Loading...
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_ore.h>
42
43#include "objlayout.h"
44
45#define NFSDBG_FACILITY NFSDBG_PNFS_LD
46
47struct objio_dev_ent {
48 struct nfs4_deviceid_node id_node;
49 struct ore_dev od;
50};
51
52static void
53objio_free_deviceid_node(struct nfs4_deviceid_node *d)
54{
55 struct objio_dev_ent *de = container_of(d, struct objio_dev_ent, id_node);
56
57 dprintk("%s: free od=%p\n", __func__, de->od.od);
58 osduld_put_device(de->od.od);
59 kfree(de);
60}
61
62static struct objio_dev_ent *_dev_list_find(const struct nfs_server *nfss,
63 const struct nfs4_deviceid *d_id)
64{
65 struct nfs4_deviceid_node *d;
66 struct objio_dev_ent *de;
67
68 d = nfs4_find_get_deviceid(nfss->pnfs_curr_ld, nfss->nfs_client, d_id);
69 if (!d)
70 return NULL;
71
72 de = container_of(d, struct objio_dev_ent, id_node);
73 return de;
74}
75
76static struct objio_dev_ent *
77_dev_list_add(const struct nfs_server *nfss,
78 const struct nfs4_deviceid *d_id, struct osd_dev *od,
79 gfp_t gfp_flags)
80{
81 struct nfs4_deviceid_node *d;
82 struct objio_dev_ent *de = kzalloc(sizeof(*de), gfp_flags);
83 struct objio_dev_ent *n;
84
85 if (!de) {
86 dprintk("%s: -ENOMEM od=%p\n", __func__, od);
87 return NULL;
88 }
89
90 dprintk("%s: Adding od=%p\n", __func__, od);
91 nfs4_init_deviceid_node(&de->id_node,
92 nfss->pnfs_curr_ld,
93 nfss->nfs_client,
94 d_id);
95 de->od.od = od;
96
97 d = nfs4_insert_deviceid_node(&de->id_node);
98 n = container_of(d, struct objio_dev_ent, id_node);
99 if (n != de) {
100 dprintk("%s: Race with other n->od=%p\n", __func__, n->od.od);
101 objio_free_deviceid_node(&de->id_node);
102 de = n;
103 }
104
105 return de;
106}
107
108struct objio_segment {
109 struct pnfs_layout_segment lseg;
110
111 struct ore_layout layout;
112 struct ore_components oc;
113};
114
115static inline struct objio_segment *
116OBJIO_LSEG(struct pnfs_layout_segment *lseg)
117{
118 return container_of(lseg, struct objio_segment, lseg);
119}
120
121struct objio_state {
122 /* Generic layer */
123 struct objlayout_io_res oir;
124
125 bool sync;
126 /*FIXME: Support for extra_bytes at ore_get_rw_state() */
127 struct ore_io_state *ios;
128};
129
130/* Send and wait for a get_device_info of devices in the layout,
131 then look them up with the osd_initiator library */
132static int objio_devices_lookup(struct pnfs_layout_hdr *pnfslay,
133 struct objio_segment *objio_seg, unsigned c, struct nfs4_deviceid *d_id,
134 gfp_t gfp_flags)
135{
136 struct pnfs_osd_deviceaddr *deviceaddr;
137 struct objio_dev_ent *ode;
138 struct osd_dev *od;
139 struct osd_dev_info odi;
140 bool retry_flag = true;
141 int err;
142
143 ode = _dev_list_find(NFS_SERVER(pnfslay->plh_inode), d_id);
144 if (ode) {
145 objio_seg->oc.ods[c] = &ode->od; /* must use container_of */
146 return 0;
147 }
148
149 err = objlayout_get_deviceinfo(pnfslay, d_id, &deviceaddr, gfp_flags);
150 if (unlikely(err)) {
151 dprintk("%s: objlayout_get_deviceinfo dev(%llx:%llx) =>%d\n",
152 __func__, _DEVID_LO(d_id), _DEVID_HI(d_id), err);
153 return err;
154 }
155
156 odi.systemid_len = deviceaddr->oda_systemid.len;
157 if (odi.systemid_len > sizeof(odi.systemid)) {
158 dprintk("%s: odi.systemid_len > sizeof(systemid=%zd)\n",
159 __func__, sizeof(odi.systemid));
160 err = -EINVAL;
161 goto out;
162 } else if (odi.systemid_len)
163 memcpy(odi.systemid, deviceaddr->oda_systemid.data,
164 odi.systemid_len);
165 odi.osdname_len = deviceaddr->oda_osdname.len;
166 odi.osdname = (u8 *)deviceaddr->oda_osdname.data;
167
168 if (!odi.osdname_len && !odi.systemid_len) {
169 dprintk("%s: !odi.osdname_len && !odi.systemid_len\n",
170 __func__);
171 err = -ENODEV;
172 goto out;
173 }
174
175retry_lookup:
176 od = osduld_info_lookup(&odi);
177 if (unlikely(IS_ERR(od))) {
178 err = PTR_ERR(od);
179 dprintk("%s: osduld_info_lookup => %d\n", __func__, err);
180 if (err == -ENODEV && retry_flag) {
181 err = objlayout_autologin(deviceaddr);
182 if (likely(!err)) {
183 retry_flag = false;
184 goto retry_lookup;
185 }
186 }
187 goto out;
188 }
189
190 ode = _dev_list_add(NFS_SERVER(pnfslay->plh_inode), d_id, od,
191 gfp_flags);
192 objio_seg->oc.ods[c] = &ode->od; /* must use container_of */
193 dprintk("Adding new dev_id(%llx:%llx)\n",
194 _DEVID_LO(d_id), _DEVID_HI(d_id));
195out:
196 objlayout_put_deviceinfo(deviceaddr);
197 return err;
198}
199
200static void copy_single_comp(struct ore_components *oc, unsigned c,
201 struct pnfs_osd_object_cred *src_comp)
202{
203 struct ore_comp *ocomp = &oc->comps[c];
204
205 WARN_ON(src_comp->oc_cap_key.cred_len > 0); /* libosd is NO_SEC only */
206 WARN_ON(src_comp->oc_cap.cred_len > sizeof(ocomp->cred));
207
208 ocomp->obj.partition = src_comp->oc_object_id.oid_partition_id;
209 ocomp->obj.id = src_comp->oc_object_id.oid_object_id;
210
211 memcpy(ocomp->cred, src_comp->oc_cap.cred, sizeof(ocomp->cred));
212}
213
214static int __alloc_objio_seg(unsigned numdevs, gfp_t gfp_flags,
215 struct objio_segment **pseg)
216{
217/* This is the in memory structure of the objio_segment
218 *
219 * struct __alloc_objio_segment {
220 * struct objio_segment olseg;
221 * struct ore_dev *ods[numdevs];
222 * struct ore_comp comps[numdevs];
223 * } *aolseg;
224 * NOTE: The code as above compiles and runs perfectly. It is elegant,
225 * type safe and compact. At some Past time Linus has decided he does not
226 * like variable length arrays, For the sake of this principal we uglify
227 * the code as below.
228 */
229 struct objio_segment *lseg;
230 size_t lseg_size = sizeof(*lseg) +
231 numdevs * sizeof(lseg->oc.ods[0]) +
232 numdevs * sizeof(*lseg->oc.comps);
233
234 lseg = kzalloc(lseg_size, gfp_flags);
235 if (unlikely(!lseg)) {
236 dprintk("%s: Faild allocation numdevs=%d size=%zd\n", __func__,
237 numdevs, lseg_size);
238 return -ENOMEM;
239 }
240
241 lseg->oc.numdevs = numdevs;
242 lseg->oc.single_comp = EC_MULTPLE_COMPS;
243 lseg->oc.ods = (void *)(lseg + 1);
244 lseg->oc.comps = (void *)(lseg->oc.ods + numdevs);
245
246 *pseg = lseg;
247 return 0;
248}
249
250int objio_alloc_lseg(struct pnfs_layout_segment **outp,
251 struct pnfs_layout_hdr *pnfslay,
252 struct pnfs_layout_range *range,
253 struct xdr_stream *xdr,
254 gfp_t gfp_flags)
255{
256 struct objio_segment *objio_seg;
257 struct pnfs_osd_xdr_decode_layout_iter iter;
258 struct pnfs_osd_layout layout;
259 struct pnfs_osd_object_cred src_comp;
260 unsigned cur_comp;
261 int err;
262
263 err = pnfs_osd_xdr_decode_layout_map(&layout, &iter, xdr);
264 if (unlikely(err))
265 return err;
266
267 err = __alloc_objio_seg(layout.olo_num_comps, gfp_flags, &objio_seg);
268 if (unlikely(err))
269 return err;
270
271 objio_seg->layout.stripe_unit = layout.olo_map.odm_stripe_unit;
272 objio_seg->layout.group_width = layout.olo_map.odm_group_width;
273 objio_seg->layout.group_depth = layout.olo_map.odm_group_depth;
274 objio_seg->layout.mirrors_p1 = layout.olo_map.odm_mirror_cnt + 1;
275 objio_seg->layout.raid_algorithm = layout.olo_map.odm_raid_algorithm;
276
277 err = ore_verify_layout(layout.olo_map.odm_num_comps,
278 &objio_seg->layout);
279 if (unlikely(err))
280 goto err;
281
282 objio_seg->oc.first_dev = layout.olo_comps_index;
283 cur_comp = 0;
284 while (pnfs_osd_xdr_decode_layout_comp(&src_comp, &iter, xdr, &err)) {
285 copy_single_comp(&objio_seg->oc, cur_comp, &src_comp);
286 err = objio_devices_lookup(pnfslay, objio_seg, cur_comp,
287 &src_comp.oc_object_id.oid_device_id,
288 gfp_flags);
289 if (err)
290 goto err;
291 ++cur_comp;
292 }
293 /* pnfs_osd_xdr_decode_layout_comp returns false on error */
294 if (unlikely(err))
295 goto err;
296
297 *outp = &objio_seg->lseg;
298 return 0;
299
300err:
301 kfree(objio_seg);
302 dprintk("%s: Error: return %d\n", __func__, err);
303 *outp = NULL;
304 return err;
305}
306
307void objio_free_lseg(struct pnfs_layout_segment *lseg)
308{
309 int i;
310 struct objio_segment *objio_seg = OBJIO_LSEG(lseg);
311
312 for (i = 0; i < objio_seg->oc.numdevs; i++) {
313 struct ore_dev *od = objio_seg->oc.ods[i];
314 struct objio_dev_ent *ode;
315
316 if (!od)
317 break;
318 ode = container_of(od, typeof(*ode), od);
319 nfs4_put_deviceid_node(&ode->id_node);
320 }
321 kfree(objio_seg);
322}
323
324static int
325objio_alloc_io_state(struct pnfs_layout_hdr *pnfs_layout_type, bool is_reading,
326 struct pnfs_layout_segment *lseg, struct page **pages, unsigned pgbase,
327 loff_t offset, size_t count, void *rpcdata, gfp_t gfp_flags,
328 struct objio_state **outp)
329{
330 struct objio_segment *objio_seg = OBJIO_LSEG(lseg);
331 struct ore_io_state *ios;
332 int ret;
333 struct __alloc_objio_state {
334 struct objio_state objios;
335 struct pnfs_osd_ioerr ioerrs[objio_seg->oc.numdevs];
336 } *aos;
337
338 aos = kzalloc(sizeof(*aos), gfp_flags);
339 if (unlikely(!aos))
340 return -ENOMEM;
341
342 objlayout_init_ioerrs(&aos->objios.oir, objio_seg->oc.numdevs,
343 aos->ioerrs, rpcdata, pnfs_layout_type);
344
345 ret = ore_get_rw_state(&objio_seg->layout, &objio_seg->oc, is_reading,
346 offset, count, &ios);
347 if (unlikely(ret)) {
348 kfree(aos);
349 return ret;
350 }
351
352 ios->pages = pages;
353 ios->pgbase = pgbase;
354 ios->private = aos;
355 BUG_ON(ios->nr_pages > (pgbase + count + PAGE_SIZE - 1) >> PAGE_SHIFT);
356
357 aos->objios.sync = 0;
358 aos->objios.ios = ios;
359 *outp = &aos->objios;
360 return 0;
361}
362
363void objio_free_result(struct objlayout_io_res *oir)
364{
365 struct objio_state *objios = container_of(oir, struct objio_state, oir);
366
367 ore_put_io_state(objios->ios);
368 kfree(objios);
369}
370
371enum pnfs_osd_errno osd_pri_2_pnfs_err(enum osd_err_priority oep)
372{
373 switch (oep) {
374 case OSD_ERR_PRI_NO_ERROR:
375 return (enum pnfs_osd_errno)0;
376
377 case OSD_ERR_PRI_CLEAR_PAGES:
378 BUG_ON(1);
379 return 0;
380
381 case OSD_ERR_PRI_RESOURCE:
382 return PNFS_OSD_ERR_RESOURCE;
383 case OSD_ERR_PRI_BAD_CRED:
384 return PNFS_OSD_ERR_BAD_CRED;
385 case OSD_ERR_PRI_NO_ACCESS:
386 return PNFS_OSD_ERR_NO_ACCESS;
387 case OSD_ERR_PRI_UNREACHABLE:
388 return PNFS_OSD_ERR_UNREACHABLE;
389 case OSD_ERR_PRI_NOT_FOUND:
390 return PNFS_OSD_ERR_NOT_FOUND;
391 case OSD_ERR_PRI_NO_SPACE:
392 return PNFS_OSD_ERR_NO_SPACE;
393 default:
394 WARN_ON(1);
395 /* fallthrough */
396 case OSD_ERR_PRI_EIO:
397 return PNFS_OSD_ERR_EIO;
398 }
399}
400
401static void __on_dev_error(struct ore_io_state *ios,
402 struct ore_dev *od, unsigned dev_index, enum osd_err_priority oep,
403 u64 dev_offset, u64 dev_len)
404{
405 struct objio_state *objios = ios->private;
406 struct pnfs_osd_objid pooid;
407 struct objio_dev_ent *ode = container_of(od, typeof(*ode), od);
408 /* FIXME: what to do with more-then-one-group layouts. We need to
409 * translate from ore_io_state index to oc->comps index
410 */
411 unsigned comp = dev_index;
412
413 pooid.oid_device_id = ode->id_node.deviceid;
414 pooid.oid_partition_id = ios->oc->comps[comp].obj.partition;
415 pooid.oid_object_id = ios->oc->comps[comp].obj.id;
416
417 objlayout_io_set_result(&objios->oir, comp,
418 &pooid, osd_pri_2_pnfs_err(oep),
419 dev_offset, dev_len, !ios->reading);
420}
421
422/*
423 * read
424 */
425static void _read_done(struct ore_io_state *ios, void *private)
426{
427 struct objio_state *objios = private;
428 ssize_t status;
429 int ret = ore_check_io(ios, &__on_dev_error);
430
431 /* FIXME: _io_free(ios) can we dealocate the libosd resources; */
432
433 if (likely(!ret))
434 status = ios->length;
435 else
436 status = ret;
437
438 objlayout_read_done(&objios->oir, status, objios->sync);
439}
440
441int objio_read_pagelist(struct nfs_read_data *rdata)
442{
443 struct nfs_pgio_header *hdr = rdata->header;
444 struct objio_state *objios;
445 int ret;
446
447 ret = objio_alloc_io_state(NFS_I(hdr->inode)->layout, true,
448 hdr->lseg, rdata->args.pages, rdata->args.pgbase,
449 rdata->args.offset, rdata->args.count, rdata,
450 GFP_KERNEL, &objios);
451 if (unlikely(ret))
452 return ret;
453
454 objios->ios->done = _read_done;
455 dprintk("%s: offset=0x%llx length=0x%x\n", __func__,
456 rdata->args.offset, rdata->args.count);
457 ret = ore_read(objios->ios);
458 if (unlikely(ret))
459 objio_free_result(&objios->oir);
460 return ret;
461}
462
463/*
464 * write
465 */
466static void _write_done(struct ore_io_state *ios, void *private)
467{
468 struct objio_state *objios = private;
469 ssize_t status;
470 int ret = ore_check_io(ios, &__on_dev_error);
471
472 /* FIXME: _io_free(ios) can we dealocate the libosd resources; */
473
474 if (likely(!ret)) {
475 /* FIXME: should be based on the OSD's persistence model
476 * See OSD2r05 Section 4.13 Data persistence model */
477 objios->oir.committed = NFS_FILE_SYNC;
478 status = ios->length;
479 } else {
480 status = ret;
481 }
482
483 objlayout_write_done(&objios->oir, status, objios->sync);
484}
485
486static struct page *__r4w_get_page(void *priv, u64 offset, bool *uptodate)
487{
488 struct objio_state *objios = priv;
489 struct nfs_write_data *wdata = objios->oir.rpcdata;
490 struct address_space *mapping = wdata->header->inode->i_mapping;
491 pgoff_t index = offset / PAGE_SIZE;
492 struct page *page;
493 loff_t i_size = i_size_read(wdata->header->inode);
494
495 if (offset >= i_size) {
496 *uptodate = true;
497 dprintk("%s: g_zero_page index=0x%lx\n", __func__, index);
498 return ZERO_PAGE(0);
499 }
500
501 page = find_get_page(mapping, index);
502 if (!page) {
503 page = find_or_create_page(mapping, index, GFP_NOFS);
504 if (unlikely(!page)) {
505 dprintk("%s: grab_cache_page Failed index=0x%lx\n",
506 __func__, index);
507 return NULL;
508 }
509 unlock_page(page);
510 }
511 if (PageDirty(page) || PageWriteback(page))
512 *uptodate = true;
513 else
514 *uptodate = PageUptodate(page);
515 dprintk("%s: index=0x%lx uptodate=%d\n", __func__, index, *uptodate);
516 return page;
517}
518
519static void __r4w_put_page(void *priv, struct page *page)
520{
521 dprintk("%s: index=0x%lx\n", __func__,
522 (page == ZERO_PAGE(0)) ? -1UL : page->index);
523 if (ZERO_PAGE(0) != page)
524 page_cache_release(page);
525 return;
526}
527
528static const struct _ore_r4w_op _r4w_op = {
529 .get_page = &__r4w_get_page,
530 .put_page = &__r4w_put_page,
531};
532
533int objio_write_pagelist(struct nfs_write_data *wdata, int how)
534{
535 struct nfs_pgio_header *hdr = wdata->header;
536 struct objio_state *objios;
537 int ret;
538
539 ret = objio_alloc_io_state(NFS_I(hdr->inode)->layout, false,
540 hdr->lseg, wdata->args.pages, wdata->args.pgbase,
541 wdata->args.offset, wdata->args.count, wdata, GFP_NOFS,
542 &objios);
543 if (unlikely(ret))
544 return ret;
545
546 objios->sync = 0 != (how & FLUSH_SYNC);
547 objios->ios->r4w = &_r4w_op;
548
549 if (!objios->sync)
550 objios->ios->done = _write_done;
551
552 dprintk("%s: offset=0x%llx length=0x%x\n", __func__,
553 wdata->args.offset, wdata->args.count);
554 ret = ore_write(objios->ios);
555 if (unlikely(ret)) {
556 objio_free_result(&objios->oir);
557 return ret;
558 }
559
560 if (objios->sync)
561 _write_done(objios->ios, objios);
562
563 return 0;
564}
565
566static bool objio_pg_test(struct nfs_pageio_descriptor *pgio,
567 struct nfs_page *prev, struct nfs_page *req)
568{
569 if (!pnfs_generic_pg_test(pgio, prev, req))
570 return false;
571
572 return pgio->pg_count + req->wb_bytes <=
573 OBJIO_LSEG(pgio->pg_lseg)->layout.max_io_length;
574}
575
576static const struct nfs_pageio_ops objio_pg_read_ops = {
577 .pg_init = pnfs_generic_pg_init_read,
578 .pg_test = objio_pg_test,
579 .pg_doio = pnfs_generic_pg_readpages,
580};
581
582static const struct nfs_pageio_ops objio_pg_write_ops = {
583 .pg_init = pnfs_generic_pg_init_write,
584 .pg_test = objio_pg_test,
585 .pg_doio = pnfs_generic_pg_writepages,
586};
587
588static struct pnfs_layoutdriver_type objlayout_type = {
589 .id = LAYOUT_OSD2_OBJECTS,
590 .name = "LAYOUT_OSD2_OBJECTS",
591 .flags = PNFS_LAYOUTRET_ON_SETATTR |
592 PNFS_LAYOUTRET_ON_ERROR,
593
594 .alloc_layout_hdr = objlayout_alloc_layout_hdr,
595 .free_layout_hdr = objlayout_free_layout_hdr,
596
597 .alloc_lseg = objlayout_alloc_lseg,
598 .free_lseg = objlayout_free_lseg,
599
600 .read_pagelist = objlayout_read_pagelist,
601 .write_pagelist = objlayout_write_pagelist,
602 .pg_read_ops = &objio_pg_read_ops,
603 .pg_write_ops = &objio_pg_write_ops,
604
605 .free_deviceid_node = objio_free_deviceid_node,
606
607 .encode_layoutcommit = objlayout_encode_layoutcommit,
608 .encode_layoutreturn = objlayout_encode_layoutreturn,
609};
610
611MODULE_DESCRIPTION("pNFS Layout Driver for OSD2 objects");
612MODULE_AUTHOR("Benny Halevy <bhalevy@panasas.com>");
613MODULE_LICENSE("GPL");
614
615static int __init
616objlayout_init(void)
617{
618 int ret = pnfs_register_layoutdriver(&objlayout_type);
619
620 if (ret)
621 printk(KERN_INFO
622 "NFS: %s: Registering OSD pNFS Layout Driver failed: error=%d\n",
623 __func__, ret);
624 else
625 printk(KERN_INFO "NFS: %s: Registered OSD pNFS Layout Driver\n",
626 __func__);
627 return ret;
628}
629
630static void __exit
631objlayout_exit(void)
632{
633 pnfs_unregister_layoutdriver(&objlayout_type);
634 printk(KERN_INFO "NFS: %s: Unregistered OSD pNFS Layout Driver\n",
635 __func__);
636}
637
638MODULE_ALIAS("nfs-layouttype4-2");
639
640module_init(objlayout_init);
641module_exit(objlayout_exit);