1 /*
2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <ooo@electrozaur.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
25 #include <linux/slab.h>
26 #include <linux/module.h>
27 #include <asm/div64.h>
28 #include <linux/lcm.h>
29
30 #include "ore_raid.h"
31
32 MODULE_AUTHOR("Boaz Harrosh <ooo@electrozaur.com>");
33 MODULE_DESCRIPTION("Objects Raid Engine ore.ko");
34 MODULE_LICENSE("GPL");
35
36 /* ore_verify_layout does a couple of things:
37 * 1. Given a minimum number of needed parameters fixes up the rest of the
38 * members to be operatonals for the ore. The needed parameters are those
39 * that are defined by the pnfs-objects layout STD.
40 * 2. Check to see if the current ore code actually supports these parameters
41 * for example stripe_unit must be a multple of the system PAGE_SIZE,
42 * and etc...
43 * 3. Cache some havily used calculations that will be needed by users.
44 */
45
46 enum { BIO_MAX_PAGES_KMALLOC =
47 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),};
48
ore_verify_layout(unsigned total_comps,struct ore_layout * layout)49 int ore_verify_layout(unsigned total_comps, struct ore_layout *layout)
50 {
51 u64 stripe_length;
52
53 switch (layout->raid_algorithm) {
54 case PNFS_OSD_RAID_0:
55 layout->parity = 0;
56 break;
57 case PNFS_OSD_RAID_5:
58 layout->parity = 1;
59 break;
60 case PNFS_OSD_RAID_PQ:
61 layout->parity = 2;
62 break;
63 case PNFS_OSD_RAID_4:
64 default:
65 ORE_ERR("Only RAID_0/5/6 for now received-enum=%d\n",
66 layout->raid_algorithm);
67 return -EINVAL;
68 }
69 if (0 != (layout->stripe_unit & ~PAGE_MASK)) {
70 ORE_ERR("Stripe Unit(0x%llx)"
71 " must be Multples of PAGE_SIZE(0x%lx)\n",
72 _LLU(layout->stripe_unit), PAGE_SIZE);
73 return -EINVAL;
74 }
75 if (layout->group_width) {
76 if (!layout->group_depth) {
77 ORE_ERR("group_depth == 0 && group_width != 0\n");
78 return -EINVAL;
79 }
80 if (total_comps < (layout->group_width * layout->mirrors_p1)) {
81 ORE_ERR("Data Map wrong, "
82 "numdevs=%d < group_width=%d * mirrors=%d\n",
83 total_comps, layout->group_width,
84 layout->mirrors_p1);
85 return -EINVAL;
86 }
87 layout->group_count = total_comps / layout->mirrors_p1 /
88 layout->group_width;
89 } else {
90 if (layout->group_depth) {
91 printk(KERN_NOTICE "Warning: group_depth ignored "
92 "group_width == 0 && group_depth == %lld\n",
93 _LLU(layout->group_depth));
94 }
95 layout->group_width = total_comps / layout->mirrors_p1;
96 layout->group_depth = -1;
97 layout->group_count = 1;
98 }
99
100 stripe_length = (u64)layout->group_width * layout->stripe_unit;
101 if (stripe_length >= (1ULL << 32)) {
102 ORE_ERR("Stripe_length(0x%llx) >= 32bit is not supported\n",
103 _LLU(stripe_length));
104 return -EINVAL;
105 }
106
107 layout->max_io_length =
108 (BIO_MAX_PAGES_KMALLOC * PAGE_SIZE - layout->stripe_unit) *
109 (layout->group_width - layout->parity);
110 if (layout->parity) {
111 unsigned stripe_length =
112 (layout->group_width - layout->parity) *
113 layout->stripe_unit;
114
115 layout->max_io_length /= stripe_length;
116 layout->max_io_length *= stripe_length;
117 }
118 ORE_DBGMSG("max_io_length=0x%lx\n", layout->max_io_length);
119
120 return 0;
121 }
122 EXPORT_SYMBOL(ore_verify_layout);
123
_ios_cred(struct ore_io_state * ios,unsigned index)124 static u8 *_ios_cred(struct ore_io_state *ios, unsigned index)
125 {
126 return ios->oc->comps[index & ios->oc->single_comp].cred;
127 }
128
_ios_obj(struct ore_io_state * ios,unsigned index)129 static struct osd_obj_id *_ios_obj(struct ore_io_state *ios, unsigned index)
130 {
131 return &ios->oc->comps[index & ios->oc->single_comp].obj;
132 }
133
_ios_od(struct ore_io_state * ios,unsigned index)134 static struct osd_dev *_ios_od(struct ore_io_state *ios, unsigned index)
135 {
136 ORE_DBGMSG2("oc->first_dev=%d oc->numdevs=%d i=%d oc->ods=%p\n",
137 ios->oc->first_dev, ios->oc->numdevs, index,
138 ios->oc->ods);
139
140 return ore_comp_dev(ios->oc, index);
141 }
142
_ore_get_io_state(struct ore_layout * layout,struct ore_components * oc,unsigned numdevs,unsigned sgs_per_dev,unsigned num_par_pages,struct ore_io_state ** pios)143 int _ore_get_io_state(struct ore_layout *layout,
144 struct ore_components *oc, unsigned numdevs,
145 unsigned sgs_per_dev, unsigned num_par_pages,
146 struct ore_io_state **pios)
147 {
148 struct ore_io_state *ios;
149 size_t size_ios, size_extra, size_total;
150 void *ios_extra;
151
152 /*
153 * The desired layout looks like this, with the extra_allocation
154 * items pointed at from fields within ios or per_dev:
155
156 struct __alloc_all_io_state {
157 struct ore_io_state ios;
158 struct ore_per_dev_state per_dev[numdevs];
159 union {
160 struct osd_sg_entry sglist[sgs_per_dev * numdevs];
161 struct page *pages[num_par_pages];
162 } extra_allocation;
163 } whole_allocation;
164
165 */
166
167 /* This should never happen, so abort early if it ever does. */
168 if (sgs_per_dev && num_par_pages) {
169 ORE_DBGMSG("Tried to use both pages and sglist\n");
170 *pios = NULL;
171 return -EINVAL;
172 }
173
174 if (numdevs > (INT_MAX - sizeof(*ios)) /
175 sizeof(struct ore_per_dev_state))
176 return -ENOMEM;
177 size_ios = sizeof(*ios) + sizeof(struct ore_per_dev_state) * numdevs;
178
179 if (sgs_per_dev * numdevs > INT_MAX / sizeof(struct osd_sg_entry))
180 return -ENOMEM;
181 if (num_par_pages > INT_MAX / sizeof(struct page *))
182 return -ENOMEM;
183 size_extra = max(sizeof(struct osd_sg_entry) * (sgs_per_dev * numdevs),
184 sizeof(struct page *) * num_par_pages);
185
186 size_total = size_ios + size_extra;
187
188 if (likely(size_total <= PAGE_SIZE)) {
189 ios = kzalloc(size_total, GFP_KERNEL);
190 if (unlikely(!ios)) {
191 ORE_DBGMSG("Failed kzalloc bytes=%zd\n", size_total);
192 *pios = NULL;
193 return -ENOMEM;
194 }
195 ios_extra = (char *)ios + size_ios;
196 } else {
197 ios = kzalloc(size_ios, GFP_KERNEL);
198 if (unlikely(!ios)) {
199 ORE_DBGMSG("Failed alloc first part bytes=%zd\n",
200 size_ios);
201 *pios = NULL;
202 return -ENOMEM;
203 }
204 ios_extra = kzalloc(size_extra, GFP_KERNEL);
205 if (unlikely(!ios_extra)) {
206 ORE_DBGMSG("Failed alloc second part bytes=%zd\n",
207 size_extra);
208 kfree(ios);
209 *pios = NULL;
210 return -ENOMEM;
211 }
212
213 /* In this case the per_dev[0].sgilist holds the pointer to
214 * be freed
215 */
216 ios->extra_part_alloc = true;
217 }
218
219 if (num_par_pages) {
220 ios->parity_pages = ios_extra;
221 ios->max_par_pages = num_par_pages;
222 }
223 if (sgs_per_dev) {
224 struct osd_sg_entry *sgilist = ios_extra;
225 unsigned d;
226
227 for (d = 0; d < numdevs; ++d) {
228 ios->per_dev[d].sglist = sgilist;
229 sgilist += sgs_per_dev;
230 }
231 ios->sgs_per_dev = sgs_per_dev;
232 }
233
234 ios->layout = layout;
235 ios->oc = oc;
236 *pios = ios;
237 return 0;
238 }
239
240 /* Allocate an io_state for only a single group of devices
241 *
242 * If a user needs to call ore_read/write() this version must be used becase it
243 * allocates extra stuff for striping and raid.
244 * The ore might decide to only IO less then @length bytes do to alignmets
245 * and constrains as follows:
246 * - The IO cannot cross group boundary.
247 * - In raid5/6 The end of the IO must align at end of a stripe eg.
248 * (@offset + @length) % strip_size == 0. Or the complete range is within a
249 * single stripe.
250 * - Memory condition only permitted a shorter IO. (A user can use @length=~0
251 * And check the returned ios->length for max_io_size.)
252 *
253 * The caller must check returned ios->length (and/or ios->nr_pages) and
254 * re-issue these pages that fall outside of ios->length
255 */
ore_get_rw_state(struct ore_layout * layout,struct ore_components * oc,bool is_reading,u64 offset,u64 length,struct ore_io_state ** pios)256 int ore_get_rw_state(struct ore_layout *layout, struct ore_components *oc,
257 bool is_reading, u64 offset, u64 length,
258 struct ore_io_state **pios)
259 {
260 struct ore_io_state *ios;
261 unsigned numdevs = layout->group_width * layout->mirrors_p1;
262 unsigned sgs_per_dev = 0, max_par_pages = 0;
263 int ret;
264
265 if (layout->parity && length) {
266 unsigned data_devs = layout->group_width - layout->parity;
267 unsigned stripe_size = layout->stripe_unit * data_devs;
268 unsigned pages_in_unit = layout->stripe_unit / PAGE_SIZE;
269 u32 remainder;
270 u64 num_stripes;
271 u64 num_raid_units;
272
273 num_stripes = div_u64_rem(length, stripe_size, &remainder);
274 if (remainder)
275 ++num_stripes;
276
277 num_raid_units = num_stripes * layout->parity;
278
279 if (is_reading) {
280 /* For reads add per_dev sglist array */
281 /* TODO: Raid 6 we need twice more. Actually:
282 * num_stripes / LCMdP(W,P);
283 * if (W%P != 0) num_stripes *= parity;
284 */
285
286 /* first/last seg is split */
287 num_raid_units += layout->group_width;
288 sgs_per_dev = div_u64(num_raid_units, data_devs) + 2;
289 } else {
290 /* For Writes add parity pages array. */
291 max_par_pages = num_raid_units * pages_in_unit *
292 sizeof(struct page *);
293 }
294 }
295
296 ret = _ore_get_io_state(layout, oc, numdevs, sgs_per_dev, max_par_pages,
297 pios);
298 if (unlikely(ret))
299 return ret;
300
301 ios = *pios;
302 ios->reading = is_reading;
303 ios->offset = offset;
304
305 if (length) {
306 ore_calc_stripe_info(layout, offset, length, &ios->si);
307 ios->length = ios->si.length;
308 ios->nr_pages = ((ios->offset & (PAGE_SIZE - 1)) +
309 ios->length + PAGE_SIZE - 1) / PAGE_SIZE;
310 if (layout->parity)
311 _ore_post_alloc_raid_stuff(ios);
312 }
313
314 return 0;
315 }
316 EXPORT_SYMBOL(ore_get_rw_state);
317
318 /* Allocate an io_state for all the devices in the comps array
319 *
320 * This version of io_state allocation is used mostly by create/remove
321 * and trunc where we currently need all the devices. The only wastful
322 * bit is the read/write_attributes with no IO. Those sites should
323 * be converted to use ore_get_rw_state() with length=0
324 */
ore_get_io_state(struct ore_layout * layout,struct ore_components * oc,struct ore_io_state ** pios)325 int ore_get_io_state(struct ore_layout *layout, struct ore_components *oc,
326 struct ore_io_state **pios)
327 {
328 return _ore_get_io_state(layout, oc, oc->numdevs, 0, 0, pios);
329 }
330 EXPORT_SYMBOL(ore_get_io_state);
331
ore_put_io_state(struct ore_io_state * ios)332 void ore_put_io_state(struct ore_io_state *ios)
333 {
334 if (ios) {
335 unsigned i;
336
337 for (i = 0; i < ios->numdevs; i++) {
338 struct ore_per_dev_state *per_dev = &ios->per_dev[i];
339
340 if (per_dev->or)
341 osd_end_request(per_dev->or);
342 if (per_dev->bio)
343 bio_put(per_dev->bio);
344 }
345
346 _ore_free_raid_stuff(ios);
347 kfree(ios);
348 }
349 }
350 EXPORT_SYMBOL(ore_put_io_state);
351
_sync_done(struct ore_io_state * ios,void * p)352 static void _sync_done(struct ore_io_state *ios, void *p)
353 {
354 struct completion *waiting = p;
355
356 complete(waiting);
357 }
358
_last_io(struct kref * kref)359 static void _last_io(struct kref *kref)
360 {
361 struct ore_io_state *ios = container_of(
362 kref, struct ore_io_state, kref);
363
364 ios->done(ios, ios->private);
365 }
366
_done_io(struct osd_request * or,void * p)367 static void _done_io(struct osd_request *or, void *p)
368 {
369 struct ore_io_state *ios = p;
370
371 kref_put(&ios->kref, _last_io);
372 }
373
ore_io_execute(struct ore_io_state * ios)374 int ore_io_execute(struct ore_io_state *ios)
375 {
376 DECLARE_COMPLETION_ONSTACK(wait);
377 bool sync = (ios->done == NULL);
378 int i, ret;
379
380 if (sync) {
381 ios->done = _sync_done;
382 ios->private = &wait;
383 }
384
385 for (i = 0; i < ios->numdevs; i++) {
386 struct osd_request *or = ios->per_dev[i].or;
387 if (unlikely(!or))
388 continue;
389
390 ret = osd_finalize_request(or, 0, _ios_cred(ios, i), NULL);
391 if (unlikely(ret)) {
392 ORE_DBGMSG("Failed to osd_finalize_request() => %d\n",
393 ret);
394 return ret;
395 }
396 }
397
398 kref_init(&ios->kref);
399
400 for (i = 0; i < ios->numdevs; i++) {
401 struct osd_request *or = ios->per_dev[i].or;
402 if (unlikely(!or))
403 continue;
404
405 kref_get(&ios->kref);
406 osd_execute_request_async(or, _done_io, ios);
407 }
408
409 kref_put(&ios->kref, _last_io);
410 ret = 0;
411
412 if (sync) {
413 wait_for_completion(&wait);
414 ret = ore_check_io(ios, NULL);
415 }
416 return ret;
417 }
418
_clear_bio(struct bio * bio)419 static void _clear_bio(struct bio *bio)
420 {
421 struct bio_vec *bv;
422 unsigned i;
423
424 bio_for_each_segment_all(bv, bio, i) {
425 unsigned this_count = bv->bv_len;
426
427 if (likely(PAGE_SIZE == this_count))
428 clear_highpage(bv->bv_page);
429 else
430 zero_user(bv->bv_page, bv->bv_offset, this_count);
431 }
432 }
433
ore_check_io(struct ore_io_state * ios,ore_on_dev_error on_dev_error)434 int ore_check_io(struct ore_io_state *ios, ore_on_dev_error on_dev_error)
435 {
436 enum osd_err_priority acumulated_osd_err = 0;
437 int acumulated_lin_err = 0;
438 int i;
439
440 for (i = 0; i < ios->numdevs; i++) {
441 struct osd_sense_info osi;
442 struct ore_per_dev_state *per_dev = &ios->per_dev[i];
443 struct osd_request *or = per_dev->or;
444 int ret;
445
446 if (unlikely(!or))
447 continue;
448
449 ret = osd_req_decode_sense(or, &osi);
450 if (likely(!ret))
451 continue;
452
453 if ((OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) &&
454 per_dev->bio) {
455 /* start read offset passed endof file.
456 * Note: if we do not have bio it means read-attributes
457 * In this case we should return error to caller.
458 */
459 _clear_bio(per_dev->bio);
460 ORE_DBGMSG("start read offset passed end of file "
461 "offset=0x%llx, length=0x%llx\n",
462 _LLU(per_dev->offset),
463 _LLU(per_dev->length));
464
465 continue; /* we recovered */
466 }
467
468 if (on_dev_error) {
469 u64 residual = ios->reading ?
470 or->in.residual : or->out.residual;
471 u64 offset = (ios->offset + ios->length) - residual;
472 unsigned dev = per_dev->dev - ios->oc->first_dev;
473 struct ore_dev *od = ios->oc->ods[dev];
474
475 on_dev_error(ios, od, dev, osi.osd_err_pri,
476 offset, residual);
477 }
478 if (osi.osd_err_pri >= acumulated_osd_err) {
479 acumulated_osd_err = osi.osd_err_pri;
480 acumulated_lin_err = ret;
481 }
482 }
483
484 return acumulated_lin_err;
485 }
486 EXPORT_SYMBOL(ore_check_io);
487
488 /*
489 * L - logical offset into the file
490 *
491 * D - number of Data devices
492 * D = group_width - parity
493 *
494 * U - The number of bytes in a stripe within a group
495 * U = stripe_unit * D
496 *
497 * T - The number of bytes striped within a group of component objects
498 * (before advancing to the next group)
499 * T = U * group_depth
500 *
501 * S - The number of bytes striped across all component objects
502 * before the pattern repeats
503 * S = T * group_count
504 *
505 * M - The "major" (i.e., across all components) cycle number
506 * M = L / S
507 *
508 * G - Counts the groups from the beginning of the major cycle
509 * G = (L - (M * S)) / T [or (L % S) / T]
510 *
511 * H - The byte offset within the group
512 * H = (L - (M * S)) % T [or (L % S) % T]
513 *
514 * N - The "minor" (i.e., across the group) stripe number
515 * N = H / U
516 *
517 * C - The component index coresponding to L
518 *
519 * C = (H - (N * U)) / stripe_unit + G * D
520 * [or (L % U) / stripe_unit + G * D]
521 *
522 * O - The component offset coresponding to L
523 * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
524 *
525 * LCMdP – Parity cycle: Lowest Common Multiple of group_width, parity
526 * divide by parity
527 * LCMdP = lcm(group_width, parity) / parity
528 *
529 * R - The parity Rotation stripe
530 * (Note parity cycle always starts at a group's boundary)
531 * R = N % LCMdP
532 *
533 * I = the first parity device index
534 * I = (group_width + group_width - R*parity - parity) % group_width
535 *
536 * Craid - The component index Rotated
537 * Craid = (group_width + C - R*parity) % group_width
538 * (We add the group_width to avoid negative numbers modulo math)
539 */
ore_calc_stripe_info(struct ore_layout * layout,u64 file_offset,u64 length,struct ore_striping_info * si)540 void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset,
541 u64 length, struct ore_striping_info *si)
542 {
543 u32 stripe_unit = layout->stripe_unit;
544 u32 group_width = layout->group_width;
545 u64 group_depth = layout->group_depth;
546 u32 parity = layout->parity;
547
548 u32 D = group_width - parity;
549 u32 U = D * stripe_unit;
550 u64 T = U * group_depth;
551 u64 S = T * layout->group_count;
552 u64 M = div64_u64(file_offset, S);
553
554 /*
555 G = (L - (M * S)) / T
556 H = (L - (M * S)) % T
557 */
558 u64 LmodS = file_offset - M * S;
559 u32 G = div64_u64(LmodS, T);
560 u64 H = LmodS - G * T;
561
562 u32 N = div_u64(H, U);
563 u32 Nlast;
564
565 /* "H - (N * U)" is just "H % U" so it's bound to u32 */
566 u32 C = (u32)(H - (N * U)) / stripe_unit + G * group_width;
567 u32 first_dev = C - C % group_width;
568
569 div_u64_rem(file_offset, stripe_unit, &si->unit_off);
570
571 si->obj_offset = si->unit_off + (N * stripe_unit) +
572 (M * group_depth * stripe_unit);
573 si->cur_comp = C - first_dev;
574 si->cur_pg = si->unit_off / PAGE_SIZE;
575
576 if (parity) {
577 u32 LCMdP = lcm(group_width, parity) / parity;
578 /* R = N % LCMdP; */
579 u32 RxP = (N % LCMdP) * parity;
580
581 si->par_dev = (group_width + group_width - parity - RxP) %
582 group_width + first_dev;
583 si->dev = (group_width + group_width + C - RxP) %
584 group_width + first_dev;
585 si->bytes_in_stripe = U;
586 si->first_stripe_start = M * S + G * T + N * U;
587 } else {
588 /* Make the math correct see _prepare_one_group */
589 si->par_dev = group_width;
590 si->dev = C;
591 }
592
593 si->dev *= layout->mirrors_p1;
594 si->par_dev *= layout->mirrors_p1;
595 si->offset = file_offset;
596 si->length = T - H;
597 if (si->length > length)
598 si->length = length;
599
600 Nlast = div_u64(H + si->length + U - 1, U);
601 si->maxdevUnits = Nlast - N;
602
603 si->M = M;
604 }
605 EXPORT_SYMBOL(ore_calc_stripe_info);
606
_ore_add_stripe_unit(struct ore_io_state * ios,unsigned * cur_pg,unsigned pgbase,struct page ** pages,struct ore_per_dev_state * per_dev,int cur_len)607 int _ore_add_stripe_unit(struct ore_io_state *ios, unsigned *cur_pg,
608 unsigned pgbase, struct page **pages,
609 struct ore_per_dev_state *per_dev, int cur_len)
610 {
611 unsigned pg = *cur_pg;
612 struct request_queue *q =
613 osd_request_queue(_ios_od(ios, per_dev->dev));
614 unsigned len = cur_len;
615 int ret;
616
617 if (per_dev->bio == NULL) {
618 unsigned bio_size;
619
620 if (!ios->reading) {
621 bio_size = ios->si.maxdevUnits;
622 } else {
623 bio_size = (ios->si.maxdevUnits + 1) *
624 (ios->layout->group_width - ios->layout->parity) /
625 ios->layout->group_width;
626 }
627 bio_size *= (ios->layout->stripe_unit / PAGE_SIZE);
628
629 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
630 if (unlikely(!per_dev->bio)) {
631 ORE_DBGMSG("Failed to allocate BIO size=%u\n",
632 bio_size);
633 ret = -ENOMEM;
634 goto out;
635 }
636 }
637
638 while (cur_len > 0) {
639 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
640 unsigned added_len;
641
642 cur_len -= pglen;
643
644 added_len = bio_add_pc_page(q, per_dev->bio, pages[pg],
645 pglen, pgbase);
646 if (unlikely(pglen != added_len)) {
647 /* If bi_vcnt == bi_max then this is a SW BUG */
648 ORE_DBGMSG("Failed bio_add_pc_page bi_vcnt=0x%x "
649 "bi_max=0x%x BIO_MAX=0x%x cur_len=0x%x\n",
650 per_dev->bio->bi_vcnt,
651 per_dev->bio->bi_max_vecs,
652 BIO_MAX_PAGES_KMALLOC, cur_len);
653 ret = -ENOMEM;
654 goto out;
655 }
656 _add_stripe_page(ios->sp2d, &ios->si, pages[pg]);
657
658 pgbase = 0;
659 ++pg;
660 }
661 BUG_ON(cur_len);
662
663 per_dev->length += len;
664 *cur_pg = pg;
665 ret = 0;
666 out: /* we fail the complete unit on an error eg don't advance
667 * per_dev->length and cur_pg. This means that we might have a bigger
668 * bio than the CDB requested length (per_dev->length). That's fine
669 * only the oposite is fatal.
670 */
671 return ret;
672 }
673
_add_parity_units(struct ore_io_state * ios,struct ore_striping_info * si,unsigned dev,unsigned first_dev,unsigned mirrors_p1,unsigned devs_in_group,unsigned cur_len)674 static int _add_parity_units(struct ore_io_state *ios,
675 struct ore_striping_info *si,
676 unsigned dev, unsigned first_dev,
677 unsigned mirrors_p1, unsigned devs_in_group,
678 unsigned cur_len)
679 {
680 unsigned do_parity;
681 int ret = 0;
682
683 for (do_parity = ios->layout->parity; do_parity; --do_parity) {
684 struct ore_per_dev_state *per_dev;
685
686 per_dev = &ios->per_dev[dev - first_dev];
687 if (!per_dev->length && !per_dev->offset) {
688 /* Only/always the parity unit of the first
689 * stripe will be empty. So this is a chance to
690 * initialize the per_dev info.
691 */
692 per_dev->dev = dev;
693 per_dev->offset = si->obj_offset - si->unit_off;
694 }
695
696 ret = _ore_add_parity_unit(ios, si, per_dev, cur_len,
697 do_parity == 1);
698 if (unlikely(ret))
699 break;
700
701 if (do_parity != 1) {
702 dev = ((dev + mirrors_p1) % devs_in_group) + first_dev;
703 si->cur_comp = (si->cur_comp + 1) %
704 ios->layout->group_width;
705 }
706 }
707
708 return ret;
709 }
710
_prepare_for_striping(struct ore_io_state * ios)711 static int _prepare_for_striping(struct ore_io_state *ios)
712 {
713 struct ore_striping_info *si = &ios->si;
714 unsigned stripe_unit = ios->layout->stripe_unit;
715 unsigned mirrors_p1 = ios->layout->mirrors_p1;
716 unsigned group_width = ios->layout->group_width;
717 unsigned devs_in_group = group_width * mirrors_p1;
718 unsigned dev = si->dev;
719 unsigned first_dev = dev - (dev % devs_in_group);
720 unsigned cur_pg = ios->pages_consumed;
721 u64 length = ios->length;
722 int ret = 0;
723
724 if (!ios->pages) {
725 ios->numdevs = ios->layout->mirrors_p1;
726 return 0;
727 }
728
729 BUG_ON(length > si->length);
730
731 while (length) {
732 struct ore_per_dev_state *per_dev =
733 &ios->per_dev[dev - first_dev];
734 unsigned cur_len, page_off = 0;
735
736 if (!per_dev->length && !per_dev->offset) {
737 /* First time initialize the per_dev info. */
738 per_dev->dev = dev;
739 if (dev == si->dev) {
740 WARN_ON(dev == si->par_dev);
741 per_dev->offset = si->obj_offset;
742 cur_len = stripe_unit - si->unit_off;
743 page_off = si->unit_off & ~PAGE_MASK;
744 BUG_ON(page_off && (page_off != ios->pgbase));
745 } else {
746 per_dev->offset = si->obj_offset - si->unit_off;
747 cur_len = stripe_unit;
748 }
749 } else {
750 cur_len = stripe_unit;
751 }
752 if (cur_len >= length)
753 cur_len = length;
754
755 ret = _ore_add_stripe_unit(ios, &cur_pg, page_off, ios->pages,
756 per_dev, cur_len);
757 if (unlikely(ret))
758 goto out;
759
760 length -= cur_len;
761
762 dev = ((dev + mirrors_p1) % devs_in_group) + first_dev;
763 si->cur_comp = (si->cur_comp + 1) % group_width;
764 if (unlikely((dev == si->par_dev) || (!length && ios->sp2d))) {
765 if (!length && ios->sp2d) {
766 /* If we are writing and this is the very last
767 * stripe. then operate on parity dev.
768 */
769 dev = si->par_dev;
770 /* If last stripe operate on parity comp */
771 si->cur_comp = group_width - ios->layout->parity;
772 }
773
774 /* In writes cur_len just means if it's the
775 * last one. See _ore_add_parity_unit.
776 */
777 ret = _add_parity_units(ios, si, dev, first_dev,
778 mirrors_p1, devs_in_group,
779 ios->sp2d ? length : cur_len);
780 if (unlikely(ret))
781 goto out;
782
783 /* Rotate next par_dev backwards with wraping */
784 si->par_dev = (devs_in_group + si->par_dev -
785 ios->layout->parity * mirrors_p1) %
786 devs_in_group + first_dev;
787 /* Next stripe, start fresh */
788 si->cur_comp = 0;
789 si->cur_pg = 0;
790 si->obj_offset += cur_len;
791 si->unit_off = 0;
792 }
793 }
794 out:
795 ios->numdevs = devs_in_group;
796 ios->pages_consumed = cur_pg;
797 return ret;
798 }
799
ore_create(struct ore_io_state * ios)800 int ore_create(struct ore_io_state *ios)
801 {
802 int i, ret;
803
804 for (i = 0; i < ios->oc->numdevs; i++) {
805 struct osd_request *or;
806
807 or = osd_start_request(_ios_od(ios, i));
808 if (unlikely(!or)) {
809 ORE_ERR("%s: osd_start_request failed\n", __func__);
810 ret = -ENOMEM;
811 goto out;
812 }
813 ios->per_dev[i].or = or;
814 ios->numdevs++;
815
816 osd_req_create_object(or, _ios_obj(ios, i));
817 }
818 ret = ore_io_execute(ios);
819
820 out:
821 return ret;
822 }
823 EXPORT_SYMBOL(ore_create);
824
ore_remove(struct ore_io_state * ios)825 int ore_remove(struct ore_io_state *ios)
826 {
827 int i, ret;
828
829 for (i = 0; i < ios->oc->numdevs; i++) {
830 struct osd_request *or;
831
832 or = osd_start_request(_ios_od(ios, i));
833 if (unlikely(!or)) {
834 ORE_ERR("%s: osd_start_request failed\n", __func__);
835 ret = -ENOMEM;
836 goto out;
837 }
838 ios->per_dev[i].or = or;
839 ios->numdevs++;
840
841 osd_req_remove_object(or, _ios_obj(ios, i));
842 }
843 ret = ore_io_execute(ios);
844
845 out:
846 return ret;
847 }
848 EXPORT_SYMBOL(ore_remove);
849
_write_mirror(struct ore_io_state * ios,int cur_comp)850 static int _write_mirror(struct ore_io_state *ios, int cur_comp)
851 {
852 struct ore_per_dev_state *master_dev = &ios->per_dev[cur_comp];
853 unsigned dev = ios->per_dev[cur_comp].dev;
854 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
855 int ret = 0;
856
857 if (ios->pages && !master_dev->length)
858 return 0; /* Just an empty slot */
859
860 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
861 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
862 struct osd_request *or;
863
864 or = osd_start_request(_ios_od(ios, dev));
865 if (unlikely(!or)) {
866 ORE_ERR("%s: osd_start_request failed\n", __func__);
867 ret = -ENOMEM;
868 goto out;
869 }
870 per_dev->or = or;
871
872 if (ios->pages) {
873 struct bio *bio;
874
875 if (per_dev != master_dev) {
876 bio = bio_clone_fast(master_dev->bio,
877 GFP_KERNEL, NULL);
878 if (unlikely(!bio)) {
879 ORE_DBGMSG(
880 "Failed to allocate BIO size=%u\n",
881 master_dev->bio->bi_max_vecs);
882 ret = -ENOMEM;
883 goto out;
884 }
885
886 bio->bi_disk = NULL;
887 bio->bi_next = NULL;
888 per_dev->offset = master_dev->offset;
889 per_dev->length = master_dev->length;
890 per_dev->bio = bio;
891 per_dev->dev = dev;
892 } else {
893 bio = master_dev->bio;
894 /* FIXME: bio_set_dir() */
895 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
896 }
897
898 osd_req_write(or, _ios_obj(ios, cur_comp),
899 per_dev->offset, bio, per_dev->length);
900 ORE_DBGMSG("write(0x%llx) offset=0x%llx "
901 "length=0x%llx dev=%d\n",
902 _LLU(_ios_obj(ios, cur_comp)->id),
903 _LLU(per_dev->offset),
904 _LLU(per_dev->length), dev);
905 } else if (ios->kern_buff) {
906 per_dev->offset = ios->si.obj_offset;
907 per_dev->dev = ios->si.dev + dev;
908
909 /* no cross device without page array */
910 BUG_ON((ios->layout->group_width > 1) &&
911 (ios->si.unit_off + ios->length >
912 ios->layout->stripe_unit));
913
914 ret = osd_req_write_kern(or, _ios_obj(ios, cur_comp),
915 per_dev->offset,
916 ios->kern_buff, ios->length);
917 if (unlikely(ret))
918 goto out;
919 ORE_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
920 "length=0x%llx dev=%d\n",
921 _LLU(_ios_obj(ios, cur_comp)->id),
922 _LLU(per_dev->offset),
923 _LLU(ios->length), per_dev->dev);
924 } else {
925 osd_req_set_attributes(or, _ios_obj(ios, cur_comp));
926 ORE_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
927 _LLU(_ios_obj(ios, cur_comp)->id),
928 ios->out_attr_len, dev);
929 }
930
931 if (ios->out_attr)
932 osd_req_add_set_attr_list(or, ios->out_attr,
933 ios->out_attr_len);
934
935 if (ios->in_attr)
936 osd_req_add_get_attr_list(or, ios->in_attr,
937 ios->in_attr_len);
938 }
939
940 out:
941 return ret;
942 }
943
ore_write(struct ore_io_state * ios)944 int ore_write(struct ore_io_state *ios)
945 {
946 int i;
947 int ret;
948
949 if (unlikely(ios->sp2d && !ios->r4w)) {
950 /* A library is attempting a RAID-write without providing
951 * a pages lock interface.
952 */
953 WARN_ON_ONCE(1);
954 return -ENOTSUPP;
955 }
956
957 ret = _prepare_for_striping(ios);
958 if (unlikely(ret))
959 return ret;
960
961 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
962 ret = _write_mirror(ios, i);
963 if (unlikely(ret))
964 return ret;
965 }
966
967 ret = ore_io_execute(ios);
968 return ret;
969 }
970 EXPORT_SYMBOL(ore_write);
971
_ore_read_mirror(struct ore_io_state * ios,unsigned cur_comp)972 int _ore_read_mirror(struct ore_io_state *ios, unsigned cur_comp)
973 {
974 struct osd_request *or;
975 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
976 struct osd_obj_id *obj = _ios_obj(ios, cur_comp);
977 unsigned first_dev = (unsigned)obj->id;
978
979 if (ios->pages && !per_dev->length)
980 return 0; /* Just an empty slot */
981
982 first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
983 or = osd_start_request(_ios_od(ios, first_dev));
984 if (unlikely(!or)) {
985 ORE_ERR("%s: osd_start_request failed\n", __func__);
986 return -ENOMEM;
987 }
988 per_dev->or = or;
989
990 if (ios->pages) {
991 if (per_dev->cur_sg) {
992 /* finalize the last sg_entry */
993 _ore_add_sg_seg(per_dev, 0, false);
994 if (unlikely(!per_dev->cur_sg))
995 return 0; /* Skip parity only device */
996
997 osd_req_read_sg(or, obj, per_dev->bio,
998 per_dev->sglist, per_dev->cur_sg);
999 } else {
1000 /* The no raid case */
1001 osd_req_read(or, obj, per_dev->offset,
1002 per_dev->bio, per_dev->length);
1003 }
1004
1005 ORE_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
1006 " dev=%d sg_len=%d\n", _LLU(obj->id),
1007 _LLU(per_dev->offset), _LLU(per_dev->length),
1008 first_dev, per_dev->cur_sg);
1009 } else {
1010 BUG_ON(ios->kern_buff);
1011
1012 osd_req_get_attributes(or, obj);
1013 ORE_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
1014 _LLU(obj->id),
1015 ios->in_attr_len, first_dev);
1016 }
1017 if (ios->out_attr)
1018 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
1019
1020 if (ios->in_attr)
1021 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
1022
1023 return 0;
1024 }
1025
ore_read(struct ore_io_state * ios)1026 int ore_read(struct ore_io_state *ios)
1027 {
1028 int i;
1029 int ret;
1030
1031 ret = _prepare_for_striping(ios);
1032 if (unlikely(ret))
1033 return ret;
1034
1035 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
1036 ret = _ore_read_mirror(ios, i);
1037 if (unlikely(ret))
1038 return ret;
1039 }
1040
1041 ret = ore_io_execute(ios);
1042 return ret;
1043 }
1044 EXPORT_SYMBOL(ore_read);
1045
extract_attr_from_ios(struct ore_io_state * ios,struct osd_attr * attr)1046 int extract_attr_from_ios(struct ore_io_state *ios, struct osd_attr *attr)
1047 {
1048 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
1049 void *iter = NULL;
1050 int nelem;
1051
1052 do {
1053 nelem = 1;
1054 osd_req_decode_get_attr_list(ios->per_dev[0].or,
1055 &cur_attr, &nelem, &iter);
1056 if ((cur_attr.attr_page == attr->attr_page) &&
1057 (cur_attr.attr_id == attr->attr_id)) {
1058 attr->len = cur_attr.len;
1059 attr->val_ptr = cur_attr.val_ptr;
1060 return 0;
1061 }
1062 } while (iter);
1063
1064 return -EIO;
1065 }
1066 EXPORT_SYMBOL(extract_attr_from_ios);
1067
_truncate_mirrors(struct ore_io_state * ios,unsigned cur_comp,struct osd_attr * attr)1068 static int _truncate_mirrors(struct ore_io_state *ios, unsigned cur_comp,
1069 struct osd_attr *attr)
1070 {
1071 int last_comp = cur_comp + ios->layout->mirrors_p1;
1072
1073 for (; cur_comp < last_comp; ++cur_comp) {
1074 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
1075 struct osd_request *or;
1076
1077 or = osd_start_request(_ios_od(ios, cur_comp));
1078 if (unlikely(!or)) {
1079 ORE_ERR("%s: osd_start_request failed\n", __func__);
1080 return -ENOMEM;
1081 }
1082 per_dev->or = or;
1083
1084 osd_req_set_attributes(or, _ios_obj(ios, cur_comp));
1085 osd_req_add_set_attr_list(or, attr, 1);
1086 }
1087
1088 return 0;
1089 }
1090
1091 struct _trunc_info {
1092 struct ore_striping_info si;
1093 u64 prev_group_obj_off;
1094 u64 next_group_obj_off;
1095
1096 unsigned first_group_dev;
1097 unsigned nex_group_dev;
1098 };
1099
_calc_trunk_info(struct ore_layout * layout,u64 file_offset,struct _trunc_info * ti)1100 static void _calc_trunk_info(struct ore_layout *layout, u64 file_offset,
1101 struct _trunc_info *ti)
1102 {
1103 unsigned stripe_unit = layout->stripe_unit;
1104
1105 ore_calc_stripe_info(layout, file_offset, 0, &ti->si);
1106
1107 ti->prev_group_obj_off = ti->si.M * stripe_unit;
1108 ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
1109
1110 ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
1111 ti->nex_group_dev = ti->first_group_dev + layout->group_width;
1112 }
1113
ore_truncate(struct ore_layout * layout,struct ore_components * oc,u64 size)1114 int ore_truncate(struct ore_layout *layout, struct ore_components *oc,
1115 u64 size)
1116 {
1117 struct ore_io_state *ios;
1118 struct exofs_trunc_attr {
1119 struct osd_attr attr;
1120 __be64 newsize;
1121 } *size_attrs;
1122 struct _trunc_info ti;
1123 int i, ret;
1124
1125 ret = ore_get_io_state(layout, oc, &ios);
1126 if (unlikely(ret))
1127 return ret;
1128
1129 _calc_trunk_info(ios->layout, size, &ti);
1130
1131 size_attrs = kcalloc(ios->oc->numdevs, sizeof(*size_attrs),
1132 GFP_KERNEL);
1133 if (unlikely(!size_attrs)) {
1134 ret = -ENOMEM;
1135 goto out;
1136 }
1137
1138 ios->numdevs = ios->oc->numdevs;
1139
1140 for (i = 0; i < ios->numdevs; ++i) {
1141 struct exofs_trunc_attr *size_attr = &size_attrs[i];
1142 u64 obj_size;
1143
1144 if (i < ti.first_group_dev)
1145 obj_size = ti.prev_group_obj_off;
1146 else if (i >= ti.nex_group_dev)
1147 obj_size = ti.next_group_obj_off;
1148 else if (i < ti.si.dev) /* dev within this group */
1149 obj_size = ti.si.obj_offset +
1150 ios->layout->stripe_unit - ti.si.unit_off;
1151 else if (i == ti.si.dev)
1152 obj_size = ti.si.obj_offset;
1153 else /* i > ti.dev */
1154 obj_size = ti.si.obj_offset - ti.si.unit_off;
1155
1156 size_attr->newsize = cpu_to_be64(obj_size);
1157 size_attr->attr = g_attr_logical_length;
1158 size_attr->attr.val_ptr = &size_attr->newsize;
1159
1160 ORE_DBGMSG2("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
1161 _LLU(oc->comps->obj.id), _LLU(obj_size), i);
1162 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
1163 &size_attr->attr);
1164 if (unlikely(ret))
1165 goto out;
1166 }
1167 ret = ore_io_execute(ios);
1168
1169 out:
1170 kfree(size_attrs);
1171 ore_put_io_state(ios);
1172 return ret;
1173 }
1174 EXPORT_SYMBOL(ore_truncate);
1175
1176 const struct osd_attr g_attr_logical_length = ATTR_DEF(
1177 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
1178 EXPORT_SYMBOL(g_attr_logical_length);
1179