROC_ECHO_READ_BYTES = 1,
62 LPROC_ECHO_WRITE_BYTES = 2,
63 LPROC_ECHO_LAST = LPROC_ECHO_WRITE_BYTES +1
66 static int echo_connect(const struct lu_env *env,
67 struct obd_export **exp, struct obd_device *obd,
68 struct obd_uuid *cluuid, struct obd_connect_data *data,
71 struct lustre_handle conn = { 0 };
74 data->ocd_connect_flags &= ECHO_CONNECT_SUPPORTED;
76 if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
77 data->ocd_connect_flags2 &= ECHO_CONNECT_SUPPORTED2;
79 rc = class_connect(&conn, obd, cluuid);
81 CERROR("can't connect %d\n", rc);
84 *exp = class_conn2export(&conn);
89 static int echo_disconnect(struct obd_export *exp)
91 LASSERT (exp != NULL);
93 return server_disconnect_export(exp);
96 static int echo_init_export(struct obd_export *exp)
98 return ldlm_init_export(exp);
101 static int echo_destroy_export(struct obd_export *exp)
105 target_destroy_export(exp);
106 ldlm_destroy_export(exp);
111 static u64 echo_next_id(struct obd_device *obddev)
115 spin_lock(&obddev->u.echo.eo_lock);
116 id = ++obddev->u.echo.eo_lastino;
117 spin_unlock(&obddev->u.echo.eo_lock);
122 static int echo_create(const struct lu_env *env, struct obd_export *exp,
125 struct obd_device *obd = class_exp2obd(exp);
128 CERROR("invalid client cookie %#llx\n",
129 exp->exp_handle.h_cookie);
133 if (!(oa->o_mode & S_IFMT)) {
134 CERROR("echo obd: no type!\n");
138 if (!(oa->o_valid & OBD_MD_FLTYPE)) {
139 CERROR("invalid o_valid %#llx\n", oa->o_valid);
143 ostid_set_seq_echo(&oa->o_oi);
144 ostid_set_id(&oa->o_oi, echo_next_id(obd));
145 oa->o_valid = OBD_MD_FLID;
150 static int echo_destroy(const struct lu_env *env, struct obd_export *exp,
153 struct obd_device *obd = class_exp2obd(exp);
157 CERROR("invalid client cookie %#llx\n",
158 exp->exp_handle.h_cookie);
162 if (!(oa->o_valid & OBD_MD_FLID)) {
163 CERROR("obdo missing FLID valid flag: %#llx\n", oa->o_valid);
167 if (ostid_id(&oa->o_oi) > obd->u.echo.eo_lastino ||
168 ostid_id(&oa->o_oi) < ECHO_INIT_OID) {
169 CERROR("bad destroy objid: "DOSTID"\n", POSTID(&oa->o_oi));
176 static int echo_getattr(const struct lu_env *env, struct obd_export *exp,
179 struct obd_device *obd = class_exp2obd(exp);
180 u64 id = ostid_id(&oa->o_oi);
184 CERROR("invalid client cookie %#llx\n",
185 exp->exp_handle.h_cookie);
189 if (!(oa->o_valid & OBD_MD_FLID)) {
190 CERROR("obdo missing FLID valid flag: %#llx\n", oa->o_valid);
194 obdo_cpy_md(oa, &obd->u.echo.eo_oa, oa->o_valid);
195 ostid_set_seq_echo(&oa->o_oi);
196 ostid_set_id(&oa->o_oi, id);
201 static int echo_setattr(const struct lu_env *env, struct obd_export *exp,
204 struct obd_device *obd = class_exp2obd(exp);
208 CERROR("invalid client cookie %#llx\n",
209 exp->exp_handle.h_cookie);
213 if (!(oa->o_valid & OBD_MD_FLID)) {
214 CERROR("obdo missing FLID valid flag: %#llx\n", oa->o_valid);
218 obd->u.echo.eo_oa = *oa;
224 echo_page_debug_setup(struct page *page, int rw, u64 id,
225 __u64 offset, int len)
227 int page_offset = offset & ~PAGE_MASK;
228 char *addr = ((char *)kmap(page)) + page_offset;
230 if (len % OBD_ECHO_BLOCK_SIZE != 0)
231 CERROR("Unexpected block size %d\n", len);
234 if (rw & OBD_BRW_READ)
235 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
238 block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE,
239 0xecc0ecc0ecc0ecc0ULL,
240 0xecc0ecc0ecc0ecc0ULL);
242 addr += OBD_ECHO_BLOCK_SIZE;
243 offset += OBD_ECHO_BLOCK_SIZE;
244 len -= OBD_ECHO_BLOCK_SIZE;
251 echo_page_debug_check(struct page *page, u64 id,
252 __u64 offset, int len)
254 int page_offset = offset & ~PAGE_MASK;
255 char *addr = ((char *)kmap(page)) + page_offset;
259 if (len % OBD_ECHO_BLOCK_SIZE != 0)
260 CERROR("Unexpected block size %d\n", len);
263 rc2 = block_debug_check("echo", addr, OBD_ECHO_BLOCK_SIZE,
266 if (rc2 != 0 && rc == 0)
269 addr += OBD_ECHO_BLOCK_SIZE;
270 offset += OBD_ECHO_BLOCK_SIZE;
271 len -= OBD_ECHO_BLOCK_SIZE;
279 static int echo_map_nb_to_lb(struct obdo *oa, struct obd_ioobj *obj,
280 struct niobuf_remote *nb, int *pages,
281 struct niobuf_local *lb, int cmd, int *left)
283 gfp_t gfp_mask = (ostid_id(&obj->ioo_oid) & 1) ?
284 GFP_HIGHUSER : GFP_KERNEL;
285 int ispersistent = ostid_id(&obj->ioo_oid) == ECHO_PERSISTENT_OBJID;
286 int debug_setup = (!ispersistent &&
287 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
288 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
289 struct niobuf_local *res = lb;
290 u64 offset = nb->rnb_offset;
291 int len = nb->rnb_len;
294 int plen = PAGE_SIZE - (offset & (PAGE_SIZE-1));
298 /* check for local buf overflow */
302 res->lnb_file_offset = offset;
304 LASSERT((res->lnb_file_offset & ~PAGE_MASK) +
305 res->lnb_len <= PAGE_SIZE);
308 ((res->lnb_file_offset >> PAGE_SHIFT) <
309 ECHO_PERSISTENT_PAGES)) {
311 echo_persistent_pages[res->lnb_file_offset >>
313 /* Take extra ref so __free_pages() can be called OK */
314 get_page(res->lnb_page);
316 res->lnb_page = alloc_page(gfp_mask);
317 if (res->lnb_page == NULL) {
318 CERROR("can't get page for id " DOSTID"\n",
319 POSTID(&obj->ioo_oid));
324 CDEBUG(D_PAGE, "$$$$ get page %p @ %llu for %d\n",
325 res->lnb_page, res->lnb_file_offset, res->lnb_len);
327 if (cmd & OBD_BRW_READ)
328 res->lnb_rc = res->lnb_len;
331 echo_page_debug_setup(res->lnb_page, cmd,
332 ostid_id(&obj->ioo_oid),
333 res->lnb_file_offset,
347 static int echo_finalize_lb(struct obdo *oa, struct obd_ioobj *obj,
348 struct niobuf_remote *rb, int *pgs,
349 struct niobuf_local *lb, int verify)
351 struct niobuf_local *res = lb;
352 u64 start = rb->rnb_offset >> PAGE_SHIFT;
353 u64 end = (rb->rnb_offset + rb->rnb_len + PAGE_SIZE - 1) >>
355 int count = (int)(end - start);
359 for (i = 0; i < count; i++, (*pgs) ++, res++) {
360 struct page *page = res->lnb_page;
364 CERROR("null page objid %llu:%p, buf %d/%d\n",
365 ostid_id(&obj->ioo_oid), page, i,
372 CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@%llu\n",
373 res->lnb_page, addr, res->lnb_file_offset);
376 int vrc = echo_page_debug_check(page,
377 ostid_id(&obj->ioo_oid),
378 res->lnb_file_offset,
380 /* check all the pages always */
381 if (vrc != 0 && rc == 0)
386 /* NB see comment above regarding persistent pages */
393 static int echo_preprw(const struct lu_env *env, int cmd,
394 struct obd_export *export, struct obdo *oa,
395 int objcount, struct obd_ioobj *obj,
396 struct niobuf_remote *nb, int *pages,
397 struct niobuf_local *res)
399 struct obd_device *obd;
405 obd = export->exp_obd;
409 /* Temp fix to stop falling foul of osc_announce_cached() */
410 oa->o_valid &= ~(OBD_MD_FLBLOCKS | OBD_MD_FLGRANT);
412 memset(res, 0, sizeof(*res) * *pages);
414 CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n",
415 cmd == OBD_BRW_READ ? "reading" : "writing", objcount, *pages);
420 for (i = 0; i < objcount; i++, obj++) {
423 for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++) {
425 rc = echo_map_nb_to_lb(oa, obj, nb, pages,
426 res + *pages, cmd, &left);
428 GOTO(preprw_cleanup, rc);
430 tot_bytes += nb->rnb_len;
434 atomic_add(*pages, &obd->u.echo.eo_prep);
436 if (cmd & OBD_BRW_READ)
437 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES,
440 lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
443 CDEBUG(D_PAGE, "%d pages allocated after prep\n",
444 atomic_read(&obd->u.echo.eo_prep));
449 /* It is possible that we would rather handle errors by allow
450 * any already-set-up pages to complete, rather than tearing them
451 * all down again. I believe that this is what the in-kernel
452 * prep/commit operations do.
454 CERROR("cleaning up %u pages (%d obdos)\n", *pages, objcount);
455 for (i = 0; i < *pages; i++) {
456 kunmap(res[i].lnb_page);
457 /* NB if this is a persistent page, __free_page() will just
458 * lose the extra ref gained above */
459 __free_page(res[i].lnb_page);
460 res[i].lnb_page = NULL;
461 atomic_dec(&obd->u.echo.eo_prep);
467 static int echo_commitrw(const struct lu_env *env, int cmd,
468 struct obd_export *export, struct obdo *oa,
469 int objcount, struct obd_ioobj *obj,
470 struct niobuf_remote *rb, int niocount,
471 struct niobuf_local *res, int rc)
473 struct obd_device *obd;
478 obd = export->exp_obd;
483 GOTO(commitrw_cleanup, rc);
485 if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) {
486 CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n",
489 CDEBUG(D_PAGE, "writing %d obdos with %d IOs\n",
493 if (niocount && res == NULL) {
494 CERROR("NULL res niobuf with niocount %d\n", niocount);
498 for (i = 0; i < objcount; i++, obj++) {
499 int verify = (rc == 0 &&
500 ostid_id(&obj->ioo_oid) != ECHO_PERSISTENT_OBJID &&
501 (oa->o_valid & OBD_MD_FLFLAGS) != 0 &&
502 (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0);
505 for (j = 0 ; j < obj->ioo_bufcnt ; j++, rb++) {
506 int vrc = echo_finalize_lb(oa, obj, rb, &pgs, &res[pgs],
512 GOTO(commitrw_cleanup, rc = vrc);
520 atomic_sub(pgs, &obd->u.echo.eo_prep);
522 CDEBUG(D_PAGE, "%d pages remain after commit\n",
523 atomic_read(&obd->u.echo.eo_prep));
527 atomic_sub(pgs, &obd->u.echo.eo_prep);
529 CERROR("cleaning up %d pages (%d obdos)\n",
530 niocount - pgs - 1, objcount);
532 while (pgs < niocount) {
533 struct page *page = res[pgs++].lnb_page;
538 /* NB see comment above regarding persistent pages */
540 atomic_dec(&obd->u.echo.eo_prep);
545 LPROC_SEQ_FOPS_RO_TYPE(echo, uuid);
546 static struct lprocfs_vars lprocfs_echo_obd_vars[] = {
548 .fops = &echo_uuid_fops },
552 static int echo_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
555 __u64 lock_flags = 0;
556 struct ldlm_res_id res_id = {.name = {1}};
560 obd->u.echo.eo_obt.obt_magic = OBT_MAGIC;
561 spin_lock_init(&obd->u.echo.eo_lock);
562 obd->u.echo.eo_lastino = ECHO_INIT_OID;
564 sprintf(ns_name, "echotgt-%s", obd->obd_uuid.uuid);
565 obd->obd_namespace = ldlm_namespace_new(obd, ns_name,
566 LDLM_NAMESPACE_SERVER,
567 LDLM_NAMESPACE_MODEST,
569 if (obd->obd_namespace == NULL) {
574 rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id, LDLM_PLAIN,
575 NULL, LCK_NL, &lock_flags, NULL,
576 ldlm_completion_ast, NULL, NULL, 0,
577 LVB_T_NONE, NULL, &obd->u.echo.eo_nl_lock);
578 LASSERT (rc == ELDLM_OK);
580 obd->obd_vars = lprocfs_echo_obd_vars;
581 if (lprocfs_obd_setup(obd) == 0 &&
582 lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST) == 0) {
583 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES,
584 LPROCFS_CNTR_AVGMINMAX,
585 "read_bytes", "bytes");
586 lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES,
587 LPROCFS_CNTR_AVGMINMAX,
588 "write_bytes", "bytes");
591 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
592 "echo_ldlm_cb_client", &obd->obd_ldlm_client);
596 static int echo_cleanup(struct obd_device *obd)
601 lprocfs_obd_cleanup(obd);
602 lprocfs_free_obd_stats(obd);
604 ldlm_lock_decref(&obd->u.echo.eo_nl_lock, LCK_NL);
606 /* XXX Bug 3413; wait for a bit to ensure the BL callback has
607 * happened before calling ldlm_namespace_free() */
608 set_current_state(TASK_UNINTERRUPTIBLE);
609 schedule_timeout(cfs_time_seconds(1));
611 ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force);
612 obd->obd_namespace = NULL;
614 leaked = atomic_read(&obd->u.echo.eo_prep);
616 CERROR("%d prep/commitrw pages leaked\n", leaked);
621 struct obd_ops echo_obd_ops = {
622 .o_owner = THIS_MODULE,
623 .o_connect = echo_connect,
624 .o_disconnect = echo_disconnect,
625 .o_init_export = echo_init_export,
626 .o_destroy_export = echo_destroy_export,
627 .o_create = echo_create,
628 .o_destroy = echo_destroy,
629 .o_getattr = echo_getattr,
630 .o_setattr = echo_setattr,
631 .o_preprw = echo_preprw,
632 .o_commitrw = echo_commitrw,
633 .o_setup = echo_setup,
634 .o_cleanup = echo_cleanup
637 void echo_persistent_pages_fini(void)
641 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++)
642 if (echo_persistent_pages[i] != NULL) {
643 __free_page(echo_persistent_pages[i]);
644 echo_persistent_pages[i] = NULL;
648 int echo_persistent_pages_init(void)
653 for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) {
654 gfp_t gfp_mask = (i < ECHO_PERSISTENT_PAGES/2) ?
655 GFP_KERNEL : GFP_HIGHUSER;
657 pg = alloc_page(gfp_mask);
659 echo_persistent_pages_fini();
663 memset(kmap(pg), 0, PAGE_SIZE);
666 echo_persistent_pages[i] = pg;