Loading...
Note: File does not exist in v6.9.4.
1/*
2 * pNFS Objects layout implementation over open-osd initiator library
3 *
4 * Copyright (C) 2009 Panasas Inc. [year of first publication]
5 * All rights reserved.
6 *
7 * Benny Halevy <bhalevy@panasas.com>
8 * Boaz Harrosh <bharrosh@panasas.com>
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2
12 * See the file COPYING included with this distribution for more details.
13 *
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
16 * are met:
17 *
18 * 1. Redistributions of source code must retain the above copyright
19 * notice, this list of conditions and the following disclaimer.
20 * 2. Redistributions in binary form must reproduce the above copyright
21 * notice, this list of conditions and the following disclaimer in the
22 * documentation and/or other materials provided with the distribution.
23 * 3. Neither the name of the Panasas company nor the names of its
24 * contributors may be used to endorse or promote products derived
25 * from this software without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
28 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
29 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
30 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
34 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
35 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
36 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
37 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
38 */
39
40#include <linux/module.h>
41#include <scsi/osd_initiator.h>
42
43#include "objlayout.h"
44
45#define NFSDBG_FACILITY NFSDBG_PNFS_LD
46
47#define _LLU(x) ((unsigned long long)x)
48
49enum { BIO_MAX_PAGES_KMALLOC =
50 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
51};
52
53struct objio_dev_ent {
54 struct nfs4_deviceid_node id_node;
55 struct osd_dev *od;
56};
57
58static void
59objio_free_deviceid_node(struct nfs4_deviceid_node *d)
60{
61 struct objio_dev_ent *de = container_of(d, struct objio_dev_ent, id_node);
62
63 dprintk("%s: free od=%p\n", __func__, de->od);
64 osduld_put_device(de->od);
65 kfree(de);
66}
67
68static struct objio_dev_ent *_dev_list_find(const struct nfs_server *nfss,
69 const struct nfs4_deviceid *d_id)
70{
71 struct nfs4_deviceid_node *d;
72 struct objio_dev_ent *de;
73
74 d = nfs4_find_get_deviceid(nfss->pnfs_curr_ld, nfss->nfs_client, d_id);
75 if (!d)
76 return NULL;
77
78 de = container_of(d, struct objio_dev_ent, id_node);
79 return de;
80}
81
82static struct objio_dev_ent *
83_dev_list_add(const struct nfs_server *nfss,
84 const struct nfs4_deviceid *d_id, struct osd_dev *od,
85 gfp_t gfp_flags)
86{
87 struct nfs4_deviceid_node *d;
88 struct objio_dev_ent *de = kzalloc(sizeof(*de), gfp_flags);
89 struct objio_dev_ent *n;
90
91 if (!de) {
92 dprintk("%s: -ENOMEM od=%p\n", __func__, od);
93 return NULL;
94 }
95
96 dprintk("%s: Adding od=%p\n", __func__, od);
97 nfs4_init_deviceid_node(&de->id_node,
98 nfss->pnfs_curr_ld,
99 nfss->nfs_client,
100 d_id);
101 de->od = od;
102
103 d = nfs4_insert_deviceid_node(&de->id_node);
104 n = container_of(d, struct objio_dev_ent, id_node);
105 if (n != de) {
106 dprintk("%s: Race with other n->od=%p\n", __func__, n->od);
107 objio_free_deviceid_node(&de->id_node);
108 de = n;
109 }
110
111 return de;
112}
113
114struct caps_buffers {
115 u8 caps_key[OSD_CRYPTO_KEYID_SIZE];
116 u8 creds[OSD_CAP_LEN];
117};
118
119struct objio_segment {
120 struct pnfs_layout_segment lseg;
121
122 struct pnfs_osd_object_cred *comps;
123
124 unsigned mirrors_p1;
125 unsigned stripe_unit;
126 unsigned group_width; /* Data stripe_units without integrity comps */
127 u64 group_depth;
128 unsigned group_count;
129
130 unsigned max_io_size;
131
132 unsigned comps_index;
133 unsigned num_comps;
134 /* variable length */
135 struct objio_dev_ent *ods[];
136};
137
138static inline struct objio_segment *
139OBJIO_LSEG(struct pnfs_layout_segment *lseg)
140{
141 return container_of(lseg, struct objio_segment, lseg);
142}
143
144struct objio_state;
145typedef ssize_t (*objio_done_fn)(struct objio_state *ios);
146
147struct objio_state {
148 /* Generic layer */
149 struct objlayout_io_state ol_state;
150
151 struct objio_segment *layout;
152
153 struct kref kref;
154 objio_done_fn done;
155 void *private;
156
157 unsigned long length;
158 unsigned numdevs; /* Actually used devs in this IO */
159 /* A per-device variable array of size numdevs */
160 struct _objio_per_comp {
161 struct bio *bio;
162 struct osd_request *or;
163 unsigned long length;
164 u64 offset;
165 unsigned dev;
166 } per_dev[];
167};
168
169/* Send and wait for a get_device_info of devices in the layout,
170 then look them up with the osd_initiator library */
171static struct objio_dev_ent *_device_lookup(struct pnfs_layout_hdr *pnfslay,
172 struct objio_segment *objio_seg, unsigned comp,
173 gfp_t gfp_flags)
174{
175 struct pnfs_osd_deviceaddr *deviceaddr;
176 struct nfs4_deviceid *d_id;
177 struct objio_dev_ent *ode;
178 struct osd_dev *od;
179 struct osd_dev_info odi;
180 int err;
181
182 d_id = &objio_seg->comps[comp].oc_object_id.oid_device_id;
183
184 ode = _dev_list_find(NFS_SERVER(pnfslay->plh_inode), d_id);
185 if (ode)
186 return ode;
187
188 err = objlayout_get_deviceinfo(pnfslay, d_id, &deviceaddr, gfp_flags);
189 if (unlikely(err)) {
190 dprintk("%s: objlayout_get_deviceinfo dev(%llx:%llx) =>%d\n",
191 __func__, _DEVID_LO(d_id), _DEVID_HI(d_id), err);
192 return ERR_PTR(err);
193 }
194
195 odi.systemid_len = deviceaddr->oda_systemid.len;
196 if (odi.systemid_len > sizeof(odi.systemid)) {
197 err = -EINVAL;
198 goto out;
199 } else if (odi.systemid_len)
200 memcpy(odi.systemid, deviceaddr->oda_systemid.data,
201 odi.systemid_len);
202 odi.osdname_len = deviceaddr->oda_osdname.len;
203 odi.osdname = (u8 *)deviceaddr->oda_osdname.data;
204
205 if (!odi.osdname_len && !odi.systemid_len) {
206 dprintk("%s: !odi.osdname_len && !odi.systemid_len\n",
207 __func__);
208 err = -ENODEV;
209 goto out;
210 }
211
212 od = osduld_info_lookup(&odi);
213 if (unlikely(IS_ERR(od))) {
214 err = PTR_ERR(od);
215 dprintk("%s: osduld_info_lookup => %d\n", __func__, err);
216 goto out;
217 }
218
219 ode = _dev_list_add(NFS_SERVER(pnfslay->plh_inode), d_id, od,
220 gfp_flags);
221
222out:
223 dprintk("%s: return=%d\n", __func__, err);
224 objlayout_put_deviceinfo(deviceaddr);
225 return err ? ERR_PTR(err) : ode;
226}
227
228static int objio_devices_lookup(struct pnfs_layout_hdr *pnfslay,
229 struct objio_segment *objio_seg,
230 gfp_t gfp_flags)
231{
232 unsigned i;
233 int err;
234
235 /* lookup all devices */
236 for (i = 0; i < objio_seg->num_comps; i++) {
237 struct objio_dev_ent *ode;
238
239 ode = _device_lookup(pnfslay, objio_seg, i, gfp_flags);
240 if (unlikely(IS_ERR(ode))) {
241 err = PTR_ERR(ode);
242 goto out;
243 }
244 objio_seg->ods[i] = ode;
245 }
246 err = 0;
247
248out:
249 dprintk("%s: return=%d\n", __func__, err);
250 return err;
251}
252
253static int _verify_data_map(struct pnfs_osd_layout *layout)
254{
255 struct pnfs_osd_data_map *data_map = &layout->olo_map;
256 u64 stripe_length;
257 u32 group_width;
258
259/* FIXME: Only raid0 for now. if not go through MDS */
260 if (data_map->odm_raid_algorithm != PNFS_OSD_RAID_0) {
261 printk(KERN_ERR "Only RAID_0 for now\n");
262 return -ENOTSUPP;
263 }
264 if (0 != (data_map->odm_num_comps % (data_map->odm_mirror_cnt + 1))) {
265 printk(KERN_ERR "Data Map wrong, num_comps=%u mirrors=%u\n",
266 data_map->odm_num_comps, data_map->odm_mirror_cnt);
267 return -EINVAL;
268 }
269
270 if (data_map->odm_group_width)
271 group_width = data_map->odm_group_width;
272 else
273 group_width = data_map->odm_num_comps /
274 (data_map->odm_mirror_cnt + 1);
275
276 stripe_length = (u64)data_map->odm_stripe_unit * group_width;
277 if (stripe_length >= (1ULL << 32)) {
278 printk(KERN_ERR "Total Stripe length(0x%llx)"
279 " >= 32bit is not supported\n", _LLU(stripe_length));
280 return -ENOTSUPP;
281 }
282
283 if (0 != (data_map->odm_stripe_unit & ~PAGE_MASK)) {
284 printk(KERN_ERR "Stripe Unit(0x%llx)"
285 " must be Multples of PAGE_SIZE(0x%lx)\n",
286 _LLU(data_map->odm_stripe_unit), PAGE_SIZE);
287 return -ENOTSUPP;
288 }
289
290 return 0;
291}
292
293static void copy_single_comp(struct pnfs_osd_object_cred *cur_comp,
294 struct pnfs_osd_object_cred *src_comp,
295 struct caps_buffers *caps_p)
296{
297 WARN_ON(src_comp->oc_cap_key.cred_len > sizeof(caps_p->caps_key));
298 WARN_ON(src_comp->oc_cap.cred_len > sizeof(caps_p->creds));
299
300 *cur_comp = *src_comp;
301
302 memcpy(caps_p->caps_key, src_comp->oc_cap_key.cred,
303 sizeof(caps_p->caps_key));
304 cur_comp->oc_cap_key.cred = caps_p->caps_key;
305
306 memcpy(caps_p->creds, src_comp->oc_cap.cred,
307 sizeof(caps_p->creds));
308 cur_comp->oc_cap.cred = caps_p->creds;
309}
310
311int objio_alloc_lseg(struct pnfs_layout_segment **outp,
312 struct pnfs_layout_hdr *pnfslay,
313 struct pnfs_layout_range *range,
314 struct xdr_stream *xdr,
315 gfp_t gfp_flags)
316{
317 struct objio_segment *objio_seg;
318 struct pnfs_osd_xdr_decode_layout_iter iter;
319 struct pnfs_osd_layout layout;
320 struct pnfs_osd_object_cred *cur_comp, src_comp;
321 struct caps_buffers *caps_p;
322 int err;
323
324 err = pnfs_osd_xdr_decode_layout_map(&layout, &iter, xdr);
325 if (unlikely(err))
326 return err;
327
328 err = _verify_data_map(&layout);
329 if (unlikely(err))
330 return err;
331
332 objio_seg = kzalloc(sizeof(*objio_seg) +
333 sizeof(objio_seg->ods[0]) * layout.olo_num_comps +
334 sizeof(*objio_seg->comps) * layout.olo_num_comps +
335 sizeof(struct caps_buffers) * layout.olo_num_comps,
336 gfp_flags);
337 if (!objio_seg)
338 return -ENOMEM;
339
340 objio_seg->comps = (void *)(objio_seg->ods + layout.olo_num_comps);
341 cur_comp = objio_seg->comps;
342 caps_p = (void *)(cur_comp + layout.olo_num_comps);
343 while (pnfs_osd_xdr_decode_layout_comp(&src_comp, &iter, xdr, &err))
344 copy_single_comp(cur_comp++, &src_comp, caps_p++);
345 if (unlikely(err))
346 goto err;
347
348 objio_seg->num_comps = layout.olo_num_comps;
349 objio_seg->comps_index = layout.olo_comps_index;
350 err = objio_devices_lookup(pnfslay, objio_seg, gfp_flags);
351 if (err)
352 goto err;
353
354 objio_seg->mirrors_p1 = layout.olo_map.odm_mirror_cnt + 1;
355 objio_seg->stripe_unit = layout.olo_map.odm_stripe_unit;
356 if (layout.olo_map.odm_group_width) {
357 objio_seg->group_width = layout.olo_map.odm_group_width;
358 objio_seg->group_depth = layout.olo_map.odm_group_depth;
359 objio_seg->group_count = layout.olo_map.odm_num_comps /
360 objio_seg->mirrors_p1 /
361 objio_seg->group_width;
362 } else {
363 objio_seg->group_width = layout.olo_map.odm_num_comps /
364 objio_seg->mirrors_p1;
365 objio_seg->group_depth = -1;
366 objio_seg->group_count = 1;
367 }
368
369 /* Cache this calculation it will hit for every page */
370 objio_seg->max_io_size = (BIO_MAX_PAGES_KMALLOC * PAGE_SIZE -
371 objio_seg->stripe_unit) *
372 objio_seg->group_width;
373
374 *outp = &objio_seg->lseg;
375 return 0;
376
377err:
378 kfree(objio_seg);
379 dprintk("%s: Error: return %d\n", __func__, err);
380 *outp = NULL;
381 return err;
382}
383
384void objio_free_lseg(struct pnfs_layout_segment *lseg)
385{
386 int i;
387 struct objio_segment *objio_seg = OBJIO_LSEG(lseg);
388
389 for (i = 0; i < objio_seg->num_comps; i++) {
390 if (!objio_seg->ods[i])
391 break;
392 nfs4_put_deviceid_node(&objio_seg->ods[i]->id_node);
393 }
394 kfree(objio_seg);
395}
396
397int objio_alloc_io_state(struct pnfs_layout_segment *lseg,
398 struct objlayout_io_state **outp,
399 gfp_t gfp_flags)
400{
401 struct objio_segment *objio_seg = OBJIO_LSEG(lseg);
402 struct objio_state *ios;
403 const unsigned first_size = sizeof(*ios) +
404 objio_seg->num_comps * sizeof(ios->per_dev[0]);
405 const unsigned sec_size = objio_seg->num_comps *
406 sizeof(ios->ol_state.ioerrs[0]);
407
408 ios = kzalloc(first_size + sec_size, gfp_flags);
409 if (unlikely(!ios))
410 return -ENOMEM;
411
412 ios->layout = objio_seg;
413 ios->ol_state.ioerrs = ((void *)ios) + first_size;
414 ios->ol_state.num_comps = objio_seg->num_comps;
415
416 *outp = &ios->ol_state;
417 return 0;
418}
419
420void objio_free_io_state(struct objlayout_io_state *ol_state)
421{
422 struct objio_state *ios = container_of(ol_state, struct objio_state,
423 ol_state);
424
425 kfree(ios);
426}
427
428enum pnfs_osd_errno osd_pri_2_pnfs_err(enum osd_err_priority oep)
429{
430 switch (oep) {
431 case OSD_ERR_PRI_NO_ERROR:
432 return (enum pnfs_osd_errno)0;
433
434 case OSD_ERR_PRI_CLEAR_PAGES:
435 BUG_ON(1);
436 return 0;
437
438 case OSD_ERR_PRI_RESOURCE:
439 return PNFS_OSD_ERR_RESOURCE;
440 case OSD_ERR_PRI_BAD_CRED:
441 return PNFS_OSD_ERR_BAD_CRED;
442 case OSD_ERR_PRI_NO_ACCESS:
443 return PNFS_OSD_ERR_NO_ACCESS;
444 case OSD_ERR_PRI_UNREACHABLE:
445 return PNFS_OSD_ERR_UNREACHABLE;
446 case OSD_ERR_PRI_NOT_FOUND:
447 return PNFS_OSD_ERR_NOT_FOUND;
448 case OSD_ERR_PRI_NO_SPACE:
449 return PNFS_OSD_ERR_NO_SPACE;
450 default:
451 WARN_ON(1);
452 /* fallthrough */
453 case OSD_ERR_PRI_EIO:
454 return PNFS_OSD_ERR_EIO;
455 }
456}
457
458static void _clear_bio(struct bio *bio)
459{
460 struct bio_vec *bv;
461 unsigned i;
462
463 __bio_for_each_segment(bv, bio, i, 0) {
464 unsigned this_count = bv->bv_len;
465
466 if (likely(PAGE_SIZE == this_count))
467 clear_highpage(bv->bv_page);
468 else
469 zero_user(bv->bv_page, bv->bv_offset, this_count);
470 }
471}
472
473static int _io_check(struct objio_state *ios, bool is_write)
474{
475 enum osd_err_priority oep = OSD_ERR_PRI_NO_ERROR;
476 int lin_ret = 0;
477 int i;
478
479 for (i = 0; i < ios->numdevs; i++) {
480 struct osd_sense_info osi;
481 struct osd_request *or = ios->per_dev[i].or;
482 int ret;
483
484 if (!or)
485 continue;
486
487 ret = osd_req_decode_sense(or, &osi);
488 if (likely(!ret))
489 continue;
490
491 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
492 /* start read offset passed endof file */
493 BUG_ON(is_write);
494 _clear_bio(ios->per_dev[i].bio);
495 dprintk("%s: start read offset passed end of file "
496 "offset=0x%llx, length=0x%lx\n", __func__,
497 _LLU(ios->per_dev[i].offset),
498 ios->per_dev[i].length);
499
500 continue; /* we recovered */
501 }
502 objlayout_io_set_result(&ios->ol_state, i,
503 &ios->layout->comps[i].oc_object_id,
504 osd_pri_2_pnfs_err(osi.osd_err_pri),
505 ios->per_dev[i].offset,
506 ios->per_dev[i].length,
507 is_write);
508
509 if (osi.osd_err_pri >= oep) {
510 oep = osi.osd_err_pri;
511 lin_ret = ret;
512 }
513 }
514
515 return lin_ret;
516}
517
518/*
519 * Common IO state helpers.
520 */
521static void _io_free(struct objio_state *ios)
522{
523 unsigned i;
524
525 for (i = 0; i < ios->numdevs; i++) {
526 struct _objio_per_comp *per_dev = &ios->per_dev[i];
527
528 if (per_dev->or) {
529 osd_end_request(per_dev->or);
530 per_dev->or = NULL;
531 }
532
533 if (per_dev->bio) {
534 bio_put(per_dev->bio);
535 per_dev->bio = NULL;
536 }
537 }
538}
539
540struct osd_dev *_io_od(struct objio_state *ios, unsigned dev)
541{
542 unsigned min_dev = ios->layout->comps_index;
543 unsigned max_dev = min_dev + ios->layout->num_comps;
544
545 BUG_ON(dev < min_dev || max_dev <= dev);
546 return ios->layout->ods[dev - min_dev]->od;
547}
548
549struct _striping_info {
550 u64 obj_offset;
551 u64 group_length;
552 unsigned dev;
553 unsigned unit_off;
554};
555
556static void _calc_stripe_info(struct objio_state *ios, u64 file_offset,
557 struct _striping_info *si)
558{
559 u32 stripe_unit = ios->layout->stripe_unit;
560 u32 group_width = ios->layout->group_width;
561 u64 group_depth = ios->layout->group_depth;
562 u32 U = stripe_unit * group_width;
563
564 u64 T = U * group_depth;
565 u64 S = T * ios->layout->group_count;
566 u64 M = div64_u64(file_offset, S);
567
568 /*
569 G = (L - (M * S)) / T
570 H = (L - (M * S)) % T
571 */
572 u64 LmodU = file_offset - M * S;
573 u32 G = div64_u64(LmodU, T);
574 u64 H = LmodU - G * T;
575
576 u32 N = div_u64(H, U);
577
578 div_u64_rem(file_offset, stripe_unit, &si->unit_off);
579 si->obj_offset = si->unit_off + (N * stripe_unit) +
580 (M * group_depth * stripe_unit);
581
582 /* "H - (N * U)" is just "H % U" so it's bound to u32 */
583 si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
584 si->dev *= ios->layout->mirrors_p1;
585
586 si->group_length = T - H;
587}
588
589static int _add_stripe_unit(struct objio_state *ios, unsigned *cur_pg,
590 unsigned pgbase, struct _objio_per_comp *per_dev, int len,
591 gfp_t gfp_flags)
592{
593 unsigned pg = *cur_pg;
594 int cur_len = len;
595 struct request_queue *q =
596 osd_request_queue(_io_od(ios, per_dev->dev));
597
598 if (per_dev->bio == NULL) {
599 unsigned pages_in_stripe = ios->layout->group_width *
600 (ios->layout->stripe_unit / PAGE_SIZE);
601 unsigned bio_size = (ios->ol_state.nr_pages + pages_in_stripe) /
602 ios->layout->group_width;
603
604 if (BIO_MAX_PAGES_KMALLOC < bio_size)
605 bio_size = BIO_MAX_PAGES_KMALLOC;
606
607 per_dev->bio = bio_kmalloc(gfp_flags, bio_size);
608 if (unlikely(!per_dev->bio)) {
609 dprintk("Faild to allocate BIO size=%u\n", bio_size);
610 return -ENOMEM;
611 }
612 }
613
614 while (cur_len > 0) {
615 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
616 unsigned added_len;
617
618 BUG_ON(ios->ol_state.nr_pages <= pg);
619 cur_len -= pglen;
620
621 added_len = bio_add_pc_page(q, per_dev->bio,
622 ios->ol_state.pages[pg], pglen, pgbase);
623 if (unlikely(pglen != added_len))
624 return -ENOMEM;
625 pgbase = 0;
626 ++pg;
627 }
628 BUG_ON(cur_len);
629
630 per_dev->length += len;
631 *cur_pg = pg;
632 return 0;
633}
634
635static int _prepare_one_group(struct objio_state *ios, u64 length,
636 struct _striping_info *si, unsigned *last_pg,
637 gfp_t gfp_flags)
638{
639 unsigned stripe_unit = ios->layout->stripe_unit;
640 unsigned mirrors_p1 = ios->layout->mirrors_p1;
641 unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
642 unsigned dev = si->dev;
643 unsigned first_dev = dev - (dev % devs_in_group);
644 unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
645 unsigned cur_pg = *last_pg;
646 int ret = 0;
647
648 while (length) {
649 struct _objio_per_comp *per_dev = &ios->per_dev[dev - first_dev];
650 unsigned cur_len, page_off = 0;
651
652 if (!per_dev->length) {
653 per_dev->dev = dev;
654 if (dev < si->dev) {
655 per_dev->offset = si->obj_offset + stripe_unit -
656 si->unit_off;
657 cur_len = stripe_unit;
658 } else if (dev == si->dev) {
659 per_dev->offset = si->obj_offset;
660 cur_len = stripe_unit - si->unit_off;
661 page_off = si->unit_off & ~PAGE_MASK;
662 BUG_ON(page_off &&
663 (page_off != ios->ol_state.pgbase));
664 } else { /* dev > si->dev */
665 per_dev->offset = si->obj_offset - si->unit_off;
666 cur_len = stripe_unit;
667 }
668
669 if (max_comp < dev - first_dev)
670 max_comp = dev - first_dev;
671 } else {
672 cur_len = stripe_unit;
673 }
674 if (cur_len >= length)
675 cur_len = length;
676
677 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
678 cur_len, gfp_flags);
679 if (unlikely(ret))
680 goto out;
681
682 dev += mirrors_p1;
683 dev = (dev % devs_in_group) + first_dev;
684
685 length -= cur_len;
686 ios->length += cur_len;
687 }
688out:
689 ios->numdevs = max_comp + mirrors_p1;
690 *last_pg = cur_pg;
691 return ret;
692}
693
694static int _io_rw_pagelist(struct objio_state *ios, gfp_t gfp_flags)
695{
696 u64 length = ios->ol_state.count;
697 u64 offset = ios->ol_state.offset;
698 struct _striping_info si;
699 unsigned last_pg = 0;
700 int ret = 0;
701
702 while (length) {
703 _calc_stripe_info(ios, offset, &si);
704
705 if (length < si.group_length)
706 si.group_length = length;
707
708 ret = _prepare_one_group(ios, si.group_length, &si, &last_pg, gfp_flags);
709 if (unlikely(ret))
710 goto out;
711
712 offset += si.group_length;
713 length -= si.group_length;
714 }
715
716out:
717 if (!ios->length)
718 return ret;
719
720 return 0;
721}
722
723static ssize_t _sync_done(struct objio_state *ios)
724{
725 struct completion *waiting = ios->private;
726
727 complete(waiting);
728 return 0;
729}
730
731static void _last_io(struct kref *kref)
732{
733 struct objio_state *ios = container_of(kref, struct objio_state, kref);
734
735 ios->done(ios);
736}
737
738static void _done_io(struct osd_request *or, void *p)
739{
740 struct objio_state *ios = p;
741
742 kref_put(&ios->kref, _last_io);
743}
744
745static ssize_t _io_exec(struct objio_state *ios)
746{
747 DECLARE_COMPLETION_ONSTACK(wait);
748 ssize_t status = 0; /* sync status */
749 unsigned i;
750 objio_done_fn saved_done_fn = ios->done;
751 bool sync = ios->ol_state.sync;
752
753 if (sync) {
754 ios->done = _sync_done;
755 ios->private = &wait;
756 }
757
758 kref_init(&ios->kref);
759
760 for (i = 0; i < ios->numdevs; i++) {
761 struct osd_request *or = ios->per_dev[i].or;
762
763 if (!or)
764 continue;
765
766 kref_get(&ios->kref);
767 osd_execute_request_async(or, _done_io, ios);
768 }
769
770 kref_put(&ios->kref, _last_io);
771
772 if (sync) {
773 wait_for_completion(&wait);
774 status = saved_done_fn(ios);
775 }
776
777 return status;
778}
779
780/*
781 * read
782 */
783static ssize_t _read_done(struct objio_state *ios)
784{
785 ssize_t status;
786 int ret = _io_check(ios, false);
787
788 _io_free(ios);
789
790 if (likely(!ret))
791 status = ios->length;
792 else
793 status = ret;
794
795 objlayout_read_done(&ios->ol_state, status, ios->ol_state.sync);
796 return status;
797}
798
799static int _read_mirrors(struct objio_state *ios, unsigned cur_comp)
800{
801 struct osd_request *or = NULL;
802 struct _objio_per_comp *per_dev = &ios->per_dev[cur_comp];
803 unsigned dev = per_dev->dev;
804 struct pnfs_osd_object_cred *cred =
805 &ios->layout->comps[cur_comp];
806 struct osd_obj_id obj = {
807 .partition = cred->oc_object_id.oid_partition_id,
808 .id = cred->oc_object_id.oid_object_id,
809 };
810 int ret;
811
812 or = osd_start_request(_io_od(ios, dev), GFP_KERNEL);
813 if (unlikely(!or)) {
814 ret = -ENOMEM;
815 goto err;
816 }
817 per_dev->or = or;
818
819 osd_req_read(or, &obj, per_dev->offset, per_dev->bio, per_dev->length);
820
821 ret = osd_finalize_request(or, 0, cred->oc_cap.cred, NULL);
822 if (ret) {
823 dprintk("%s: Faild to osd_finalize_request() => %d\n",
824 __func__, ret);
825 goto err;
826 }
827
828 dprintk("%s:[%d] dev=%d obj=0x%llx start=0x%llx length=0x%lx\n",
829 __func__, cur_comp, dev, obj.id, _LLU(per_dev->offset),
830 per_dev->length);
831
832err:
833 return ret;
834}
835
836static ssize_t _read_exec(struct objio_state *ios)
837{
838 unsigned i;
839 int ret;
840
841 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
842 if (!ios->per_dev[i].length)
843 continue;
844 ret = _read_mirrors(ios, i);
845 if (unlikely(ret))
846 goto err;
847 }
848
849 ios->done = _read_done;
850 return _io_exec(ios); /* In sync mode exec returns the io status */
851
852err:
853 _io_free(ios);
854 return ret;
855}
856
857ssize_t objio_read_pagelist(struct objlayout_io_state *ol_state)
858{
859 struct objio_state *ios = container_of(ol_state, struct objio_state,
860 ol_state);
861 int ret;
862
863 ret = _io_rw_pagelist(ios, GFP_KERNEL);
864 if (unlikely(ret))
865 return ret;
866
867 return _read_exec(ios);
868}
869
870/*
871 * write
872 */
873static ssize_t _write_done(struct objio_state *ios)
874{
875 ssize_t status;
876 int ret = _io_check(ios, true);
877
878 _io_free(ios);
879
880 if (likely(!ret)) {
881 /* FIXME: should be based on the OSD's persistence model
882 * See OSD2r05 Section 4.13 Data persistence model */
883 ios->ol_state.committed = NFS_FILE_SYNC;
884 status = ios->length;
885 } else {
886 status = ret;
887 }
888
889 objlayout_write_done(&ios->ol_state, status, ios->ol_state.sync);
890 return status;
891}
892
893static int _write_mirrors(struct objio_state *ios, unsigned cur_comp)
894{
895 struct _objio_per_comp *master_dev = &ios->per_dev[cur_comp];
896 unsigned dev = ios->per_dev[cur_comp].dev;
897 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
898 int ret;
899
900 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
901 struct osd_request *or = NULL;
902 struct pnfs_osd_object_cred *cred =
903 &ios->layout->comps[cur_comp];
904 struct osd_obj_id obj = {
905 .partition = cred->oc_object_id.oid_partition_id,
906 .id = cred->oc_object_id.oid_object_id,
907 };
908 struct _objio_per_comp *per_dev = &ios->per_dev[cur_comp];
909 struct bio *bio;
910
911 or = osd_start_request(_io_od(ios, dev), GFP_NOFS);
912 if (unlikely(!or)) {
913 ret = -ENOMEM;
914 goto err;
915 }
916 per_dev->or = or;
917
918 if (per_dev != master_dev) {
919 bio = bio_kmalloc(GFP_NOFS,
920 master_dev->bio->bi_max_vecs);
921 if (unlikely(!bio)) {
922 dprintk("Faild to allocate BIO size=%u\n",
923 master_dev->bio->bi_max_vecs);
924 ret = -ENOMEM;
925 goto err;
926 }
927
928 __bio_clone(bio, master_dev->bio);
929 bio->bi_bdev = NULL;
930 bio->bi_next = NULL;
931 per_dev->bio = bio;
932 per_dev->dev = dev;
933 per_dev->length = master_dev->length;
934 per_dev->offset = master_dev->offset;
935 } else {
936 bio = master_dev->bio;
937 bio->bi_rw |= REQ_WRITE;
938 }
939
940 osd_req_write(or, &obj, per_dev->offset, bio, per_dev->length);
941
942 ret = osd_finalize_request(or, 0, cred->oc_cap.cred, NULL);
943 if (ret) {
944 dprintk("%s: Faild to osd_finalize_request() => %d\n",
945 __func__, ret);
946 goto err;
947 }
948
949 dprintk("%s:[%d] dev=%d obj=0x%llx start=0x%llx length=0x%lx\n",
950 __func__, cur_comp, dev, obj.id, _LLU(per_dev->offset),
951 per_dev->length);
952 }
953
954err:
955 return ret;
956}
957
958static ssize_t _write_exec(struct objio_state *ios)
959{
960 unsigned i;
961 int ret;
962
963 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
964 if (!ios->per_dev[i].length)
965 continue;
966 ret = _write_mirrors(ios, i);
967 if (unlikely(ret))
968 goto err;
969 }
970
971 ios->done = _write_done;
972 return _io_exec(ios); /* In sync mode exec returns the io->status */
973
974err:
975 _io_free(ios);
976 return ret;
977}
978
979ssize_t objio_write_pagelist(struct objlayout_io_state *ol_state, bool stable)
980{
981 struct objio_state *ios = container_of(ol_state, struct objio_state,
982 ol_state);
983 int ret;
984
985 /* TODO: ios->stable = stable; */
986 ret = _io_rw_pagelist(ios, GFP_NOFS);
987 if (unlikely(ret))
988 return ret;
989
990 return _write_exec(ios);
991}
992
993static bool objio_pg_test(struct nfs_pageio_descriptor *pgio,
994 struct nfs_page *prev, struct nfs_page *req)
995{
996 if (!pnfs_generic_pg_test(pgio, prev, req))
997 return false;
998
999 return pgio->pg_count + req->wb_bytes <=
1000 OBJIO_LSEG(pgio->pg_lseg)->max_io_size;
1001}
1002
1003static const struct nfs_pageio_ops objio_pg_read_ops = {
1004 .pg_init = pnfs_generic_pg_init_read,
1005 .pg_test = objio_pg_test,
1006 .pg_doio = pnfs_generic_pg_readpages,
1007};
1008
1009static const struct nfs_pageio_ops objio_pg_write_ops = {
1010 .pg_init = pnfs_generic_pg_init_write,
1011 .pg_test = objio_pg_test,
1012 .pg_doio = pnfs_generic_pg_writepages,
1013};
1014
1015static struct pnfs_layoutdriver_type objlayout_type = {
1016 .id = LAYOUT_OSD2_OBJECTS,
1017 .name = "LAYOUT_OSD2_OBJECTS",
1018 .flags = PNFS_LAYOUTRET_ON_SETATTR,
1019
1020 .alloc_layout_hdr = objlayout_alloc_layout_hdr,
1021 .free_layout_hdr = objlayout_free_layout_hdr,
1022
1023 .alloc_lseg = objlayout_alloc_lseg,
1024 .free_lseg = objlayout_free_lseg,
1025
1026 .read_pagelist = objlayout_read_pagelist,
1027 .write_pagelist = objlayout_write_pagelist,
1028 .pg_read_ops = &objio_pg_read_ops,
1029 .pg_write_ops = &objio_pg_write_ops,
1030
1031 .free_deviceid_node = objio_free_deviceid_node,
1032
1033 .encode_layoutcommit = objlayout_encode_layoutcommit,
1034 .encode_layoutreturn = objlayout_encode_layoutreturn,
1035};
1036
1037MODULE_DESCRIPTION("pNFS Layout Driver for OSD2 objects");
1038MODULE_AUTHOR("Benny Halevy <bhalevy@panasas.com>");
1039MODULE_LICENSE("GPL");
1040
1041static int __init
1042objlayout_init(void)
1043{
1044 int ret = pnfs_register_layoutdriver(&objlayout_type);
1045
1046 if (ret)
1047 printk(KERN_INFO
1048 "%s: Registering OSD pNFS Layout Driver failed: error=%d\n",
1049 __func__, ret);
1050 else
1051 printk(KERN_INFO "%s: Registered OSD pNFS Layout Driver\n",
1052 __func__);
1053 return ret;
1054}
1055
1056static void __exit
1057objlayout_exit(void)
1058{
1059 pnfs_unregister_layoutdriver(&objlayout_type);
1060 printk(KERN_INFO "%s: Unregistered OSD pNFS Layout Driver\n",
1061 __func__);
1062}
1063
1064MODULE_ALIAS("nfs-layouttype4-2");
1065
1066module_init(objlayout_init);
1067module_exit(objlayout_exit);