Merge branch 'for-3.0' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound...
[sfrench/cifs-2.6.git] / fs / nfs / objlayout / objio_osd.c
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
49 enum { BIO_MAX_PAGES_KMALLOC =
50                 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
51 };
52
53 struct objio_dev_ent {
54         struct nfs4_deviceid_node id_node;
55         struct osd_dev *od;
56 };
57
58 static void
59 objio_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
68 static 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
82 static 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         atomic_inc(&de->id_node.ref);
112         return de;
113 }
114
115 struct caps_buffers {
116         u8 caps_key[OSD_CRYPTO_KEYID_SIZE];
117         u8 creds[OSD_CAP_LEN];
118 };
119
120 struct objio_segment {
121         struct pnfs_layout_segment lseg;
122
123         struct pnfs_osd_object_cred *comps;
124
125         unsigned mirrors_p1;
126         unsigned stripe_unit;
127         unsigned group_width;   /* Data stripe_units without integrity comps */
128         u64 group_depth;
129         unsigned group_count;
130
131         unsigned max_io_size;
132
133         unsigned comps_index;
134         unsigned num_comps;
135         /* variable length */
136         struct objio_dev_ent *ods[];
137 };
138
139 static inline struct objio_segment *
140 OBJIO_LSEG(struct pnfs_layout_segment *lseg)
141 {
142         return container_of(lseg, struct objio_segment, lseg);
143 }
144
145 struct objio_state;
146 typedef ssize_t (*objio_done_fn)(struct objio_state *ios);
147
148 struct objio_state {
149         /* Generic layer */
150         struct objlayout_io_state ol_state;
151
152         struct objio_segment *layout;
153
154         struct kref kref;
155         objio_done_fn done;
156         void *private;
157
158         unsigned long length;
159         unsigned numdevs; /* Actually used devs in this IO */
160         /* A per-device variable array of size numdevs */
161         struct _objio_per_comp {
162                 struct bio *bio;
163                 struct osd_request *or;
164                 unsigned long length;
165                 u64 offset;
166                 unsigned dev;
167         } per_dev[];
168 };
169
170 /* Send and wait for a get_device_info of devices in the layout,
171    then look them up with the osd_initiator library */
172 static struct objio_dev_ent *_device_lookup(struct pnfs_layout_hdr *pnfslay,
173                                 struct objio_segment *objio_seg, unsigned comp,
174                                 gfp_t gfp_flags)
175 {
176         struct pnfs_osd_deviceaddr *deviceaddr;
177         struct nfs4_deviceid *d_id;
178         struct objio_dev_ent *ode;
179         struct osd_dev *od;
180         struct osd_dev_info odi;
181         int err;
182
183         d_id = &objio_seg->comps[comp].oc_object_id.oid_device_id;
184
185         ode = _dev_list_find(NFS_SERVER(pnfslay->plh_inode), d_id);
186         if (ode)
187                 return ode;
188
189         err = objlayout_get_deviceinfo(pnfslay, d_id, &deviceaddr, gfp_flags);
190         if (unlikely(err)) {
191                 dprintk("%s: objlayout_get_deviceinfo dev(%llx:%llx) =>%d\n",
192                         __func__, _DEVID_LO(d_id), _DEVID_HI(d_id), err);
193                 return ERR_PTR(err);
194         }
195
196         odi.systemid_len = deviceaddr->oda_systemid.len;
197         if (odi.systemid_len > sizeof(odi.systemid)) {
198                 err = -EINVAL;
199                 goto out;
200         } else if (odi.systemid_len)
201                 memcpy(odi.systemid, deviceaddr->oda_systemid.data,
202                        odi.systemid_len);
203         odi.osdname_len  = deviceaddr->oda_osdname.len;
204         odi.osdname      = (u8 *)deviceaddr->oda_osdname.data;
205
206         if (!odi.osdname_len && !odi.systemid_len) {
207                 dprintk("%s: !odi.osdname_len && !odi.systemid_len\n",
208                         __func__);
209                 err = -ENODEV;
210                 goto out;
211         }
212
213         od = osduld_info_lookup(&odi);
214         if (unlikely(IS_ERR(od))) {
215                 err = PTR_ERR(od);
216                 dprintk("%s: osduld_info_lookup => %d\n", __func__, err);
217                 goto out;
218         }
219
220         ode = _dev_list_add(NFS_SERVER(pnfslay->plh_inode), d_id, od,
221                             gfp_flags);
222
223 out:
224         dprintk("%s: return=%d\n", __func__, err);
225         objlayout_put_deviceinfo(deviceaddr);
226         return err ? ERR_PTR(err) : ode;
227 }
228
229 static int objio_devices_lookup(struct pnfs_layout_hdr *pnfslay,
230         struct objio_segment *objio_seg,
231         gfp_t gfp_flags)
232 {
233         unsigned i;
234         int err;
235
236         /* lookup all devices */
237         for (i = 0; i < objio_seg->num_comps; i++) {
238                 struct objio_dev_ent *ode;
239
240                 ode = _device_lookup(pnfslay, objio_seg, i, gfp_flags);
241                 if (unlikely(IS_ERR(ode))) {
242                         err = PTR_ERR(ode);
243                         goto out;
244                 }
245                 objio_seg->ods[i] = ode;
246         }
247         err = 0;
248
249 out:
250         dprintk("%s: return=%d\n", __func__, err);
251         return err;
252 }
253
254 static int _verify_data_map(struct pnfs_osd_layout *layout)
255 {
256         struct pnfs_osd_data_map *data_map = &layout->olo_map;
257         u64 stripe_length;
258         u32 group_width;
259
260 /* FIXME: Only raid0 for now. if not go through MDS */
261         if (data_map->odm_raid_algorithm != PNFS_OSD_RAID_0) {
262                 printk(KERN_ERR "Only RAID_0 for now\n");
263                 return -ENOTSUPP;
264         }
265         if (0 != (data_map->odm_num_comps % (data_map->odm_mirror_cnt + 1))) {
266                 printk(KERN_ERR "Data Map wrong, num_comps=%u mirrors=%u\n",
267                           data_map->odm_num_comps, data_map->odm_mirror_cnt);
268                 return -EINVAL;
269         }
270
271         if (data_map->odm_group_width)
272                 group_width = data_map->odm_group_width;
273         else
274                 group_width = data_map->odm_num_comps /
275                                                 (data_map->odm_mirror_cnt + 1);
276
277         stripe_length = (u64)data_map->odm_stripe_unit * group_width;
278         if (stripe_length >= (1ULL << 32)) {
279                 printk(KERN_ERR "Total Stripe length(0x%llx)"
280                           " >= 32bit is not supported\n", _LLU(stripe_length));
281                 return -ENOTSUPP;
282         }
283
284         if (0 != (data_map->odm_stripe_unit & ~PAGE_MASK)) {
285                 printk(KERN_ERR "Stripe Unit(0x%llx)"
286                           " must be Multples of PAGE_SIZE(0x%lx)\n",
287                           _LLU(data_map->odm_stripe_unit), PAGE_SIZE);
288                 return -ENOTSUPP;
289         }
290
291         return 0;
292 }
293
294 static void copy_single_comp(struct pnfs_osd_object_cred *cur_comp,
295                              struct pnfs_osd_object_cred *src_comp,
296                              struct caps_buffers *caps_p)
297 {
298         WARN_ON(src_comp->oc_cap_key.cred_len > sizeof(caps_p->caps_key));
299         WARN_ON(src_comp->oc_cap.cred_len > sizeof(caps_p->creds));
300
301         *cur_comp = *src_comp;
302
303         memcpy(caps_p->caps_key, src_comp->oc_cap_key.cred,
304                sizeof(caps_p->caps_key));
305         cur_comp->oc_cap_key.cred = caps_p->caps_key;
306
307         memcpy(caps_p->creds, src_comp->oc_cap.cred,
308                sizeof(caps_p->creds));
309         cur_comp->oc_cap.cred = caps_p->creds;
310 }
311
312 int objio_alloc_lseg(struct pnfs_layout_segment **outp,
313         struct pnfs_layout_hdr *pnfslay,
314         struct pnfs_layout_range *range,
315         struct xdr_stream *xdr,
316         gfp_t gfp_flags)
317 {
318         struct objio_segment *objio_seg;
319         struct pnfs_osd_xdr_decode_layout_iter iter;
320         struct pnfs_osd_layout layout;
321         struct pnfs_osd_object_cred *cur_comp, src_comp;
322         struct caps_buffers *caps_p;
323         int err;
324
325         err = pnfs_osd_xdr_decode_layout_map(&layout, &iter, xdr);
326         if (unlikely(err))
327                 return err;
328
329         err = _verify_data_map(&layout);
330         if (unlikely(err))
331                 return err;
332
333         objio_seg = kzalloc(sizeof(*objio_seg) +
334                             sizeof(objio_seg->ods[0]) * layout.olo_num_comps +
335                             sizeof(*objio_seg->comps) * layout.olo_num_comps +
336                             sizeof(struct caps_buffers) * layout.olo_num_comps,
337                             gfp_flags);
338         if (!objio_seg)
339                 return -ENOMEM;
340
341         objio_seg->comps = (void *)(objio_seg->ods + layout.olo_num_comps);
342         cur_comp = objio_seg->comps;
343         caps_p = (void *)(cur_comp + layout.olo_num_comps);
344         while (pnfs_osd_xdr_decode_layout_comp(&src_comp, &iter, xdr, &err))
345                 copy_single_comp(cur_comp++, &src_comp, caps_p++);
346         if (unlikely(err))
347                 goto err;
348
349         objio_seg->num_comps = layout.olo_num_comps;
350         objio_seg->comps_index = layout.olo_comps_index;
351         err = objio_devices_lookup(pnfslay, objio_seg, gfp_flags);
352         if (err)
353                 goto err;
354
355         objio_seg->mirrors_p1 = layout.olo_map.odm_mirror_cnt + 1;
356         objio_seg->stripe_unit = layout.olo_map.odm_stripe_unit;
357         if (layout.olo_map.odm_group_width) {
358                 objio_seg->group_width = layout.olo_map.odm_group_width;
359                 objio_seg->group_depth = layout.olo_map.odm_group_depth;
360                 objio_seg->group_count = layout.olo_map.odm_num_comps /
361                                                 objio_seg->mirrors_p1 /
362                                                 objio_seg->group_width;
363         } else {
364                 objio_seg->group_width = layout.olo_map.odm_num_comps /
365                                                 objio_seg->mirrors_p1;
366                 objio_seg->group_depth = -1;
367                 objio_seg->group_count = 1;
368         }
369
370         /* Cache this calculation it will hit for every page */
371         objio_seg->max_io_size = (BIO_MAX_PAGES_KMALLOC * PAGE_SIZE -
372                                   objio_seg->stripe_unit) *
373                                  objio_seg->group_width;
374
375         *outp = &objio_seg->lseg;
376         return 0;
377
378 err:
379         kfree(objio_seg);
380         dprintk("%s: Error: return %d\n", __func__, err);
381         *outp = NULL;
382         return err;
383 }
384
385 void objio_free_lseg(struct pnfs_layout_segment *lseg)
386 {
387         int i;
388         struct objio_segment *objio_seg = OBJIO_LSEG(lseg);
389
390         for (i = 0; i < objio_seg->num_comps; i++) {
391                 if (!objio_seg->ods[i])
392                         break;
393                 nfs4_put_deviceid_node(&objio_seg->ods[i]->id_node);
394         }
395         kfree(objio_seg);
396 }
397
398 int objio_alloc_io_state(struct pnfs_layout_segment *lseg,
399                          struct objlayout_io_state **outp,
400                          gfp_t gfp_flags)
401 {
402         struct objio_segment *objio_seg = OBJIO_LSEG(lseg);
403         struct objio_state *ios;
404         const unsigned first_size = sizeof(*ios) +
405                                 objio_seg->num_comps * sizeof(ios->per_dev[0]);
406         const unsigned sec_size = objio_seg->num_comps *
407                                                 sizeof(ios->ol_state.ioerrs[0]);
408
409         ios = kzalloc(first_size + sec_size, gfp_flags);
410         if (unlikely(!ios))
411                 return -ENOMEM;
412
413         ios->layout = objio_seg;
414         ios->ol_state.ioerrs = ((void *)ios) + first_size;
415         ios->ol_state.num_comps = objio_seg->num_comps;
416
417         *outp = &ios->ol_state;
418         return 0;
419 }
420
421 void objio_free_io_state(struct objlayout_io_state *ol_state)
422 {
423         struct objio_state *ios = container_of(ol_state, struct objio_state,
424                                                ol_state);
425
426         kfree(ios);
427 }
428
429 enum pnfs_osd_errno osd_pri_2_pnfs_err(enum osd_err_priority oep)
430 {
431         switch (oep) {
432         case OSD_ERR_PRI_NO_ERROR:
433                 return (enum pnfs_osd_errno)0;
434
435         case OSD_ERR_PRI_CLEAR_PAGES:
436                 BUG_ON(1);
437                 return 0;
438
439         case OSD_ERR_PRI_RESOURCE:
440                 return PNFS_OSD_ERR_RESOURCE;
441         case OSD_ERR_PRI_BAD_CRED:
442                 return PNFS_OSD_ERR_BAD_CRED;
443         case OSD_ERR_PRI_NO_ACCESS:
444                 return PNFS_OSD_ERR_NO_ACCESS;
445         case OSD_ERR_PRI_UNREACHABLE:
446                 return PNFS_OSD_ERR_UNREACHABLE;
447         case OSD_ERR_PRI_NOT_FOUND:
448                 return PNFS_OSD_ERR_NOT_FOUND;
449         case OSD_ERR_PRI_NO_SPACE:
450                 return PNFS_OSD_ERR_NO_SPACE;
451         default:
452                 WARN_ON(1);
453                 /* fallthrough */
454         case OSD_ERR_PRI_EIO:
455                 return PNFS_OSD_ERR_EIO;
456         }
457 }
458
459 static void _clear_bio(struct bio *bio)
460 {
461         struct bio_vec *bv;
462         unsigned i;
463
464         __bio_for_each_segment(bv, bio, i, 0) {
465                 unsigned this_count = bv->bv_len;
466
467                 if (likely(PAGE_SIZE == this_count))
468                         clear_highpage(bv->bv_page);
469                 else
470                         zero_user(bv->bv_page, bv->bv_offset, this_count);
471         }
472 }
473
474 static int _io_check(struct objio_state *ios, bool is_write)
475 {
476         enum osd_err_priority oep = OSD_ERR_PRI_NO_ERROR;
477         int lin_ret = 0;
478         int i;
479
480         for (i = 0; i <  ios->numdevs; i++) {
481                 struct osd_sense_info osi;
482                 struct osd_request *or = ios->per_dev[i].or;
483                 unsigned dev;
484                 int ret;
485
486                 if (!or)
487                         continue;
488
489                 ret = osd_req_decode_sense(or, &osi);
490                 if (likely(!ret))
491                         continue;
492
493                 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
494                         /* start read offset passed endof file */
495                         BUG_ON(is_write);
496                         _clear_bio(ios->per_dev[i].bio);
497                         dprintk("%s: start read offset passed end of file "
498                                 "offset=0x%llx, length=0x%lx\n", __func__,
499                                 _LLU(ios->per_dev[i].offset),
500                                 ios->per_dev[i].length);
501
502                         continue; /* we recovered */
503                 }
504                 dev = ios->per_dev[i].dev;
505                 objlayout_io_set_result(&ios->ol_state, dev,
506                                         &ios->layout->comps[dev].oc_object_id,
507                                         osd_pri_2_pnfs_err(osi.osd_err_pri),
508                                         ios->per_dev[i].offset,
509                                         ios->per_dev[i].length,
510                                         is_write);
511
512                 if (osi.osd_err_pri >= oep) {
513                         oep = osi.osd_err_pri;
514                         lin_ret = ret;
515                 }
516         }
517
518         return lin_ret;
519 }
520
521 /*
522  * Common IO state helpers.
523  */
524 static void _io_free(struct objio_state *ios)
525 {
526         unsigned i;
527
528         for (i = 0; i < ios->numdevs; i++) {
529                 struct _objio_per_comp *per_dev = &ios->per_dev[i];
530
531                 if (per_dev->or) {
532                         osd_end_request(per_dev->or);
533                         per_dev->or = NULL;
534                 }
535
536                 if (per_dev->bio) {
537                         bio_put(per_dev->bio);
538                         per_dev->bio = NULL;
539                 }
540         }
541 }
542
543 struct osd_dev *_io_od(struct objio_state *ios, unsigned dev)
544 {
545         unsigned min_dev = ios->layout->comps_index;
546         unsigned max_dev = min_dev + ios->layout->num_comps;
547
548         BUG_ON(dev < min_dev || max_dev <= dev);
549         return ios->layout->ods[dev - min_dev]->od;
550 }
551
552 struct _striping_info {
553         u64 obj_offset;
554         u64 group_length;
555         unsigned dev;
556         unsigned unit_off;
557 };
558
559 static void _calc_stripe_info(struct objio_state *ios, u64 file_offset,
560                               struct _striping_info *si)
561 {
562         u32     stripe_unit = ios->layout->stripe_unit;
563         u32     group_width = ios->layout->group_width;
564         u64     group_depth = ios->layout->group_depth;
565         u32     U = stripe_unit * group_width;
566
567         u64     T = U * group_depth;
568         u64     S = T * ios->layout->group_count;
569         u64     M = div64_u64(file_offset, S);
570
571         /*
572         G = (L - (M * S)) / T
573         H = (L - (M * S)) % T
574         */
575         u64     LmodU = file_offset - M * S;
576         u32     G = div64_u64(LmodU, T);
577         u64     H = LmodU - G * T;
578
579         u32     N = div_u64(H, U);
580
581         div_u64_rem(file_offset, stripe_unit, &si->unit_off);
582         si->obj_offset = si->unit_off + (N * stripe_unit) +
583                                   (M * group_depth * stripe_unit);
584
585         /* "H - (N * U)" is just "H % U" so it's bound to u32 */
586         si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
587         si->dev *= ios->layout->mirrors_p1;
588
589         si->group_length = T - H;
590 }
591
592 static int _add_stripe_unit(struct objio_state *ios,  unsigned *cur_pg,
593                 unsigned pgbase, struct _objio_per_comp *per_dev, int cur_len,
594                 gfp_t gfp_flags)
595 {
596         unsigned pg = *cur_pg;
597         struct request_queue *q =
598                         osd_request_queue(_io_od(ios, per_dev->dev));
599
600         per_dev->length += cur_len;
601
602         if (per_dev->bio == NULL) {
603                 unsigned stripes = ios->layout->num_comps /
604                                                      ios->layout->mirrors_p1;
605                 unsigned pages_in_stripe = stripes *
606                                       (ios->layout->stripe_unit / PAGE_SIZE);
607                 unsigned bio_size = (ios->ol_state.nr_pages + pages_in_stripe) /
608                                     stripes;
609
610                 if (BIO_MAX_PAGES_KMALLOC < bio_size)
611                         bio_size = BIO_MAX_PAGES_KMALLOC;
612
613                 per_dev->bio = bio_kmalloc(gfp_flags, bio_size);
614                 if (unlikely(!per_dev->bio)) {
615                         dprintk("Faild to allocate BIO size=%u\n", bio_size);
616                         return -ENOMEM;
617                 }
618         }
619
620         while (cur_len > 0) {
621                 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
622                 unsigned added_len;
623
624                 BUG_ON(ios->ol_state.nr_pages <= pg);
625                 cur_len -= pglen;
626
627                 added_len = bio_add_pc_page(q, per_dev->bio,
628                                         ios->ol_state.pages[pg], pglen, pgbase);
629                 if (unlikely(pglen != added_len))
630                         return -ENOMEM;
631                 pgbase = 0;
632                 ++pg;
633         }
634         BUG_ON(cur_len);
635
636         *cur_pg = pg;
637         return 0;
638 }
639
640 static int _prepare_one_group(struct objio_state *ios, u64 length,
641                               struct _striping_info *si, unsigned *last_pg,
642                               gfp_t gfp_flags)
643 {
644         unsigned stripe_unit = ios->layout->stripe_unit;
645         unsigned mirrors_p1 = ios->layout->mirrors_p1;
646         unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
647         unsigned dev = si->dev;
648         unsigned first_dev = dev - (dev % devs_in_group);
649         unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
650         unsigned cur_pg = *last_pg;
651         int ret = 0;
652
653         while (length) {
654                 struct _objio_per_comp *per_dev = &ios->per_dev[dev];
655                 unsigned cur_len, page_off = 0;
656
657                 if (!per_dev->length) {
658                         per_dev->dev = dev;
659                         if (dev < si->dev) {
660                                 per_dev->offset = si->obj_offset + stripe_unit -
661                                                                    si->unit_off;
662                                 cur_len = stripe_unit;
663                         } else if (dev == si->dev) {
664                                 per_dev->offset = si->obj_offset;
665                                 cur_len = stripe_unit - si->unit_off;
666                                 page_off = si->unit_off & ~PAGE_MASK;
667                                 BUG_ON(page_off &&
668                                       (page_off != ios->ol_state.pgbase));
669                         } else { /* dev > si->dev */
670                                 per_dev->offset = si->obj_offset - si->unit_off;
671                                 cur_len = stripe_unit;
672                         }
673
674                         if (max_comp < dev)
675                                 max_comp = dev;
676                 } else {
677                         cur_len = stripe_unit;
678                 }
679                 if (cur_len >= length)
680                         cur_len = length;
681
682                 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
683                                        cur_len, gfp_flags);
684                 if (unlikely(ret))
685                         goto out;
686
687                 dev += mirrors_p1;
688                 dev = (dev % devs_in_group) + first_dev;
689
690                 length -= cur_len;
691                 ios->length += cur_len;
692         }
693 out:
694         ios->numdevs = max_comp + mirrors_p1;
695         *last_pg = cur_pg;
696         return ret;
697 }
698
699 static int _io_rw_pagelist(struct objio_state *ios, gfp_t gfp_flags)
700 {
701         u64 length = ios->ol_state.count;
702         u64 offset = ios->ol_state.offset;
703         struct _striping_info si;
704         unsigned last_pg = 0;
705         int ret = 0;
706
707         while (length) {
708                 _calc_stripe_info(ios, offset, &si);
709
710                 if (length < si.group_length)
711                         si.group_length = length;
712
713                 ret = _prepare_one_group(ios, si.group_length, &si, &last_pg, gfp_flags);
714                 if (unlikely(ret))
715                         goto out;
716
717                 offset += si.group_length;
718                 length -= si.group_length;
719         }
720
721 out:
722         if (!ios->length)
723                 return ret;
724
725         return 0;
726 }
727
728 static ssize_t _sync_done(struct objio_state *ios)
729 {
730         struct completion *waiting = ios->private;
731
732         complete(waiting);
733         return 0;
734 }
735
736 static void _last_io(struct kref *kref)
737 {
738         struct objio_state *ios = container_of(kref, struct objio_state, kref);
739
740         ios->done(ios);
741 }
742
743 static void _done_io(struct osd_request *or, void *p)
744 {
745         struct objio_state *ios = p;
746
747         kref_put(&ios->kref, _last_io);
748 }
749
750 static ssize_t _io_exec(struct objio_state *ios)
751 {
752         DECLARE_COMPLETION_ONSTACK(wait);
753         ssize_t status = 0; /* sync status */
754         unsigned i;
755         objio_done_fn saved_done_fn = ios->done;
756         bool sync = ios->ol_state.sync;
757
758         if (sync) {
759                 ios->done = _sync_done;
760                 ios->private = &wait;
761         }
762
763         kref_init(&ios->kref);
764
765         for (i = 0; i < ios->numdevs; i++) {
766                 struct osd_request *or = ios->per_dev[i].or;
767
768                 if (!or)
769                         continue;
770
771                 kref_get(&ios->kref);
772                 osd_execute_request_async(or, _done_io, ios);
773         }
774
775         kref_put(&ios->kref, _last_io);
776
777         if (sync) {
778                 wait_for_completion(&wait);
779                 status = saved_done_fn(ios);
780         }
781
782         return status;
783 }
784
785 /*
786  * read
787  */
788 static ssize_t _read_done(struct objio_state *ios)
789 {
790         ssize_t status;
791         int ret = _io_check(ios, false);
792
793         _io_free(ios);
794
795         if (likely(!ret))
796                 status = ios->length;
797         else
798                 status = ret;
799
800         objlayout_read_done(&ios->ol_state, status, ios->ol_state.sync);
801         return status;
802 }
803
804 static int _read_mirrors(struct objio_state *ios, unsigned cur_comp)
805 {
806         struct osd_request *or = NULL;
807         struct _objio_per_comp *per_dev = &ios->per_dev[cur_comp];
808         unsigned dev = per_dev->dev;
809         struct pnfs_osd_object_cred *cred =
810                         &ios->layout->comps[dev];
811         struct osd_obj_id obj = {
812                 .partition = cred->oc_object_id.oid_partition_id,
813                 .id = cred->oc_object_id.oid_object_id,
814         };
815         int ret;
816
817         or = osd_start_request(_io_od(ios, dev), GFP_KERNEL);
818         if (unlikely(!or)) {
819                 ret = -ENOMEM;
820                 goto err;
821         }
822         per_dev->or = or;
823
824         osd_req_read(or, &obj, per_dev->offset, per_dev->bio, per_dev->length);
825
826         ret = osd_finalize_request(or, 0, cred->oc_cap.cred, NULL);
827         if (ret) {
828                 dprintk("%s: Faild to osd_finalize_request() => %d\n",
829                         __func__, ret);
830                 goto err;
831         }
832
833         dprintk("%s:[%d] dev=%d obj=0x%llx start=0x%llx length=0x%lx\n",
834                 __func__, cur_comp, dev, obj.id, _LLU(per_dev->offset),
835                 per_dev->length);
836
837 err:
838         return ret;
839 }
840
841 static ssize_t _read_exec(struct objio_state *ios)
842 {
843         unsigned i;
844         int ret;
845
846         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
847                 if (!ios->per_dev[i].length)
848                         continue;
849                 ret = _read_mirrors(ios, i);
850                 if (unlikely(ret))
851                         goto err;
852         }
853
854         ios->done = _read_done;
855         return _io_exec(ios); /* In sync mode exec returns the io status */
856
857 err:
858         _io_free(ios);
859         return ret;
860 }
861
862 ssize_t objio_read_pagelist(struct objlayout_io_state *ol_state)
863 {
864         struct objio_state *ios = container_of(ol_state, struct objio_state,
865                                                ol_state);
866         int ret;
867
868         ret = _io_rw_pagelist(ios, GFP_KERNEL);
869         if (unlikely(ret))
870                 return ret;
871
872         return _read_exec(ios);
873 }
874
875 /*
876  * write
877  */
878 static ssize_t _write_done(struct objio_state *ios)
879 {
880         ssize_t status;
881         int ret = _io_check(ios, true);
882
883         _io_free(ios);
884
885         if (likely(!ret)) {
886                 /* FIXME: should be based on the OSD's persistence model
887                  * See OSD2r05 Section 4.13 Data persistence model */
888                 ios->ol_state.committed = NFS_FILE_SYNC;
889                 status = ios->length;
890         } else {
891                 status = ret;
892         }
893
894         objlayout_write_done(&ios->ol_state, status, ios->ol_state.sync);
895         return status;
896 }
897
898 static int _write_mirrors(struct objio_state *ios, unsigned cur_comp)
899 {
900         struct _objio_per_comp *master_dev = &ios->per_dev[cur_comp];
901         unsigned dev = ios->per_dev[cur_comp].dev;
902         unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
903         int ret;
904
905         for (; cur_comp < last_comp; ++cur_comp, ++dev) {
906                 struct osd_request *or = NULL;
907                 struct pnfs_osd_object_cred *cred =
908                                         &ios->layout->comps[dev];
909                 struct osd_obj_id obj = {
910                         .partition = cred->oc_object_id.oid_partition_id,
911                         .id = cred->oc_object_id.oid_object_id,
912                 };
913                 struct _objio_per_comp *per_dev = &ios->per_dev[cur_comp];
914                 struct bio *bio;
915
916                 or = osd_start_request(_io_od(ios, dev), GFP_NOFS);
917                 if (unlikely(!or)) {
918                         ret = -ENOMEM;
919                         goto err;
920                 }
921                 per_dev->or = or;
922
923                 if (per_dev != master_dev) {
924                         bio = bio_kmalloc(GFP_NOFS,
925                                           master_dev->bio->bi_max_vecs);
926                         if (unlikely(!bio)) {
927                                 dprintk("Faild to allocate BIO size=%u\n",
928                                         master_dev->bio->bi_max_vecs);
929                                 ret = -ENOMEM;
930                                 goto err;
931                         }
932
933                         __bio_clone(bio, master_dev->bio);
934                         bio->bi_bdev = NULL;
935                         bio->bi_next = NULL;
936                         per_dev->bio = bio;
937                         per_dev->dev = dev;
938                         per_dev->length = master_dev->length;
939                         per_dev->offset =  master_dev->offset;
940                 } else {
941                         bio = master_dev->bio;
942                         bio->bi_rw |= REQ_WRITE;
943                 }
944
945                 osd_req_write(or, &obj, per_dev->offset, bio, per_dev->length);
946
947                 ret = osd_finalize_request(or, 0, cred->oc_cap.cred, NULL);
948                 if (ret) {
949                         dprintk("%s: Faild to osd_finalize_request() => %d\n",
950                                 __func__, ret);
951                         goto err;
952                 }
953
954                 dprintk("%s:[%d] dev=%d obj=0x%llx start=0x%llx length=0x%lx\n",
955                         __func__, cur_comp, dev, obj.id, _LLU(per_dev->offset),
956                         per_dev->length);
957         }
958
959 err:
960         return ret;
961 }
962
963 static ssize_t _write_exec(struct objio_state *ios)
964 {
965         unsigned i;
966         int ret;
967
968         for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
969                 if (!ios->per_dev[i].length)
970                         continue;
971                 ret = _write_mirrors(ios, i);
972                 if (unlikely(ret))
973                         goto err;
974         }
975
976         ios->done = _write_done;
977         return _io_exec(ios); /* In sync mode exec returns the io->status */
978
979 err:
980         _io_free(ios);
981         return ret;
982 }
983
984 ssize_t objio_write_pagelist(struct objlayout_io_state *ol_state, bool stable)
985 {
986         struct objio_state *ios = container_of(ol_state, struct objio_state,
987                                                ol_state);
988         int ret;
989
990         /* TODO: ios->stable = stable; */
991         ret = _io_rw_pagelist(ios, GFP_NOFS);
992         if (unlikely(ret))
993                 return ret;
994
995         return _write_exec(ios);
996 }
997
998 static bool objio_pg_test(struct nfs_pageio_descriptor *pgio,
999                           struct nfs_page *prev, struct nfs_page *req)
1000 {
1001         if (!pnfs_generic_pg_test(pgio, prev, req))
1002                 return false;
1003
1004         return pgio->pg_count + req->wb_bytes <=
1005                         OBJIO_LSEG(pgio->pg_lseg)->max_io_size;
1006 }
1007
1008 static struct pnfs_layoutdriver_type objlayout_type = {
1009         .id = LAYOUT_OSD2_OBJECTS,
1010         .name = "LAYOUT_OSD2_OBJECTS",
1011         .flags                   = PNFS_LAYOUTRET_ON_SETATTR,
1012
1013         .alloc_layout_hdr        = objlayout_alloc_layout_hdr,
1014         .free_layout_hdr         = objlayout_free_layout_hdr,
1015
1016         .alloc_lseg              = objlayout_alloc_lseg,
1017         .free_lseg               = objlayout_free_lseg,
1018
1019         .read_pagelist           = objlayout_read_pagelist,
1020         .write_pagelist          = objlayout_write_pagelist,
1021         .pg_test                 = objio_pg_test,
1022
1023         .free_deviceid_node      = objio_free_deviceid_node,
1024
1025         .encode_layoutcommit     = objlayout_encode_layoutcommit,
1026         .encode_layoutreturn     = objlayout_encode_layoutreturn,
1027 };
1028
1029 MODULE_DESCRIPTION("pNFS Layout Driver for OSD2 objects");
1030 MODULE_AUTHOR("Benny Halevy <bhalevy@panasas.com>");
1031 MODULE_LICENSE("GPL");
1032
1033 static int __init
1034 objlayout_init(void)
1035 {
1036         int ret = pnfs_register_layoutdriver(&objlayout_type);
1037
1038         if (ret)
1039                 printk(KERN_INFO
1040                         "%s: Registering OSD pNFS Layout Driver failed: error=%d\n",
1041                         __func__, ret);
1042         else
1043                 printk(KERN_INFO "%s: Registered OSD pNFS Layout Driver\n",
1044                         __func__);
1045         return ret;
1046 }
1047
1048 static void __exit
1049 objlayout_exit(void)
1050 {
1051         pnfs_unregister_layoutdriver(&objlayout_type);
1052         printk(KERN_INFO "%s: Unregistered OSD pNFS Layout Driver\n",
1053                __func__);
1054 }
1055
1056 module_init(objlayout_init);
1057 module_exit(objlayout_exit);