]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - drivers/staging/lustre/lustre/lov/lov_io.c
staging: board: Migrate away from __pm_genpd_name_add_device()
[mirror_ubuntu-zesty-kernel.git] / drivers / staging / lustre / lustre / lov / lov_io.c
1 /*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
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
19 *
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
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26 /*
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 *
36 * Implementation of cl_io for LOV layer.
37 *
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 * Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
40 */
41
42 #define DEBUG_SUBSYSTEM S_LOV
43
44 #include "lov_cl_internal.h"
45
46 /** \addtogroup lov
47 * @{
48 */
49
50 static inline void lov_sub_enter(struct lov_io_sub *sub)
51 {
52 sub->sub_reenter++;
53 }
54 static inline void lov_sub_exit(struct lov_io_sub *sub)
55 {
56 sub->sub_reenter--;
57 }
58
59 static void lov_io_sub_fini(const struct lu_env *env, struct lov_io *lio,
60 struct lov_io_sub *sub)
61 {
62 if (sub->sub_io != NULL) {
63 if (sub->sub_io_initialized) {
64 lov_sub_enter(sub);
65 cl_io_fini(sub->sub_env, sub->sub_io);
66 lov_sub_exit(sub);
67 sub->sub_io_initialized = 0;
68 lio->lis_active_subios--;
69 }
70 if (sub->sub_stripe == lio->lis_single_subio_index)
71 lio->lis_single_subio_index = -1;
72 else if (!sub->sub_borrowed)
73 kfree(sub->sub_io);
74 sub->sub_io = NULL;
75 }
76 if (sub->sub_env != NULL && !IS_ERR(sub->sub_env)) {
77 if (!sub->sub_borrowed)
78 cl_env_put(sub->sub_env, &sub->sub_refcheck);
79 sub->sub_env = NULL;
80 }
81 }
82
83 static void lov_io_sub_inherit(struct cl_io *io, struct lov_io *lio,
84 int stripe, loff_t start, loff_t end)
85 {
86 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
87 struct cl_io *parent = lio->lis_cl.cis_io;
88
89 switch (io->ci_type) {
90 case CIT_SETATTR: {
91 io->u.ci_setattr.sa_attr = parent->u.ci_setattr.sa_attr;
92 io->u.ci_setattr.sa_valid = parent->u.ci_setattr.sa_valid;
93 io->u.ci_setattr.sa_capa = parent->u.ci_setattr.sa_capa;
94 if (cl_io_is_trunc(io)) {
95 loff_t new_size = parent->u.ci_setattr.sa_attr.lvb_size;
96
97 new_size = lov_size_to_stripe(lsm, new_size, stripe);
98 io->u.ci_setattr.sa_attr.lvb_size = new_size;
99 }
100 break;
101 }
102 case CIT_FAULT: {
103 struct cl_object *obj = parent->ci_obj;
104 loff_t off = cl_offset(obj, parent->u.ci_fault.ft_index);
105
106 io->u.ci_fault = parent->u.ci_fault;
107 off = lov_size_to_stripe(lsm, off, stripe);
108 io->u.ci_fault.ft_index = cl_index(obj, off);
109 break;
110 }
111 case CIT_FSYNC: {
112 io->u.ci_fsync.fi_start = start;
113 io->u.ci_fsync.fi_end = end;
114 io->u.ci_fsync.fi_capa = parent->u.ci_fsync.fi_capa;
115 io->u.ci_fsync.fi_fid = parent->u.ci_fsync.fi_fid;
116 io->u.ci_fsync.fi_mode = parent->u.ci_fsync.fi_mode;
117 break;
118 }
119 case CIT_READ:
120 case CIT_WRITE: {
121 io->u.ci_wr.wr_sync = cl_io_is_sync_write(parent);
122 if (cl_io_is_append(parent)) {
123 io->u.ci_wr.wr_append = 1;
124 } else {
125 io->u.ci_rw.crw_pos = start;
126 io->u.ci_rw.crw_count = end - start;
127 }
128 break;
129 }
130 default:
131 break;
132 }
133 }
134
135 static int lov_io_sub_init(const struct lu_env *env, struct lov_io *lio,
136 struct lov_io_sub *sub)
137 {
138 struct lov_object *lov = lio->lis_object;
139 struct lov_device *ld = lu2lov_dev(lov2cl(lov)->co_lu.lo_dev);
140 struct cl_io *sub_io;
141 struct cl_object *sub_obj;
142 struct cl_io *io = lio->lis_cl.cis_io;
143
144 int stripe = sub->sub_stripe;
145 int result;
146
147 LASSERT(sub->sub_io == NULL);
148 LASSERT(sub->sub_env == NULL);
149 LASSERT(sub->sub_stripe < lio->lis_stripe_count);
150
151 if (unlikely(lov_r0(lov)->lo_sub[stripe] == NULL))
152 return -EIO;
153
154 result = 0;
155 sub->sub_io_initialized = 0;
156 sub->sub_borrowed = 0;
157
158 if (lio->lis_mem_frozen) {
159 LASSERT(mutex_is_locked(&ld->ld_mutex));
160 sub->sub_io = &ld->ld_emrg[stripe]->emrg_subio;
161 sub->sub_env = ld->ld_emrg[stripe]->emrg_env;
162 sub->sub_borrowed = 1;
163 } else {
164 void *cookie;
165
166 /* obtain new environment */
167 cookie = cl_env_reenter();
168 sub->sub_env = cl_env_get(&sub->sub_refcheck);
169 cl_env_reexit(cookie);
170 if (IS_ERR(sub->sub_env))
171 result = PTR_ERR(sub->sub_env);
172
173 if (result == 0) {
174 /*
175 * First sub-io. Use ->lis_single_subio to
176 * avoid dynamic allocation.
177 */
178 if (lio->lis_active_subios == 0) {
179 sub->sub_io = &lio->lis_single_subio;
180 lio->lis_single_subio_index = stripe;
181 } else {
182 sub->sub_io = kzalloc(sizeof(*sub->sub_io),
183 GFP_NOFS);
184 if (sub->sub_io == NULL)
185 result = -ENOMEM;
186 }
187 }
188 }
189
190 if (result == 0) {
191 sub_obj = lovsub2cl(lov_r0(lov)->lo_sub[stripe]);
192 sub_io = sub->sub_io;
193
194 sub_io->ci_obj = sub_obj;
195 sub_io->ci_result = 0;
196
197 sub_io->ci_parent = io;
198 sub_io->ci_lockreq = io->ci_lockreq;
199 sub_io->ci_type = io->ci_type;
200 sub_io->ci_no_srvlock = io->ci_no_srvlock;
201 sub_io->ci_noatime = io->ci_noatime;
202
203 lov_sub_enter(sub);
204 result = cl_io_sub_init(sub->sub_env, sub_io,
205 io->ci_type, sub_obj);
206 lov_sub_exit(sub);
207 if (result >= 0) {
208 lio->lis_active_subios++;
209 sub->sub_io_initialized = 1;
210 result = 0;
211 }
212 }
213 if (result != 0)
214 lov_io_sub_fini(env, lio, sub);
215 return result;
216 }
217
218 struct lov_io_sub *lov_sub_get(const struct lu_env *env,
219 struct lov_io *lio, int stripe)
220 {
221 int rc;
222 struct lov_io_sub *sub = &lio->lis_subs[stripe];
223
224 LASSERT(stripe < lio->lis_stripe_count);
225
226 if (!sub->sub_io_initialized) {
227 sub->sub_stripe = stripe;
228 rc = lov_io_sub_init(env, lio, sub);
229 } else
230 rc = 0;
231 if (rc == 0)
232 lov_sub_enter(sub);
233 else
234 sub = ERR_PTR(rc);
235 return sub;
236 }
237
238 void lov_sub_put(struct lov_io_sub *sub)
239 {
240 lov_sub_exit(sub);
241 }
242
243 /*****************************************************************************
244 *
245 * Lov io operations.
246 *
247 */
248
249 static int lov_page_stripe(const struct cl_page *page)
250 {
251 struct lovsub_object *subobj;
252
253 subobj = lu2lovsub(
254 lu_object_locate(page->cp_child->cp_obj->co_lu.lo_header,
255 &lovsub_device_type));
256 LASSERT(subobj != NULL);
257 return subobj->lso_index;
258 }
259
260 struct lov_io_sub *lov_page_subio(const struct lu_env *env, struct lov_io *lio,
261 const struct cl_page_slice *slice)
262 {
263 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
264 struct cl_page *page = slice->cpl_page;
265 int stripe;
266
267 LASSERT(lio->lis_cl.cis_io != NULL);
268 LASSERT(cl2lov(slice->cpl_obj) == lio->lis_object);
269 LASSERT(lsm != NULL);
270 LASSERT(lio->lis_nr_subios > 0);
271
272 stripe = lov_page_stripe(page);
273 return lov_sub_get(env, lio, stripe);
274 }
275
276
277 static int lov_io_subio_init(const struct lu_env *env, struct lov_io *lio,
278 struct cl_io *io)
279 {
280 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
281 int result;
282
283 LASSERT(lio->lis_object != NULL);
284
285 /*
286 * Need to be optimized, we can't afford to allocate a piece of memory
287 * when writing a page. -jay
288 */
289 lio->lis_subs =
290 libcfs_kvzalloc(lsm->lsm_stripe_count *
291 sizeof(lio->lis_subs[0]),
292 GFP_NOFS);
293 if (lio->lis_subs != NULL) {
294 lio->lis_nr_subios = lio->lis_stripe_count;
295 lio->lis_single_subio_index = -1;
296 lio->lis_active_subios = 0;
297 result = 0;
298 } else
299 result = -ENOMEM;
300 return result;
301 }
302
303 static void lov_io_slice_init(struct lov_io *lio,
304 struct lov_object *obj, struct cl_io *io)
305 {
306 io->ci_result = 0;
307 lio->lis_object = obj;
308
309 LASSERT(obj->lo_lsm != NULL);
310 lio->lis_stripe_count = obj->lo_lsm->lsm_stripe_count;
311
312 switch (io->ci_type) {
313 case CIT_READ:
314 case CIT_WRITE:
315 lio->lis_pos = io->u.ci_rw.crw_pos;
316 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
317 lio->lis_io_endpos = lio->lis_endpos;
318 if (cl_io_is_append(io)) {
319 LASSERT(io->ci_type == CIT_WRITE);
320 lio->lis_pos = 0;
321 lio->lis_endpos = OBD_OBJECT_EOF;
322 }
323 break;
324
325 case CIT_SETATTR:
326 if (cl_io_is_trunc(io))
327 lio->lis_pos = io->u.ci_setattr.sa_attr.lvb_size;
328 else
329 lio->lis_pos = 0;
330 lio->lis_endpos = OBD_OBJECT_EOF;
331 break;
332
333 case CIT_FAULT: {
334 pgoff_t index = io->u.ci_fault.ft_index;
335 lio->lis_pos = cl_offset(io->ci_obj, index);
336 lio->lis_endpos = cl_offset(io->ci_obj, index + 1);
337 break;
338 }
339
340 case CIT_FSYNC: {
341 lio->lis_pos = io->u.ci_fsync.fi_start;
342 lio->lis_endpos = io->u.ci_fsync.fi_end;
343 break;
344 }
345
346 case CIT_MISC:
347 lio->lis_pos = 0;
348 lio->lis_endpos = OBD_OBJECT_EOF;
349 break;
350
351 default:
352 LBUG();
353 }
354 }
355
356 static void lov_io_fini(const struct lu_env *env, const struct cl_io_slice *ios)
357 {
358 struct lov_io *lio = cl2lov_io(env, ios);
359 struct lov_object *lov = cl2lov(ios->cis_obj);
360 int i;
361
362 if (lio->lis_subs != NULL) {
363 for (i = 0; i < lio->lis_nr_subios; i++)
364 lov_io_sub_fini(env, lio, &lio->lis_subs[i]);
365 kvfree(lio->lis_subs);
366 lio->lis_nr_subios = 0;
367 }
368
369 LASSERT(atomic_read(&lov->lo_active_ios) > 0);
370 if (atomic_dec_and_test(&lov->lo_active_ios))
371 wake_up_all(&lov->lo_waitq);
372 }
373
374 static u64 lov_offset_mod(u64 val, int delta)
375 {
376 if (val != OBD_OBJECT_EOF)
377 val += delta;
378 return val;
379 }
380
381 static int lov_io_iter_init(const struct lu_env *env,
382 const struct cl_io_slice *ios)
383 {
384 struct lov_io *lio = cl2lov_io(env, ios);
385 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
386 struct lov_io_sub *sub;
387 u64 endpos;
388 u64 start;
389 u64 end;
390 int stripe;
391 int rc = 0;
392
393 endpos = lov_offset_mod(lio->lis_endpos, -1);
394 for (stripe = 0; stripe < lio->lis_stripe_count; stripe++) {
395 if (!lov_stripe_intersects(lsm, stripe, lio->lis_pos,
396 endpos, &start, &end))
397 continue;
398
399 if (unlikely(lov_r0(lio->lis_object)->lo_sub[stripe] == NULL)) {
400 if (ios->cis_io->ci_type == CIT_READ ||
401 ios->cis_io->ci_type == CIT_WRITE ||
402 ios->cis_io->ci_type == CIT_FAULT)
403 return -EIO;
404
405 continue;
406 }
407
408 end = lov_offset_mod(end, 1);
409 sub = lov_sub_get(env, lio, stripe);
410 if (!IS_ERR(sub)) {
411 lov_io_sub_inherit(sub->sub_io, lio, stripe,
412 start, end);
413 rc = cl_io_iter_init(sub->sub_env, sub->sub_io);
414 lov_sub_put(sub);
415 CDEBUG(D_VFSTRACE, "shrink: %d [%llu, %llu)\n",
416 stripe, start, end);
417 } else
418 rc = PTR_ERR(sub);
419
420 if (!rc)
421 list_add_tail(&sub->sub_linkage, &lio->lis_active);
422 else
423 break;
424 }
425 return rc;
426 }
427
428 static int lov_io_rw_iter_init(const struct lu_env *env,
429 const struct cl_io_slice *ios)
430 {
431 struct lov_io *lio = cl2lov_io(env, ios);
432 struct cl_io *io = ios->cis_io;
433 struct lov_stripe_md *lsm = lio->lis_object->lo_lsm;
434 __u64 start = io->u.ci_rw.crw_pos;
435 loff_t next;
436 unsigned long ssize = lsm->lsm_stripe_size;
437
438 LASSERT(io->ci_type == CIT_READ || io->ci_type == CIT_WRITE);
439
440 /* fast path for common case. */
441 if (lio->lis_nr_subios != 1 && !cl_io_is_append(io)) {
442
443 lov_do_div64(start, ssize);
444 next = (start + 1) * ssize;
445 if (next <= start * ssize)
446 next = ~0ull;
447
448 io->ci_continue = next < lio->lis_io_endpos;
449 io->u.ci_rw.crw_count = min_t(loff_t, lio->lis_io_endpos,
450 next) - io->u.ci_rw.crw_pos;
451 lio->lis_pos = io->u.ci_rw.crw_pos;
452 lio->lis_endpos = io->u.ci_rw.crw_pos + io->u.ci_rw.crw_count;
453 CDEBUG(D_VFSTRACE, "stripe: %llu chunk: [%llu, %llu) %llu\n",
454 (__u64)start, lio->lis_pos, lio->lis_endpos,
455 (__u64)lio->lis_io_endpos);
456 }
457 /*
458 * XXX The following call should be optimized: we know, that
459 * [lio->lis_pos, lio->lis_endpos) intersects with exactly one stripe.
460 */
461 return lov_io_iter_init(env, ios);
462 }
463
464 static int lov_io_call(const struct lu_env *env, struct lov_io *lio,
465 int (*iofunc)(const struct lu_env *, struct cl_io *))
466 {
467 struct cl_io *parent = lio->lis_cl.cis_io;
468 struct lov_io_sub *sub;
469 int rc = 0;
470
471 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
472 lov_sub_enter(sub);
473 rc = iofunc(sub->sub_env, sub->sub_io);
474 lov_sub_exit(sub);
475 if (rc)
476 break;
477
478 if (parent->ci_result == 0)
479 parent->ci_result = sub->sub_io->ci_result;
480 }
481 return rc;
482 }
483
484 static int lov_io_lock(const struct lu_env *env, const struct cl_io_slice *ios)
485 {
486 return lov_io_call(env, cl2lov_io(env, ios), cl_io_lock);
487 }
488
489 static int lov_io_start(const struct lu_env *env, const struct cl_io_slice *ios)
490 {
491 return lov_io_call(env, cl2lov_io(env, ios), cl_io_start);
492 }
493
494 static int lov_io_end_wrapper(const struct lu_env *env, struct cl_io *io)
495 {
496 /*
497 * It's possible that lov_io_start() wasn't called against this
498 * sub-io, either because previous sub-io failed, or upper layer
499 * completed IO.
500 */
501 if (io->ci_state == CIS_IO_GOING)
502 cl_io_end(env, io);
503 else
504 io->ci_state = CIS_IO_FINISHED;
505 return 0;
506 }
507
508 static int lov_io_iter_fini_wrapper(const struct lu_env *env, struct cl_io *io)
509 {
510 cl_io_iter_fini(env, io);
511 return 0;
512 }
513
514 static int lov_io_unlock_wrapper(const struct lu_env *env, struct cl_io *io)
515 {
516 cl_io_unlock(env, io);
517 return 0;
518 }
519
520 static void lov_io_end(const struct lu_env *env, const struct cl_io_slice *ios)
521 {
522 int rc;
523
524 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_end_wrapper);
525 LASSERT(rc == 0);
526 }
527
528 static void lov_io_iter_fini(const struct lu_env *env,
529 const struct cl_io_slice *ios)
530 {
531 struct lov_io *lio = cl2lov_io(env, ios);
532 int rc;
533
534 rc = lov_io_call(env, lio, lov_io_iter_fini_wrapper);
535 LASSERT(rc == 0);
536 while (!list_empty(&lio->lis_active))
537 list_del_init(lio->lis_active.next);
538 }
539
540 static void lov_io_unlock(const struct lu_env *env,
541 const struct cl_io_slice *ios)
542 {
543 int rc;
544
545 rc = lov_io_call(env, cl2lov_io(env, ios), lov_io_unlock_wrapper);
546 LASSERT(rc == 0);
547 }
548
549
550 static struct cl_page_list *lov_io_submit_qin(struct lov_device *ld,
551 struct cl_page_list *qin,
552 int idx, int alloc)
553 {
554 return alloc ? &qin[idx] : &ld->ld_emrg[idx]->emrg_page_list;
555 }
556
557 /**
558 * lov implementation of cl_operations::cio_submit() method. It takes a list
559 * of pages in \a queue, splits it into per-stripe sub-lists, invokes
560 * cl_io_submit() on underlying devices to submit sub-lists, and then splices
561 * everything back.
562 *
563 * Major complication of this function is a need to handle memory cleansing:
564 * cl_io_submit() is called to write out pages as a part of VM memory
565 * reclamation, and hence it may not fail due to memory shortages (system
566 * dead-locks otherwise). To deal with this, some resources (sub-lists,
567 * sub-environment, etc.) are allocated per-device on "startup" (i.e., in a
568 * not-memory cleansing context), and in case of memory shortage, these
569 * pre-allocated resources are used by lov_io_submit() under
570 * lov_device::ld_mutex mutex.
571 */
572 static int lov_io_submit(const struct lu_env *env,
573 const struct cl_io_slice *ios,
574 enum cl_req_type crt, struct cl_2queue *queue)
575 {
576 struct lov_io *lio = cl2lov_io(env, ios);
577 struct lov_object *obj = lio->lis_object;
578 struct lov_device *ld = lu2lov_dev(lov2cl(obj)->co_lu.lo_dev);
579 struct cl_page_list *qin = &queue->c2_qin;
580 struct cl_2queue *cl2q = &lov_env_info(env)->lti_cl2q;
581 struct cl_page_list *stripes_qin = NULL;
582 struct cl_page *page;
583 struct cl_page *tmp;
584 int stripe;
585
586 #define QIN(stripe) lov_io_submit_qin(ld, stripes_qin, stripe, alloc)
587
588 int rc = 0;
589 int alloc =
590 !(current->flags & PF_MEMALLOC);
591
592 if (lio->lis_active_subios == 1) {
593 int idx = lio->lis_single_subio_index;
594 struct lov_io_sub *sub;
595
596 LASSERT(idx < lio->lis_nr_subios);
597 sub = lov_sub_get(env, lio, idx);
598 LASSERT(!IS_ERR(sub));
599 LASSERT(sub->sub_io == &lio->lis_single_subio);
600 rc = cl_io_submit_rw(sub->sub_env, sub->sub_io,
601 crt, queue);
602 lov_sub_put(sub);
603 return rc;
604 }
605
606 LASSERT(lio->lis_subs != NULL);
607 if (alloc) {
608 stripes_qin =
609 libcfs_kvzalloc(sizeof(*stripes_qin) *
610 lio->lis_nr_subios,
611 GFP_NOFS);
612 if (stripes_qin == NULL)
613 return -ENOMEM;
614
615 for (stripe = 0; stripe < lio->lis_nr_subios; stripe++)
616 cl_page_list_init(&stripes_qin[stripe]);
617 } else {
618 /*
619 * If we get here, it means pageout & swap doesn't help.
620 * In order to not make things worse, even don't try to
621 * allocate the memory with __GFP_NOWARN. -jay
622 */
623 mutex_lock(&ld->ld_mutex);
624 lio->lis_mem_frozen = 1;
625 }
626
627 cl_2queue_init(cl2q);
628 cl_page_list_for_each_safe(page, tmp, qin) {
629 stripe = lov_page_stripe(page);
630 cl_page_list_move(QIN(stripe), qin, page);
631 }
632
633 for (stripe = 0; stripe < lio->lis_nr_subios; stripe++) {
634 struct lov_io_sub *sub;
635 struct cl_page_list *sub_qin = QIN(stripe);
636
637 if (list_empty(&sub_qin->pl_pages))
638 continue;
639
640 cl_page_list_splice(sub_qin, &cl2q->c2_qin);
641 sub = lov_sub_get(env, lio, stripe);
642 if (!IS_ERR(sub)) {
643 rc = cl_io_submit_rw(sub->sub_env, sub->sub_io,
644 crt, cl2q);
645 lov_sub_put(sub);
646 } else
647 rc = PTR_ERR(sub);
648 cl_page_list_splice(&cl2q->c2_qin, &queue->c2_qin);
649 cl_page_list_splice(&cl2q->c2_qout, &queue->c2_qout);
650 if (rc != 0)
651 break;
652 }
653
654 for (stripe = 0; stripe < lio->lis_nr_subios; stripe++) {
655 struct cl_page_list *sub_qin = QIN(stripe);
656
657 if (list_empty(&sub_qin->pl_pages))
658 continue;
659
660 cl_page_list_splice(sub_qin, qin);
661 }
662
663 if (alloc) {
664 kvfree(stripes_qin);
665 } else {
666 int i;
667
668 for (i = 0; i < lio->lis_nr_subios; i++) {
669 struct cl_io *cio = lio->lis_subs[i].sub_io;
670
671 if (cio && cio == &ld->ld_emrg[i]->emrg_subio)
672 lov_io_sub_fini(env, lio, &lio->lis_subs[i]);
673 }
674 lio->lis_mem_frozen = 0;
675 mutex_unlock(&ld->ld_mutex);
676 }
677
678 return rc;
679 #undef QIN
680 }
681
682 static int lov_io_prepare_write(const struct lu_env *env,
683 const struct cl_io_slice *ios,
684 const struct cl_page_slice *slice,
685 unsigned from, unsigned to)
686 {
687 struct lov_io *lio = cl2lov_io(env, ios);
688 struct cl_page *sub_page = lov_sub_page(slice);
689 struct lov_io_sub *sub;
690 int result;
691
692 sub = lov_page_subio(env, lio, slice);
693 if (!IS_ERR(sub)) {
694 result = cl_io_prepare_write(sub->sub_env, sub->sub_io,
695 sub_page, from, to);
696 lov_sub_put(sub);
697 } else
698 result = PTR_ERR(sub);
699 return result;
700 }
701
702 static int lov_io_commit_write(const struct lu_env *env,
703 const struct cl_io_slice *ios,
704 const struct cl_page_slice *slice,
705 unsigned from, unsigned to)
706 {
707 struct lov_io *lio = cl2lov_io(env, ios);
708 struct cl_page *sub_page = lov_sub_page(slice);
709 struct lov_io_sub *sub;
710 int result;
711
712 sub = lov_page_subio(env, lio, slice);
713 if (!IS_ERR(sub)) {
714 result = cl_io_commit_write(sub->sub_env, sub->sub_io,
715 sub_page, from, to);
716 lov_sub_put(sub);
717 } else
718 result = PTR_ERR(sub);
719 return result;
720 }
721
722 static int lov_io_fault_start(const struct lu_env *env,
723 const struct cl_io_slice *ios)
724 {
725 struct cl_fault_io *fio;
726 struct lov_io *lio;
727 struct lov_io_sub *sub;
728
729 fio = &ios->cis_io->u.ci_fault;
730 lio = cl2lov_io(env, ios);
731 sub = lov_sub_get(env, lio, lov_page_stripe(fio->ft_page));
732 sub->sub_io->u.ci_fault.ft_nob = fio->ft_nob;
733 lov_sub_put(sub);
734 return lov_io_start(env, ios);
735 }
736
737 static void lov_io_fsync_end(const struct lu_env *env,
738 const struct cl_io_slice *ios)
739 {
740 struct lov_io *lio = cl2lov_io(env, ios);
741 struct lov_io_sub *sub;
742 unsigned int *written = &ios->cis_io->u.ci_fsync.fi_nr_written;
743
744 *written = 0;
745 list_for_each_entry(sub, &lio->lis_active, sub_linkage) {
746 struct cl_io *subio = sub->sub_io;
747
748 lov_sub_enter(sub);
749 lov_io_end_wrapper(sub->sub_env, subio);
750 lov_sub_exit(sub);
751
752 if (subio->ci_result == 0)
753 *written += subio->u.ci_fsync.fi_nr_written;
754 }
755 }
756
757 static const struct cl_io_operations lov_io_ops = {
758 .op = {
759 [CIT_READ] = {
760 .cio_fini = lov_io_fini,
761 .cio_iter_init = lov_io_rw_iter_init,
762 .cio_iter_fini = lov_io_iter_fini,
763 .cio_lock = lov_io_lock,
764 .cio_unlock = lov_io_unlock,
765 .cio_start = lov_io_start,
766 .cio_end = lov_io_end
767 },
768 [CIT_WRITE] = {
769 .cio_fini = lov_io_fini,
770 .cio_iter_init = lov_io_rw_iter_init,
771 .cio_iter_fini = lov_io_iter_fini,
772 .cio_lock = lov_io_lock,
773 .cio_unlock = lov_io_unlock,
774 .cio_start = lov_io_start,
775 .cio_end = lov_io_end
776 },
777 [CIT_SETATTR] = {
778 .cio_fini = lov_io_fini,
779 .cio_iter_init = lov_io_iter_init,
780 .cio_iter_fini = lov_io_iter_fini,
781 .cio_lock = lov_io_lock,
782 .cio_unlock = lov_io_unlock,
783 .cio_start = lov_io_start,
784 .cio_end = lov_io_end
785 },
786 [CIT_FAULT] = {
787 .cio_fini = lov_io_fini,
788 .cio_iter_init = lov_io_iter_init,
789 .cio_iter_fini = lov_io_iter_fini,
790 .cio_lock = lov_io_lock,
791 .cio_unlock = lov_io_unlock,
792 .cio_start = lov_io_fault_start,
793 .cio_end = lov_io_end
794 },
795 [CIT_FSYNC] = {
796 .cio_fini = lov_io_fini,
797 .cio_iter_init = lov_io_iter_init,
798 .cio_iter_fini = lov_io_iter_fini,
799 .cio_lock = lov_io_lock,
800 .cio_unlock = lov_io_unlock,
801 .cio_start = lov_io_start,
802 .cio_end = lov_io_fsync_end
803 },
804 [CIT_MISC] = {
805 .cio_fini = lov_io_fini
806 }
807 },
808 .req_op = {
809 [CRT_READ] = {
810 .cio_submit = lov_io_submit
811 },
812 [CRT_WRITE] = {
813 .cio_submit = lov_io_submit
814 }
815 },
816 .cio_prepare_write = lov_io_prepare_write,
817 .cio_commit_write = lov_io_commit_write
818 };
819
820 /*****************************************************************************
821 *
822 * Empty lov io operations.
823 *
824 */
825
826 static void lov_empty_io_fini(const struct lu_env *env,
827 const struct cl_io_slice *ios)
828 {
829 struct lov_object *lov = cl2lov(ios->cis_obj);
830
831 if (atomic_dec_and_test(&lov->lo_active_ios))
832 wake_up_all(&lov->lo_waitq);
833 }
834
835 static void lov_empty_impossible(const struct lu_env *env,
836 struct cl_io_slice *ios)
837 {
838 LBUG();
839 }
840
841 #define LOV_EMPTY_IMPOSSIBLE ((void *)lov_empty_impossible)
842
843 /**
844 * An io operation vector for files without stripes.
845 */
846 static const struct cl_io_operations lov_empty_io_ops = {
847 .op = {
848 [CIT_READ] = {
849 .cio_fini = lov_empty_io_fini,
850 #if 0
851 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
852 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
853 .cio_start = LOV_EMPTY_IMPOSSIBLE,
854 .cio_end = LOV_EMPTY_IMPOSSIBLE
855 #endif
856 },
857 [CIT_WRITE] = {
858 .cio_fini = lov_empty_io_fini,
859 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
860 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
861 .cio_start = LOV_EMPTY_IMPOSSIBLE,
862 .cio_end = LOV_EMPTY_IMPOSSIBLE
863 },
864 [CIT_SETATTR] = {
865 .cio_fini = lov_empty_io_fini,
866 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
867 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
868 .cio_start = LOV_EMPTY_IMPOSSIBLE,
869 .cio_end = LOV_EMPTY_IMPOSSIBLE
870 },
871 [CIT_FAULT] = {
872 .cio_fini = lov_empty_io_fini,
873 .cio_iter_init = LOV_EMPTY_IMPOSSIBLE,
874 .cio_lock = LOV_EMPTY_IMPOSSIBLE,
875 .cio_start = LOV_EMPTY_IMPOSSIBLE,
876 .cio_end = LOV_EMPTY_IMPOSSIBLE
877 },
878 [CIT_FSYNC] = {
879 .cio_fini = lov_empty_io_fini
880 },
881 [CIT_MISC] = {
882 .cio_fini = lov_empty_io_fini
883 }
884 },
885 .req_op = {
886 [CRT_READ] = {
887 .cio_submit = LOV_EMPTY_IMPOSSIBLE
888 },
889 [CRT_WRITE] = {
890 .cio_submit = LOV_EMPTY_IMPOSSIBLE
891 }
892 },
893 .cio_commit_write = LOV_EMPTY_IMPOSSIBLE
894 };
895
896 int lov_io_init_raid0(const struct lu_env *env, struct cl_object *obj,
897 struct cl_io *io)
898 {
899 struct lov_io *lio = lov_env_io(env);
900 struct lov_object *lov = cl2lov(obj);
901
902 INIT_LIST_HEAD(&lio->lis_active);
903 lov_io_slice_init(lio, lov, io);
904 if (io->ci_result == 0) {
905 io->ci_result = lov_io_subio_init(env, lio, io);
906 if (io->ci_result == 0) {
907 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_io_ops);
908 atomic_inc(&lov->lo_active_ios);
909 }
910 }
911 return io->ci_result;
912 }
913
914 int lov_io_init_empty(const struct lu_env *env, struct cl_object *obj,
915 struct cl_io *io)
916 {
917 struct lov_object *lov = cl2lov(obj);
918 struct lov_io *lio = lov_env_io(env);
919 int result;
920
921 lio->lis_object = lov;
922 switch (io->ci_type) {
923 default:
924 LBUG();
925 case CIT_MISC:
926 case CIT_READ:
927 result = 0;
928 break;
929 case CIT_FSYNC:
930 case CIT_SETATTR:
931 result = 1;
932 break;
933 case CIT_WRITE:
934 result = -EBADF;
935 break;
936 case CIT_FAULT:
937 result = -EFAULT;
938 CERROR("Page fault on a file without stripes: "DFID"\n",
939 PFID(lu_object_fid(&obj->co_lu)));
940 break;
941 }
942 if (result == 0) {
943 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
944 atomic_inc(&lov->lo_active_ios);
945 }
946
947 io->ci_result = result < 0 ? result : 0;
948 return result != 0;
949 }
950
951 int lov_io_init_released(const struct lu_env *env, struct cl_object *obj,
952 struct cl_io *io)
953 {
954 struct lov_object *lov = cl2lov(obj);
955 struct lov_io *lio = lov_env_io(env);
956 int result;
957
958 LASSERT(lov->lo_lsm != NULL);
959 lio->lis_object = lov;
960
961 switch (io->ci_type) {
962 default:
963 LASSERTF(0, "invalid type %d\n", io->ci_type);
964 case CIT_MISC:
965 case CIT_FSYNC:
966 result = 1;
967 break;
968 case CIT_SETATTR:
969 /* the truncate to 0 is managed by MDT:
970 * - in open, for open O_TRUNC
971 * - in setattr, for truncate
972 */
973 /* the truncate is for size > 0 so triggers a restore */
974 if (cl_io_is_trunc(io))
975 io->ci_restore_needed = 1;
976 result = -ENODATA;
977 break;
978 case CIT_READ:
979 case CIT_WRITE:
980 case CIT_FAULT:
981 io->ci_restore_needed = 1;
982 result = -ENODATA;
983 break;
984 }
985 if (result == 0) {
986 cl_io_slice_add(io, &lio->lis_cl, obj, &lov_empty_io_ops);
987 atomic_inc(&lov->lo_active_ios);
988 }
989
990 io->ci_result = result < 0 ? result : 0;
991 return result != 0;
992 }
993 /** @} lov */