osd: Kconfig remove wrong FIXME
[deliverable/linux.git] / fs / exofs / ore.c
CommitLineData
b14f8ab2
BH
1/*
2 * Copyright (C) 2005, 2006
27d2e149 3 * Avishay Traeger (avishay@gmail.com)
b14f8ab2
BH
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
6 *
7 * This file is part of exofs.
8 *
9 * exofs is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation. Since it is based on ext2, and the only
12 * valid version of GPL for the Linux kernel is version 2, the only valid
13 * version of GPL for exofs is version 2.
14 *
15 * exofs is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with exofs; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24
5a0e3ad6 25#include <linux/slab.h>
5d952b83 26#include <asm/div64.h>
b14f8ab2 27
8ff660ab 28#include <scsi/osd_ore.h>
b14f8ab2 29
8ff660ab 30#define ORE_ERR(fmt, a...) printk(KERN_ERR "ore: " fmt, ##a)
34ce4e7c 31
8ff660ab
BH
32#ifdef CONFIG_EXOFS_DEBUG
33#define ORE_DBGMSG(fmt, a...) \
34 printk(KERN_NOTICE "ore @%s:%d: " fmt, __func__, __LINE__, ##a)
35#else
36#define ORE_DBGMSG(fmt, a...) \
37 do { if (0) printk(fmt, ##a); } while (0)
38#endif
39
40/* u64 has problems with printk this will cast it to unsigned long long */
41#define _LLU(x) (unsigned long long)(x)
42
43#define ORE_DBGMSG2(M...) do {} while (0)
44/* #define ORE_DBGMSG2 ORE_DBGMSG */
45
cf283ade
BH
46MODULE_AUTHOR("Boaz Harrosh <bharrosh@panasas.com>");
47MODULE_DESCRIPTION("Objects Raid Engine ore.ko");
48MODULE_LICENSE("GPL");
49
8ff660ab 50static u8 *_ios_cred(struct ore_io_state *ios, unsigned index)
9e9db456
BH
51{
52 return ios->comps->comps[index & ios->comps->single_comp].cred;
53}
54
8ff660ab 55static struct osd_obj_id *_ios_obj(struct ore_io_state *ios, unsigned index)
9e9db456
BH
56{
57 return &ios->comps->comps[index & ios->comps->single_comp].obj;
58}
59
8ff660ab 60static struct osd_dev *_ios_od(struct ore_io_state *ios, unsigned index)
9e9db456
BH
61{
62 return ios->comps->ods[index];
63}
64
8ff660ab
BH
65int ore_get_rw_state(struct ore_layout *layout, struct ore_components *comps,
66 bool is_reading, u64 offset, u64 length,
67 struct ore_io_state **pios)
b14f8ab2 68{
8ff660ab 69 struct ore_io_state *ios;
06886a5a
BH
70
71 /*TODO: Maybe use kmem_cach per sbi of size
45d3abcb 72 * exofs_io_state_size(layout->s_numdevs)
06886a5a 73 */
8ff660ab 74 ios = kzalloc(ore_io_state_size(comps->numdevs), GFP_KERNEL);
06886a5a 75 if (unlikely(!ios)) {
8ff660ab
BH
76 ORE_DBGMSG("Failed kzalloc bytes=%d\n",
77 ore_io_state_size(comps->numdevs));
06886a5a
BH
78 *pios = NULL;
79 return -ENOMEM;
80 }
81
45d3abcb 82 ios->layout = layout;
9e9db456 83 ios->comps = comps;
e1042ba0
BH
84 ios->offset = offset;
85 ios->length = length;
86 ios->reading = is_reading;
87
06886a5a
BH
88 *pios = ios;
89 return 0;
b14f8ab2 90}
cf283ade 91EXPORT_SYMBOL(ore_get_rw_state);
b14f8ab2 92
8ff660ab
BH
93int ore_get_io_state(struct ore_layout *layout, struct ore_components *comps,
94 struct ore_io_state **ios)
e1042ba0 95{
8ff660ab 96 return ore_get_rw_state(layout, comps, true, 0, 0, ios);
e1042ba0 97}
cf283ade 98EXPORT_SYMBOL(ore_get_io_state);
e1042ba0 99
8ff660ab 100void ore_put_io_state(struct ore_io_state *ios)
b14f8ab2 101{
06886a5a
BH
102 if (ios) {
103 unsigned i;
b14f8ab2 104
06886a5a 105 for (i = 0; i < ios->numdevs; i++) {
8ff660ab 106 struct ore_per_dev_state *per_dev = &ios->per_dev[i];
06886a5a
BH
107
108 if (per_dev->or)
109 osd_end_request(per_dev->or);
110 if (per_dev->bio)
111 bio_put(per_dev->bio);
112 }
113
114 kfree(ios);
b14f8ab2 115 }
06886a5a 116}
cf283ade 117EXPORT_SYMBOL(ore_put_io_state);
b14f8ab2 118
8ff660ab 119static void _sync_done(struct ore_io_state *ios, void *p)
06886a5a
BH
120{
121 struct completion *waiting = p;
b14f8ab2 122
06886a5a
BH
123 complete(waiting);
124}
125
126static void _last_io(struct kref *kref)
127{
8ff660ab
BH
128 struct ore_io_state *ios = container_of(
129 kref, struct ore_io_state, kref);
06886a5a
BH
130
131 ios->done(ios, ios->private);
132}
133
134static void _done_io(struct osd_request *or, void *p)
135{
8ff660ab 136 struct ore_io_state *ios = p;
06886a5a
BH
137
138 kref_put(&ios->kref, _last_io);
139}
140
8ff660ab 141static int ore_io_execute(struct ore_io_state *ios)
06886a5a
BH
142{
143 DECLARE_COMPLETION_ONSTACK(wait);
144 bool sync = (ios->done == NULL);
145 int i, ret;
146
147 if (sync) {
148 ios->done = _sync_done;
149 ios->private = &wait;
150 }
151
152 for (i = 0; i < ios->numdevs; i++) {
153 struct osd_request *or = ios->per_dev[i].or;
154 if (unlikely(!or))
155 continue;
156
9e9db456 157 ret = osd_finalize_request(or, 0, _ios_cred(ios, i), NULL);
06886a5a 158 if (unlikely(ret)) {
8ff660ab 159 ORE_DBGMSG("Failed to osd_finalize_request() => %d\n",
06886a5a
BH
160 ret);
161 return ret;
162 }
163 }
164
165 kref_init(&ios->kref);
166
167 for (i = 0; i < ios->numdevs; i++) {
168 struct osd_request *or = ios->per_dev[i].or;
169 if (unlikely(!or))
170 continue;
171
172 kref_get(&ios->kref);
173 osd_execute_request_async(or, _done_io, ios);
174 }
175
176 kref_put(&ios->kref, _last_io);
177 ret = 0;
178
179 if (sync) {
180 wait_for_completion(&wait);
8ff660ab 181 ret = ore_check_io(ios, NULL);
06886a5a 182 }
b14f8ab2
BH
183 return ret;
184}
185
22ddc556
BH
186static void _clear_bio(struct bio *bio)
187{
188 struct bio_vec *bv;
189 unsigned i;
190
191 __bio_for_each_segment(bv, bio, i, 0) {
192 unsigned this_count = bv->bv_len;
193
194 if (likely(PAGE_SIZE == this_count))
195 clear_highpage(bv->bv_page);
196 else
197 zero_user(bv->bv_page, bv->bv_offset, this_count);
198 }
199}
200
8ff660ab 201int ore_check_io(struct ore_io_state *ios, u64 *resid)
b14f8ab2 202{
06886a5a
BH
203 enum osd_err_priority acumulated_osd_err = 0;
204 int acumulated_lin_err = 0;
205 int i;
b14f8ab2 206
06886a5a
BH
207 for (i = 0; i < ios->numdevs; i++) {
208 struct osd_sense_info osi;
22ddc556
BH
209 struct osd_request *or = ios->per_dev[i].or;
210 int ret;
211
212 if (unlikely(!or))
213 continue;
06886a5a 214
22ddc556 215 ret = osd_req_decode_sense(or, &osi);
06886a5a
BH
216 if (likely(!ret))
217 continue;
218
22ddc556
BH
219 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
220 /* start read offset passed endof file */
221 _clear_bio(ios->per_dev[i].bio);
8ff660ab 222 ORE_DBGMSG("start read offset passed end of file "
22ddc556 223 "offset=0x%llx, length=0x%llx\n",
5d952b83
BH
224 _LLU(ios->per_dev[i].offset),
225 _LLU(ios->per_dev[i].length));
22ddc556
BH
226
227 continue; /* we recovered */
06886a5a
BH
228 }
229
230 if (osi.osd_err_pri >= acumulated_osd_err) {
231 acumulated_osd_err = osi.osd_err_pri;
232 acumulated_lin_err = ret;
233 }
234 }
235
236 /* TODO: raid specific residual calculations */
237 if (resid) {
238 if (likely(!acumulated_lin_err))
239 *resid = 0;
240 else
241 *resid = ios->length;
242 }
243
244 return acumulated_lin_err;
245}
cf283ade 246EXPORT_SYMBOL(ore_check_io);
06886a5a 247
b367e78b
BH
248/*
249 * L - logical offset into the file
250 *
50a76fd3 251 * U - The number of bytes in a stripe within a group
b367e78b
BH
252 *
253 * U = stripe_unit * group_width
254 *
50a76fd3
BH
255 * T - The number of bytes striped within a group of component objects
256 * (before advancing to the next group)
b367e78b 257 *
50a76fd3
BH
258 * T = stripe_unit * group_width * group_depth
259 *
260 * S - The number of bytes striped across all component objects
261 * before the pattern repeats
262 *
263 * S = stripe_unit * group_width * group_depth * group_count
264 *
265 * M - The "major" (i.e., across all components) stripe number
266 *
267 * M = L / S
268 *
269 * G - Counts the groups from the beginning of the major stripe
270 *
271 * G = (L - (M * S)) / T [or (L % S) / T]
272 *
273 * H - The byte offset within the group
274 *
275 * H = (L - (M * S)) % T [or (L % S) % T]
276 *
277 * N - The "minor" (i.e., across the group) stripe number
278 *
279 * N = H / U
b367e78b
BH
280 *
281 * C - The component index coresponding to L
282 *
50a76fd3
BH
283 * C = (H - (N * U)) / stripe_unit + G * group_width
284 * [or (L % U) / stripe_unit + G * group_width]
b367e78b
BH
285 *
286 * O - The component offset coresponding to L
287 *
50a76fd3 288 * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
b367e78b 289 */
b367e78b
BH
290struct _striping_info {
291 u64 obj_offset;
50a76fd3 292 u64 group_length;
16f75bb3 293 u64 M; /* for truncate */
b367e78b
BH
294 unsigned dev;
295 unsigned unit_off;
296};
297
8ff660ab 298static void _calc_stripe_info(struct ore_layout *layout, u64 file_offset,
b367e78b 299 struct _striping_info *si)
5d952b83 300{
16f75bb3
BH
301 u32 stripe_unit = layout->stripe_unit;
302 u32 group_width = layout->group_width;
303 u64 group_depth = layout->group_depth;
50a76fd3 304
b367e78b 305 u32 U = stripe_unit * group_width;
50a76fd3 306 u64 T = U * group_depth;
16f75bb3 307 u64 S = T * layout->group_count;
50a76fd3
BH
308 u64 M = div64_u64(file_offset, S);
309
310 /*
311 G = (L - (M * S)) / T
312 H = (L - (M * S)) % T
313 */
314 u64 LmodS = file_offset - M * S;
315 u32 G = div64_u64(LmodS, T);
316 u64 H = LmodS - G * T;
317
318 u32 N = div_u64(H, U);
319
320 /* "H - (N * U)" is just "H % U" so it's bound to u32 */
321 si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
16f75bb3 322 si->dev *= layout->mirrors_p1;
b367e78b 323
50a76fd3 324 div_u64_rem(file_offset, stripe_unit, &si->unit_off);
5d952b83 325
50a76fd3
BH
326 si->obj_offset = si->unit_off + (N * stripe_unit) +
327 (M * group_depth * stripe_unit);
328
329 si->group_length = T - H;
16f75bb3 330 si->M = M;
5d952b83
BH
331}
332
8ff660ab
BH
333static int _add_stripe_unit(struct ore_io_state *ios, unsigned *cur_pg,
334 unsigned pgbase, struct ore_per_dev_state *per_dev,
86093aaf 335 int cur_len)
5d952b83 336{
86093aaf 337 unsigned pg = *cur_pg;
5d952b83 338 struct request_queue *q =
9e9db456 339 osd_request_queue(_ios_od(ios, per_dev->dev));
5d952b83
BH
340
341 per_dev->length += cur_len;
342
343 if (per_dev->bio == NULL) {
344 unsigned pages_in_stripe = ios->layout->group_width *
345 (ios->layout->stripe_unit / PAGE_SIZE);
86093aaf 346 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
5d952b83
BH
347 ios->layout->group_width;
348
349 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
350 if (unlikely(!per_dev->bio)) {
8ff660ab 351 ORE_DBGMSG("Failed to allocate BIO size=%u\n",
5d952b83
BH
352 bio_size);
353 return -ENOMEM;
354 }
355 }
356
357 while (cur_len > 0) {
86093aaf
BH
358 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
359 unsigned added_len;
5d952b83 360
86093aaf
BH
361 BUG_ON(ios->nr_pages <= pg);
362 cur_len -= pglen;
5d952b83 363
86093aaf
BH
364 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
365 pglen, pgbase);
366 if (unlikely(pglen != added_len))
5d952b83 367 return -ENOMEM;
86093aaf
BH
368 pgbase = 0;
369 ++pg;
5d952b83
BH
370 }
371 BUG_ON(cur_len);
372
86093aaf 373 *cur_pg = pg;
5d952b83
BH
374 return 0;
375}
376
8ff660ab 377static int _prepare_one_group(struct ore_io_state *ios, u64 length,
6e31609b 378 struct _striping_info *si)
5d952b83 379{
5d952b83 380 unsigned stripe_unit = ios->layout->stripe_unit;
b367e78b 381 unsigned mirrors_p1 = ios->layout->mirrors_p1;
50a76fd3 382 unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
b367e78b 383 unsigned dev = si->dev;
50a76fd3 384 unsigned first_dev = dev - (dev % devs_in_group);
50a76fd3
BH
385 unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
386 unsigned cur_pg = ios->pages_consumed;
86093aaf 387 int ret = 0;
5d952b83 388
5d952b83 389 while (length) {
8ff660ab 390 struct ore_per_dev_state *per_dev = &ios->per_dev[dev];
b367e78b 391 unsigned cur_len, page_off = 0;
5d952b83
BH
392
393 if (!per_dev->length) {
b367e78b
BH
394 per_dev->dev = dev;
395 if (dev < si->dev) {
396 per_dev->offset = si->obj_offset + stripe_unit -
397 si->unit_off;
398 cur_len = stripe_unit;
399 } else if (dev == si->dev) {
400 per_dev->offset = si->obj_offset;
401 cur_len = stripe_unit - si->unit_off;
402 page_off = si->unit_off & ~PAGE_MASK;
403 BUG_ON(page_off && (page_off != ios->pgbase));
404 } else { /* dev > si->dev */
405 per_dev->offset = si->obj_offset - si->unit_off;
406 cur_len = stripe_unit;
407 }
5d952b83 408
6e31609b
BH
409 if (max_comp < dev)
410 max_comp = dev;
5d952b83 411 } else {
b367e78b 412 cur_len = stripe_unit;
5d952b83 413 }
b367e78b
BH
414 if (cur_len >= length)
415 cur_len = length;
5d952b83 416
86093aaf
BH
417 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
418 cur_len);
5d952b83
BH
419 if (unlikely(ret))
420 goto out;
421
6e31609b
BH
422 dev += mirrors_p1;
423 dev = (dev % devs_in_group) + first_dev;
5d952b83
BH
424
425 length -= cur_len;
426 }
427out:
50a76fd3
BH
428 ios->numdevs = max_comp + mirrors_p1;
429 ios->pages_consumed = cur_pg;
5d952b83
BH
430 return ret;
431}
432
8ff660ab 433static int _prepare_for_striping(struct ore_io_state *ios)
b367e78b 434{
50a76fd3 435 u64 length = ios->length;
5002dd18 436 u64 offset = ios->offset;
b367e78b 437 struct _striping_info si;
50a76fd3 438 int ret = 0;
b367e78b 439
b367e78b
BH
440 if (!ios->pages) {
441 if (ios->kern_buff) {
8ff660ab 442 struct ore_per_dev_state *per_dev = &ios->per_dev[0];
b367e78b 443
16f75bb3 444 _calc_stripe_info(ios->layout, ios->offset, &si);
b367e78b
BH
445 per_dev->offset = si.obj_offset;
446 per_dev->dev = si.dev;
447
448 /* no cross device without page array */
449 BUG_ON((ios->layout->group_width > 1) &&
450 (si.unit_off + ios->length >
451 ios->layout->stripe_unit));
452 }
453 ios->numdevs = ios->layout->mirrors_p1;
454 return 0;
455 }
456
50a76fd3 457 while (length) {
16f75bb3 458 _calc_stripe_info(ios->layout, offset, &si);
5002dd18 459
50a76fd3
BH
460 if (length < si.group_length)
461 si.group_length = length;
462
6e31609b 463 ret = _prepare_one_group(ios, si.group_length, &si);
50a76fd3
BH
464 if (unlikely(ret))
465 goto out;
466
5002dd18 467 offset += si.group_length;
50a76fd3 468 length -= si.group_length;
50a76fd3
BH
469 }
470
471out:
472 return ret;
b367e78b
BH
473}
474
8ff660ab 475int ore_create(struct ore_io_state *ios)
06886a5a
BH
476{
477 int i, ret;
478
9e9db456 479 for (i = 0; i < ios->comps->numdevs; i++) {
06886a5a
BH
480 struct osd_request *or;
481
9e9db456 482 or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
06886a5a 483 if (unlikely(!or)) {
8ff660ab 484 ORE_ERR("%s: osd_start_request failed\n", __func__);
06886a5a
BH
485 ret = -ENOMEM;
486 goto out;
487 }
488 ios->per_dev[i].or = or;
489 ios->numdevs++;
490
9e9db456 491 osd_req_create_object(or, _ios_obj(ios, i));
06886a5a 492 }
8ff660ab 493 ret = ore_io_execute(ios);
06886a5a
BH
494
495out:
496 return ret;
497}
cf283ade 498EXPORT_SYMBOL(ore_create);
06886a5a 499
8ff660ab 500int ore_remove(struct ore_io_state *ios)
06886a5a
BH
501{
502 int i, ret;
503
9e9db456 504 for (i = 0; i < ios->comps->numdevs; i++) {
06886a5a
BH
505 struct osd_request *or;
506
9e9db456 507 or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
06886a5a 508 if (unlikely(!or)) {
8ff660ab 509 ORE_ERR("%s: osd_start_request failed\n", __func__);
06886a5a
BH
510 ret = -ENOMEM;
511 goto out;
512 }
513 ios->per_dev[i].or = or;
514 ios->numdevs++;
515
9e9db456 516 osd_req_remove_object(or, _ios_obj(ios, i));
06886a5a 517 }
8ff660ab 518 ret = ore_io_execute(ios);
06886a5a
BH
519
520out:
521 return ret;
522}
cf283ade 523EXPORT_SYMBOL(ore_remove);
06886a5a 524
8ff660ab 525static int _write_mirror(struct ore_io_state *ios, int cur_comp)
06886a5a 526{
8ff660ab 527 struct ore_per_dev_state *master_dev = &ios->per_dev[cur_comp];
5d952b83
BH
528 unsigned dev = ios->per_dev[cur_comp].dev;
529 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
530 int ret = 0;
06886a5a 531
50a76fd3
BH
532 if (ios->pages && !master_dev->length)
533 return 0; /* Just an empty slot */
534
5d952b83 535 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
8ff660ab 536 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
06886a5a
BH
537 struct osd_request *or;
538
9e9db456 539 or = osd_start_request(_ios_od(ios, dev), GFP_KERNEL);
06886a5a 540 if (unlikely(!or)) {
8ff660ab 541 ORE_ERR("%s: osd_start_request failed\n", __func__);
06886a5a
BH
542 ret = -ENOMEM;
543 goto out;
544 }
5d952b83
BH
545 per_dev->or = or;
546 per_dev->offset = master_dev->offset;
06886a5a 547
86093aaf 548 if (ios->pages) {
06886a5a
BH
549 struct bio *bio;
550
5d952b83 551 if (per_dev != master_dev) {
04dc1e88 552 bio = bio_kmalloc(GFP_KERNEL,
5d952b83 553 master_dev->bio->bi_max_vecs);
04dc1e88 554 if (unlikely(!bio)) {
8ff660ab 555 ORE_DBGMSG(
426d3107 556 "Failed to allocate BIO size=%u\n",
5d952b83 557 master_dev->bio->bi_max_vecs);
04dc1e88
BH
558 ret = -ENOMEM;
559 goto out;
560 }
561
5d952b83 562 __bio_clone(bio, master_dev->bio);
04dc1e88
BH
563 bio->bi_bdev = NULL;
564 bio->bi_next = NULL;
5d952b83
BH
565 per_dev->length = master_dev->length;
566 per_dev->bio = bio;
567 per_dev->dev = dev;
04dc1e88 568 } else {
5d952b83
BH
569 bio = master_dev->bio;
570 /* FIXME: bio_set_dir() */
7b6d91da 571 bio->bi_rw |= REQ_WRITE;
04dc1e88 572 }
06886a5a 573
9e9db456
BH
574 osd_req_write(or, _ios_obj(ios, dev), per_dev->offset,
575 bio, per_dev->length);
8ff660ab 576 ORE_DBGMSG("write(0x%llx) offset=0x%llx "
34ce4e7c 577 "length=0x%llx dev=%d\n",
9e9db456
BH
578 _LLU(_ios_obj(ios, dev)->id),
579 _LLU(per_dev->offset),
5d952b83 580 _LLU(per_dev->length), dev);
06886a5a 581 } else if (ios->kern_buff) {
9e9db456
BH
582 ret = osd_req_write_kern(or, _ios_obj(ios, dev),
583 per_dev->offset,
584 ios->kern_buff, ios->length);
5d952b83
BH
585 if (unlikely(ret))
586 goto out;
8ff660ab 587 ORE_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
34ce4e7c 588 "length=0x%llx dev=%d\n",
9e9db456
BH
589 _LLU(_ios_obj(ios, dev)->id),
590 _LLU(per_dev->offset),
5d952b83 591 _LLU(ios->length), dev);
06886a5a 592 } else {
9e9db456 593 osd_req_set_attributes(or, _ios_obj(ios, dev));
8ff660ab 594 ORE_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
9e9db456
BH
595 _LLU(_ios_obj(ios, dev)->id),
596 ios->out_attr_len, dev);
06886a5a
BH
597 }
598
599 if (ios->out_attr)
600 osd_req_add_set_attr_list(or, ios->out_attr,
601 ios->out_attr_len);
602
603 if (ios->in_attr)
604 osd_req_add_get_attr_list(or, ios->in_attr,
605 ios->in_attr_len);
b14f8ab2 606 }
06886a5a
BH
607
608out:
609 return ret;
610}
611
8ff660ab 612int ore_write(struct ore_io_state *ios)
5d952b83
BH
613{
614 int i;
615 int ret;
616
617 ret = _prepare_for_striping(ios);
618 if (unlikely(ret))
619 return ret;
620
621 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
8ff660ab 622 ret = _write_mirror(ios, i);
5d952b83
BH
623 if (unlikely(ret))
624 return ret;
625 }
626
8ff660ab 627 ret = ore_io_execute(ios);
5d952b83
BH
628 return ret;
629}
cf283ade 630EXPORT_SYMBOL(ore_write);
5d952b83 631
8ff660ab 632static int _read_mirror(struct ore_io_state *ios, unsigned cur_comp)
06886a5a 633{
46f4d973 634 struct osd_request *or;
8ff660ab 635 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
9e9db456
BH
636 struct osd_obj_id *obj = _ios_obj(ios, cur_comp);
637 unsigned first_dev = (unsigned)obj->id;
06886a5a 638
50a76fd3
BH
639 if (ios->pages && !per_dev->length)
640 return 0; /* Just an empty slot */
641
5d952b83 642 first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
9e9db456 643 or = osd_start_request(_ios_od(ios, first_dev), GFP_KERNEL);
46f4d973 644 if (unlikely(!or)) {
8ff660ab 645 ORE_ERR("%s: osd_start_request failed\n", __func__);
46f4d973
BH
646 return -ENOMEM;
647 }
648 per_dev->or = or;
46f4d973 649
86093aaf 650 if (ios->pages) {
9e9db456 651 osd_req_read(or, obj, per_dev->offset,
5d952b83 652 per_dev->bio, per_dev->length);
8ff660ab 653 ORE_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
9e9db456 654 " dev=%d\n", _LLU(obj->id),
5d952b83 655 _LLU(per_dev->offset), _LLU(per_dev->length),
46f4d973
BH
656 first_dev);
657 } else if (ios->kern_buff) {
9e9db456 658 int ret = osd_req_read_kern(or, obj, per_dev->offset,
46f4d973 659 ios->kern_buff, ios->length);
8ff660ab 660 ORE_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
46f4d973 661 "length=0x%llx dev=%d ret=>%d\n",
9e9db456 662 _LLU(obj->id), _LLU(per_dev->offset),
46f4d973
BH
663 _LLU(ios->length), first_dev, ret);
664 if (unlikely(ret))
665 return ret;
666 } else {
9e9db456 667 osd_req_get_attributes(or, obj);
8ff660ab 668 ORE_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
9e9db456
BH
669 _LLU(obj->id),
670 ios->in_attr_len, first_dev);
46f4d973 671 }
46f4d973
BH
672 if (ios->out_attr)
673 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
b14f8ab2 674
46f4d973
BH
675 if (ios->in_attr)
676 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
b14f8ab2 677
5d952b83
BH
678 return 0;
679}
680
8ff660ab 681int ore_read(struct ore_io_state *ios)
5d952b83
BH
682{
683 int i;
684 int ret;
685
686 ret = _prepare_for_striping(ios);
687 if (unlikely(ret))
688 return ret;
689
690 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
8ff660ab 691 ret = _read_mirror(ios, i);
5d952b83
BH
692 if (unlikely(ret))
693 return ret;
694 }
695
8ff660ab 696 ret = ore_io_execute(ios);
5d952b83 697 return ret;
b14f8ab2 698}
cf283ade 699EXPORT_SYMBOL(ore_read);
b14f8ab2 700
8ff660ab 701int extract_attr_from_ios(struct ore_io_state *ios, struct osd_attr *attr)
b14f8ab2
BH
702{
703 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
704 void *iter = NULL;
705 int nelem;
706
707 do {
708 nelem = 1;
06886a5a
BH
709 osd_req_decode_get_attr_list(ios->per_dev[0].or,
710 &cur_attr, &nelem, &iter);
b14f8ab2
BH
711 if ((cur_attr.attr_page == attr->attr_page) &&
712 (cur_attr.attr_id == attr->attr_id)) {
713 attr->len = cur_attr.len;
714 attr->val_ptr = cur_attr.val_ptr;
715 return 0;
716 }
717 } while (iter);
718
719 return -EIO;
720}
cf283ade 721EXPORT_SYMBOL(extract_attr_from_ios);
06886a5a 722
8ff660ab 723static int _truncate_mirrors(struct ore_io_state *ios, unsigned cur_comp,
5d952b83
BH
724 struct osd_attr *attr)
725{
726 int last_comp = cur_comp + ios->layout->mirrors_p1;
727
728 for (; cur_comp < last_comp; ++cur_comp) {
8ff660ab 729 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
5d952b83
BH
730 struct osd_request *or;
731
9e9db456 732 or = osd_start_request(_ios_od(ios, cur_comp), GFP_KERNEL);
5d952b83 733 if (unlikely(!or)) {
8ff660ab 734 ORE_ERR("%s: osd_start_request failed\n", __func__);
5d952b83
BH
735 return -ENOMEM;
736 }
737 per_dev->or = or;
738
9e9db456 739 osd_req_set_attributes(or, _ios_obj(ios, cur_comp));
5d952b83
BH
740 osd_req_add_set_attr_list(or, attr, 1);
741 }
742
743 return 0;
744}
745
16f75bb3
BH
746struct _trunc_info {
747 struct _striping_info si;
748 u64 prev_group_obj_off;
749 u64 next_group_obj_off;
750
751 unsigned first_group_dev;
752 unsigned nex_group_dev;
753 unsigned max_devs;
754};
755
8ff660ab 756void _calc_trunk_info(struct ore_layout *layout, u64 file_offset,
16f75bb3
BH
757 struct _trunc_info *ti)
758{
759 unsigned stripe_unit = layout->stripe_unit;
760
761 _calc_stripe_info(layout, file_offset, &ti->si);
762
763 ti->prev_group_obj_off = ti->si.M * stripe_unit;
764 ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
765
766 ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
767 ti->nex_group_dev = ti->first_group_dev + layout->group_width;
768 ti->max_devs = layout->group_width * layout->group_count;
769}
770
8ff660ab 771int ore_truncate(struct ore_layout *layout, struct ore_components *comps,
9e9db456 772 u64 size)
06886a5a 773{
8ff660ab 774 struct ore_io_state *ios;
5d952b83
BH
775 struct exofs_trunc_attr {
776 struct osd_attr attr;
777 __be64 newsize;
778 } *size_attrs;
16f75bb3 779 struct _trunc_info ti;
06886a5a
BH
780 int i, ret;
781
8ff660ab 782 ret = ore_get_io_state(layout, comps, &ios);
5d952b83
BH
783 if (unlikely(ret))
784 return ret;
785
16f75bb3
BH
786 _calc_trunk_info(ios->layout, size, &ti);
787
788 size_attrs = kcalloc(ti.max_devs, sizeof(*size_attrs),
5d952b83
BH
789 GFP_KERNEL);
790 if (unlikely(!size_attrs)) {
791 ret = -ENOMEM;
792 goto out;
793 }
06886a5a 794
9e9db456 795 ios->numdevs = ios->comps->numdevs;
06886a5a 796
16f75bb3 797 for (i = 0; i < ti.max_devs; ++i) {
5d952b83
BH
798 struct exofs_trunc_attr *size_attr = &size_attrs[i];
799 u64 obj_size;
06886a5a 800
16f75bb3
BH
801 if (i < ti.first_group_dev)
802 obj_size = ti.prev_group_obj_off;
803 else if (i >= ti.nex_group_dev)
804 obj_size = ti.next_group_obj_off;
805 else if (i < ti.si.dev) /* dev within this group */
806 obj_size = ti.si.obj_offset +
807 ios->layout->stripe_unit - ti.si.unit_off;
808 else if (i == ti.si.dev)
809 obj_size = ti.si.obj_offset;
810 else /* i > ti.dev */
811 obj_size = ti.si.obj_offset - ti.si.unit_off;
06886a5a 812
5d952b83
BH
813 size_attr->newsize = cpu_to_be64(obj_size);
814 size_attr->attr = g_attr_logical_length;
815 size_attr->attr.val_ptr = &size_attr->newsize;
816
8ff660ab 817 ORE_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
9e9db456 818 _LLU(comps->comps->obj.id), _LLU(obj_size), i);
5d952b83
BH
819 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
820 &size_attr->attr);
821 if (unlikely(ret))
822 goto out;
06886a5a 823 }
8ff660ab 824 ret = ore_io_execute(ios);
06886a5a
BH
825
826out:
5d952b83 827 kfree(size_attrs);
8ff660ab 828 ore_put_io_state(ios);
06886a5a
BH
829 return ret;
830}
cf283ade 831EXPORT_SYMBOL(ore_truncate);
85e44df4
BH
832
833const struct osd_attr g_attr_logical_length = ATTR_DEF(
834 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
cf283ade 835EXPORT_SYMBOL(g_attr_logical_length);
This page took 0.203222 seconds and 5 git commands to generate.