blob: 08ee454b218703ac436f580fd1a812aed9f0a34b [file] [log] [blame]
Boaz Harroshb14f8ab2008-10-27 18:27:55 +02001/*
2 * Copyright (C) 2005, 2006
Boaz Harrosh27d2e142009-06-14 17:23:09 +03003 * Avishay Traeger (avishay@gmail.com)
Boaz Harroshb14f8ab2008-10-27 18:27:55 +02004 * 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
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090025#include <linux/slab.h>
Boaz Harrosh5d952b82010-02-01 13:35:51 +020026#include <asm/div64.h>
Boaz Harrosha1fec1d2011-10-12 18:42:22 +020027#include <linux/lcm.h>
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020028
Boaz Harrosha1fec1d2011-10-12 18:42:22 +020029#include "ore_raid.h"
Boaz Harrosh8ff660a2011-08-06 19:26:31 -070030
Boaz Harroshcf283ad2011-08-06 19:22:06 -070031MODULE_AUTHOR("Boaz Harrosh <bharrosh@panasas.com>");
32MODULE_DESCRIPTION("Objects Raid Engine ore.ko");
33MODULE_LICENSE("GPL");
34
Boaz Harrosh5a51c0c2011-09-28 13:18:45 +030035/* ore_verify_layout does a couple of things:
36 * 1. Given a minimum number of needed parameters fixes up the rest of the
37 * members to be operatonals for the ore. The needed parameters are those
38 * that are defined by the pnfs-objects layout STD.
39 * 2. Check to see if the current ore code actually supports these parameters
40 * for example stripe_unit must be a multple of the system PAGE_SIZE,
41 * and etc...
42 * 3. Cache some havily used calculations that will be needed by users.
43 */
44
Boaz Harrosh5a51c0c2011-09-28 13:18:45 +030045enum { BIO_MAX_PAGES_KMALLOC =
46 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),};
47
48int ore_verify_layout(unsigned total_comps, struct ore_layout *layout)
49{
50 u64 stripe_length;
51
52/* FIXME: Only raid0 is supported for now. */
53 if (layout->raid_algorithm != PNFS_OSD_RAID_0) {
54 ORE_ERR("Only RAID_0 for now\n");
55 return -EINVAL;
56 }
57 if (0 != (layout->stripe_unit & ~PAGE_MASK)) {
58 ORE_ERR("Stripe Unit(0x%llx)"
59 " must be Multples of PAGE_SIZE(0x%lx)\n",
60 _LLU(layout->stripe_unit), PAGE_SIZE);
61 return -EINVAL;
62 }
63 if (layout->group_width) {
64 if (!layout->group_depth) {
65 ORE_ERR("group_depth == 0 && group_width != 0\n");
66 return -EINVAL;
67 }
68 if (total_comps < (layout->group_width * layout->mirrors_p1)) {
69 ORE_ERR("Data Map wrong, "
70 "numdevs=%d < group_width=%d * mirrors=%d\n",
71 total_comps, layout->group_width,
72 layout->mirrors_p1);
73 return -EINVAL;
74 }
75 layout->group_count = total_comps / layout->mirrors_p1 /
76 layout->group_width;
77 } else {
78 if (layout->group_depth) {
79 printk(KERN_NOTICE "Warning: group_depth ignored "
80 "group_width == 0 && group_depth == %lld\n",
81 _LLU(layout->group_depth));
82 }
83 layout->group_width = total_comps / layout->mirrors_p1;
84 layout->group_depth = -1;
85 layout->group_count = 1;
86 }
87
88 stripe_length = (u64)layout->group_width * layout->stripe_unit;
89 if (stripe_length >= (1ULL << 32)) {
90 ORE_ERR("Stripe_length(0x%llx) >= 32bit is not supported\n",
91 _LLU(stripe_length));
92 return -EINVAL;
93 }
94
95 layout->max_io_length =
96 (BIO_MAX_PAGES_KMALLOC * PAGE_SIZE - layout->stripe_unit) *
97 layout->group_width;
Boaz Harrosh769ba8d2011-10-14 15:33:51 +020098 if (layout->parity) {
99 unsigned stripe_length =
100 (layout->group_width - layout->parity) *
101 layout->stripe_unit;
102
103 layout->max_io_length /= stripe_length;
104 layout->max_io_length *= stripe_length;
105 }
Boaz Harrosh5a51c0c2011-09-28 13:18:45 +0300106 return 0;
107}
108EXPORT_SYMBOL(ore_verify_layout);
109
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700110static u8 *_ios_cred(struct ore_io_state *ios, unsigned index)
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700111{
Boaz Harrosh5bf696d2011-09-28 11:39:59 +0300112 return ios->oc->comps[index & ios->oc->single_comp].cred;
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700113}
114
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700115static struct osd_obj_id *_ios_obj(struct ore_io_state *ios, unsigned index)
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700116{
Boaz Harrosh5bf696d2011-09-28 11:39:59 +0300117 return &ios->oc->comps[index & ios->oc->single_comp].obj;
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700118}
119
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700120static struct osd_dev *_ios_od(struct ore_io_state *ios, unsigned index)
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700121{
Boaz Harrosh3bd98562011-09-28 12:04:23 +0300122 ORE_DBGMSG2("oc->first_dev=%d oc->numdevs=%d i=%d oc->ods=%p\n",
123 ios->oc->first_dev, ios->oc->numdevs, index,
124 ios->oc->ods);
125
Boaz Harroshd866d872011-09-28 14:43:09 +0300126 return ore_comp_dev(ios->oc, index);
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700127}
128
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200129int _ore_get_io_state(struct ore_layout *layout,
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200130 struct ore_components *oc, unsigned numdevs,
131 unsigned sgs_per_dev, unsigned num_par_pages,
132 struct ore_io_state **pios)
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200133{
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700134 struct ore_io_state *ios;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200135 struct page **pages;
136 struct osd_sg_entry *sgilist;
137 struct __alloc_all_io_state {
138 struct ore_io_state ios;
139 struct ore_per_dev_state per_dev[numdevs];
140 union {
141 struct osd_sg_entry sglist[sgs_per_dev * numdevs];
142 struct page *pages[num_par_pages];
143 };
144 } *_aios;
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200145
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200146 if (likely(sizeof(*_aios) <= PAGE_SIZE)) {
147 _aios = kzalloc(sizeof(*_aios), GFP_KERNEL);
148 if (unlikely(!_aios)) {
149 ORE_DBGMSG("Failed kzalloc bytes=%zd\n",
150 sizeof(*_aios));
151 *pios = NULL;
152 return -ENOMEM;
153 }
154 pages = num_par_pages ? _aios->pages : NULL;
155 sgilist = sgs_per_dev ? _aios->sglist : NULL;
156 ios = &_aios->ios;
157 } else {
158 struct __alloc_small_io_state {
159 struct ore_io_state ios;
160 struct ore_per_dev_state per_dev[numdevs];
161 } *_aio_small;
162 union __extra_part {
163 struct osd_sg_entry sglist[sgs_per_dev * numdevs];
164 struct page *pages[num_par_pages];
165 } *extra_part;
166
167 _aio_small = kzalloc(sizeof(*_aio_small), GFP_KERNEL);
168 if (unlikely(!_aio_small)) {
169 ORE_DBGMSG("Failed alloc first part bytes=%zd\n",
170 sizeof(*_aio_small));
171 *pios = NULL;
172 return -ENOMEM;
173 }
174 extra_part = kzalloc(sizeof(*extra_part), GFP_KERNEL);
175 if (unlikely(!extra_part)) {
176 ORE_DBGMSG("Failed alloc second part bytes=%zd\n",
177 sizeof(*extra_part));
178 kfree(_aio_small);
179 *pios = NULL;
180 return -ENOMEM;
181 }
182
183 pages = num_par_pages ? extra_part->pages : NULL;
184 sgilist = sgs_per_dev ? extra_part->sglist : NULL;
185 /* In this case the per_dev[0].sgilist holds the pointer to
186 * be freed
187 */
188 ios = &_aio_small->ios;
189 ios->extra_part_alloc = true;
190 }
191
192 if (pages) {
193 ios->parity_pages = pages;
194 ios->max_par_pages = num_par_pages;
195 }
196 if (sgilist) {
197 unsigned d;
198
199 for (d = 0; d < numdevs; ++d) {
200 ios->per_dev[d].sglist = sgilist;
201 sgilist += sgs_per_dev;
202 }
203 ios->sgs_per_dev = sgs_per_dev;
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200204 }
205
Boaz Harrosh45d3abc2010-01-28 11:46:16 +0200206 ios->layout = layout;
Boaz Harrosh5bf696d2011-09-28 11:39:59 +0300207 ios->oc = oc;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200208 *pios = ios;
209 return 0;
210}
Boaz Harroshb916c5c2011-09-28 11:55:51 +0300211
212/* Allocate an io_state for only a single group of devices
213 *
214 * If a user needs to call ore_read/write() this version must be used becase it
215 * allocates extra stuff for striping and raid.
216 * The ore might decide to only IO less then @length bytes do to alignmets
217 * and constrains as follows:
218 * - The IO cannot cross group boundary.
219 * - In raid5/6 The end of the IO must align at end of a stripe eg.
220 * (@offset + @length) % strip_size == 0. Or the complete range is within a
221 * single stripe.
222 * - Memory condition only permitted a shorter IO. (A user can use @length=~0
223 * And check the returned ios->length for max_io_size.)
224 *
225 * The caller must check returned ios->length (and/or ios->nr_pages) and
226 * re-issue these pages that fall outside of ios->length
227 */
228int ore_get_rw_state(struct ore_layout *layout, struct ore_components *oc,
229 bool is_reading, u64 offset, u64 length,
230 struct ore_io_state **pios)
231{
232 struct ore_io_state *ios;
233 unsigned numdevs = layout->group_width * layout->mirrors_p1;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200234 unsigned sgs_per_dev = 0, max_par_pages = 0;
Boaz Harroshb916c5c2011-09-28 11:55:51 +0300235 int ret;
236
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200237 if (layout->parity && length) {
238 unsigned data_devs = layout->group_width - layout->parity;
239 unsigned stripe_size = layout->stripe_unit * data_devs;
240 unsigned pages_in_unit = layout->stripe_unit / PAGE_SIZE;
241 u32 remainder;
242 u64 num_stripes;
243 u64 num_raid_units;
244
245 num_stripes = div_u64_rem(length, stripe_size, &remainder);
246 if (remainder)
247 ++num_stripes;
248
249 num_raid_units = num_stripes * layout->parity;
250
251 if (is_reading) {
252 /* For reads add per_dev sglist array */
253 /* TODO: Raid 6 we need twice more. Actually:
254 * num_stripes / LCMdP(W,P);
255 * if (W%P != 0) num_stripes *= parity;
256 */
257
258 /* first/last seg is split */
259 num_raid_units += layout->group_width;
260 sgs_per_dev = div_u64(num_raid_units, data_devs);
261 } else {
262 /* For Writes add parity pages array. */
263 max_par_pages = num_raid_units * pages_in_unit *
264 sizeof(struct page *);
265 }
266 }
267
268 ret = _ore_get_io_state(layout, oc, numdevs, sgs_per_dev, max_par_pages,
269 pios);
Boaz Harroshb916c5c2011-09-28 11:55:51 +0300270 if (unlikely(ret))
271 return ret;
272
273 ios = *pios;
274 ios->reading = is_reading;
275 ios->offset = offset;
276
277 if (length) {
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200278 ore_calc_stripe_info(layout, offset, length, &ios->si);
279 ios->length = ios->si.length;
Boaz Harroshb916c5c2011-09-28 11:55:51 +0300280 ios->nr_pages = (ios->length + PAGE_SIZE - 1) / PAGE_SIZE;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200281 if (layout->parity)
282 _ore_post_alloc_raid_stuff(ios);
Boaz Harroshb916c5c2011-09-28 11:55:51 +0300283 }
284
285 return 0;
286}
Boaz Harroshcf283ad2011-08-06 19:22:06 -0700287EXPORT_SYMBOL(ore_get_rw_state);
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200288
Boaz Harroshb916c5c2011-09-28 11:55:51 +0300289/* Allocate an io_state for all the devices in the comps array
290 *
291 * This version of io_state allocation is used mostly by create/remove
292 * and trunc where we currently need all the devices. The only wastful
293 * bit is the read/write_attributes with no IO. Those sites should
294 * be converted to use ore_get_rw_state() with length=0
295 */
Boaz Harrosh5bf696d2011-09-28 11:39:59 +0300296int ore_get_io_state(struct ore_layout *layout, struct ore_components *oc,
Boaz Harroshb916c5c2011-09-28 11:55:51 +0300297 struct ore_io_state **pios)
Boaz Harroshe1042ba2010-11-16 20:09:58 +0200298{
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200299 return _ore_get_io_state(layout, oc, oc->numdevs, 0, 0, pios);
Boaz Harroshe1042ba2010-11-16 20:09:58 +0200300}
Boaz Harroshcf283ad2011-08-06 19:22:06 -0700301EXPORT_SYMBOL(ore_get_io_state);
Boaz Harroshe1042ba2010-11-16 20:09:58 +0200302
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700303void ore_put_io_state(struct ore_io_state *ios)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200304{
305 if (ios) {
306 unsigned i;
307
308 for (i = 0; i < ios->numdevs; i++) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700309 struct ore_per_dev_state *per_dev = &ios->per_dev[i];
Boaz Harrosh06886a52009-11-08 14:54:08 +0200310
311 if (per_dev->or)
312 osd_end_request(per_dev->or);
313 if (per_dev->bio)
314 bio_put(per_dev->bio);
315 }
316
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200317 _ore_free_raid_stuff(ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200318 kfree(ios);
319 }
320}
Boaz Harroshcf283ad2011-08-06 19:22:06 -0700321EXPORT_SYMBOL(ore_put_io_state);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200322
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700323static void _sync_done(struct ore_io_state *ios, void *p)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200324{
325 struct completion *waiting = p;
326
327 complete(waiting);
328}
329
330static void _last_io(struct kref *kref)
331{
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700332 struct ore_io_state *ios = container_of(
333 kref, struct ore_io_state, kref);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200334
335 ios->done(ios, ios->private);
336}
337
338static void _done_io(struct osd_request *or, void *p)
339{
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700340 struct ore_io_state *ios = p;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200341
342 kref_put(&ios->kref, _last_io);
343}
344
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200345int ore_io_execute(struct ore_io_state *ios)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200346{
347 DECLARE_COMPLETION_ONSTACK(wait);
348 bool sync = (ios->done == NULL);
349 int i, ret;
350
351 if (sync) {
352 ios->done = _sync_done;
353 ios->private = &wait;
354 }
355
356 for (i = 0; i < ios->numdevs; i++) {
357 struct osd_request *or = ios->per_dev[i].or;
358 if (unlikely(!or))
359 continue;
360
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700361 ret = osd_finalize_request(or, 0, _ios_cred(ios, i), NULL);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200362 if (unlikely(ret)) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700363 ORE_DBGMSG("Failed to osd_finalize_request() => %d\n",
Boaz Harrosh06886a52009-11-08 14:54:08 +0200364 ret);
365 return ret;
366 }
367 }
368
369 kref_init(&ios->kref);
370
371 for (i = 0; i < ios->numdevs; i++) {
372 struct osd_request *or = ios->per_dev[i].or;
373 if (unlikely(!or))
374 continue;
375
376 kref_get(&ios->kref);
377 osd_execute_request_async(or, _done_io, ios);
378 }
379
380 kref_put(&ios->kref, _last_io);
381 ret = 0;
382
383 if (sync) {
384 wait_for_completion(&wait);
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700385 ret = ore_check_io(ios, NULL);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200386 }
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200387 return ret;
388}
389
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200390static void _clear_bio(struct bio *bio)
391{
392 struct bio_vec *bv;
393 unsigned i;
394
395 __bio_for_each_segment(bv, bio, i, 0) {
396 unsigned this_count = bv->bv_len;
397
398 if (likely(PAGE_SIZE == this_count))
399 clear_highpage(bv->bv_page);
400 else
401 zero_user(bv->bv_page, bv->bv_offset, this_count);
402 }
403}
404
Boaz Harrosh4b46c9f2011-09-28 13:25:50 +0300405int ore_check_io(struct ore_io_state *ios, ore_on_dev_error on_dev_error)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200406{
407 enum osd_err_priority acumulated_osd_err = 0;
408 int acumulated_lin_err = 0;
409 int i;
410
411 for (i = 0; i < ios->numdevs; i++) {
412 struct osd_sense_info osi;
Boaz Harrosh4b46c9f2011-09-28 13:25:50 +0300413 struct ore_per_dev_state *per_dev = &ios->per_dev[i];
414 struct osd_request *or = per_dev->or;
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200415 int ret;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200416
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200417 if (unlikely(!or))
418 continue;
419
420 ret = osd_req_decode_sense(or, &osi);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200421 if (likely(!ret))
422 continue;
423
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200424 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
425 /* start read offset passed endof file */
Boaz Harrosh4b46c9f2011-09-28 13:25:50 +0300426 _clear_bio(per_dev->bio);
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700427 ORE_DBGMSG("start read offset passed end of file "
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200428 "offset=0x%llx, length=0x%llx\n",
Boaz Harrosh4b46c9f2011-09-28 13:25:50 +0300429 _LLU(per_dev->offset),
430 _LLU(per_dev->length));
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200431
432 continue; /* we recovered */
Boaz Harrosh06886a52009-11-08 14:54:08 +0200433 }
434
Boaz Harrosh4b46c9f2011-09-28 13:25:50 +0300435 if (on_dev_error) {
436 u64 residual = ios->reading ?
437 or->in.residual : or->out.residual;
438 u64 offset = (ios->offset + ios->length) - residual;
439 struct ore_dev *od = ios->oc->ods[
440 per_dev->dev - ios->oc->first_dev];
441
442 on_dev_error(ios, od, per_dev->dev, osi.osd_err_pri,
443 offset, residual);
444 }
Boaz Harrosh06886a52009-11-08 14:54:08 +0200445 if (osi.osd_err_pri >= acumulated_osd_err) {
446 acumulated_osd_err = osi.osd_err_pri;
447 acumulated_lin_err = ret;
448 }
449 }
450
Boaz Harrosh06886a52009-11-08 14:54:08 +0200451 return acumulated_lin_err;
452}
Boaz Harroshcf283ad2011-08-06 19:22:06 -0700453EXPORT_SYMBOL(ore_check_io);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200454
Boaz Harroshb367e782010-02-07 19:18:58 +0200455/*
456 * L - logical offset into the file
457 *
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200458 * D - number of Data devices
459 * D = group_width - parity
Boaz Harroshb367e782010-02-07 19:18:58 +0200460 *
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200461 * U - The number of bytes in a stripe within a group
462 * U = stripe_unit * D
Boaz Harroshb367e782010-02-07 19:18:58 +0200463 *
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200464 * T - The number of bytes striped within a group of component objects
465 * (before advancing to the next group)
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200466 * T = U * group_depth
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200467 *
468 * S - The number of bytes striped across all component objects
469 * before the pattern repeats
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200470 * S = T * group_count
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200471 *
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200472 * M - The "major" (i.e., across all components) cycle number
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200473 * M = L / S
474 *
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200475 * G - Counts the groups from the beginning of the major cycle
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200476 * G = (L - (M * S)) / T [or (L % S) / T]
477 *
478 * H - The byte offset within the group
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200479 * H = (L - (M * S)) % T [or (L % S) % T]
480 *
481 * N - The "minor" (i.e., across the group) stripe number
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200482 * N = H / U
Boaz Harroshb367e782010-02-07 19:18:58 +0200483 *
484 * C - The component index coresponding to L
485 *
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200486 * C = (H - (N * U)) / stripe_unit + G * D
487 * [or (L % U) / stripe_unit + G * D]
Boaz Harroshb367e782010-02-07 19:18:58 +0200488 *
489 * O - The component offset coresponding to L
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200490 * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200491 *
492 * LCMdP – Parity cycle: Lowest Common Multiple of group_width, parity
493 * divide by parity
494 * LCMdP = lcm(group_width, parity) / parity
495 *
496 * R - The parity Rotation stripe
497 * (Note parity cycle always starts at a group's boundary)
498 * R = N % LCMdP
499 *
500 * I = the first parity device index
501 * I = (group_width + group_width - R*parity - parity) % group_width
502 *
503 * Craid - The component index Rotated
504 * Craid = (group_width + C - R*parity) % group_width
505 * (We add the group_width to avoid negative numbers modulo math)
Boaz Harroshb367e782010-02-07 19:18:58 +0200506 */
Boaz Harrosh611d7a52011-10-04 14:20:17 +0200507void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset,
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200508 u64 length, struct ore_striping_info *si)
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200509{
Boaz Harrosh16f75bb2011-08-03 20:44:16 -0700510 u32 stripe_unit = layout->stripe_unit;
511 u32 group_width = layout->group_width;
512 u64 group_depth = layout->group_depth;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200513 u32 parity = layout->parity;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200514
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200515 u32 D = group_width - parity;
516 u32 U = D * stripe_unit;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200517 u64 T = U * group_depth;
Boaz Harrosh16f75bb2011-08-03 20:44:16 -0700518 u64 S = T * layout->group_count;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200519 u64 M = div64_u64(file_offset, S);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200520
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200521 /*
522 G = (L - (M * S)) / T
523 H = (L - (M * S)) % T
524 */
525 u64 LmodS = file_offset - M * S;
526 u32 G = div64_u64(LmodS, T);
527 u64 H = LmodS - G * T;
Boaz Harroshb367e782010-02-07 19:18:58 +0200528
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200529 u32 N = div_u64(H, U);
530
531 /* "H - (N * U)" is just "H % U" so it's bound to u32 */
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200532 u32 C = (u32)(H - (N * U)) / stripe_unit + G * group_width;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200533
534 div_u64_rem(file_offset, stripe_unit, &si->unit_off);
535
536 si->obj_offset = si->unit_off + (N * stripe_unit) +
537 (M * group_depth * stripe_unit);
538
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200539 if (parity) {
540 u32 LCMdP = lcm(group_width, parity) / parity;
541 /* R = N % LCMdP; */
542 u32 RxP = (N % LCMdP) * parity;
543 u32 first_dev = C - C % group_width;
544
545 si->par_dev = (group_width + group_width - parity - RxP) %
546 group_width + first_dev;
547 si->dev = (group_width + C - RxP) % group_width + first_dev;
548 si->bytes_in_stripe = U;
549 si->first_stripe_start = M * S + G * T + N * U;
550 } else {
551 /* Make the math correct see _prepare_one_group */
552 si->par_dev = group_width;
553 si->dev = C;
554 }
555
556 si->dev *= layout->mirrors_p1;
557 si->par_dev *= layout->mirrors_p1;
558 si->offset = file_offset;
559 si->length = T - H;
560 if (si->length > length)
561 si->length = length;
Boaz Harrosh16f75bb2011-08-03 20:44:16 -0700562 si->M = M;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200563}
Boaz Harrosh611d7a52011-10-04 14:20:17 +0200564EXPORT_SYMBOL(ore_calc_stripe_info);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200565
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200566int _ore_add_stripe_unit(struct ore_io_state *ios, unsigned *cur_pg,
567 unsigned pgbase, struct page **pages,
568 struct ore_per_dev_state *per_dev, int cur_len)
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200569{
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200570 unsigned pg = *cur_pg;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200571 struct request_queue *q =
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700572 osd_request_queue(_ios_od(ios, per_dev->dev));
Boaz Harroshbbf9a312011-08-26 21:04:52 -0700573 unsigned len = cur_len;
574 int ret;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200575
576 if (per_dev->bio == NULL) {
577 unsigned pages_in_stripe = ios->layout->group_width *
578 (ios->layout->stripe_unit / PAGE_SIZE);
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200579 unsigned nr_pages = ios->nr_pages * ios->layout->group_width /
580 (ios->layout->group_width -
581 ios->layout->parity);
582 unsigned bio_size = (nr_pages + pages_in_stripe) /
583 ios->layout->group_width;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200584
585 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
586 if (unlikely(!per_dev->bio)) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700587 ORE_DBGMSG("Failed to allocate BIO size=%u\n",
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200588 bio_size);
Boaz Harroshbbf9a312011-08-26 21:04:52 -0700589 ret = -ENOMEM;
590 goto out;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200591 }
592 }
593
594 while (cur_len > 0) {
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200595 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
596 unsigned added_len;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200597
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200598 cur_len -= pglen;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200599
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200600 added_len = bio_add_pc_page(q, per_dev->bio, pages[pg],
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200601 pglen, pgbase);
Boaz Harroshbbf9a312011-08-26 21:04:52 -0700602 if (unlikely(pglen != added_len)) {
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200603 ORE_DBGMSG("Failed bio_add_pc_page bi_vcnt=%u\n",
604 per_dev->bio->bi_vcnt);
Boaz Harroshbbf9a312011-08-26 21:04:52 -0700605 ret = -ENOMEM;
606 goto out;
607 }
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200608 _add_stripe_page(ios->sp2d, &ios->si, pages[pg]);
609
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200610 pgbase = 0;
611 ++pg;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200612 }
613 BUG_ON(cur_len);
614
Boaz Harroshbbf9a312011-08-26 21:04:52 -0700615 per_dev->length += len;
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200616 *cur_pg = pg;
Boaz Harroshbbf9a312011-08-26 21:04:52 -0700617 ret = 0;
618out: /* we fail the complete unit on an error eg don't advance
619 * per_dev->length and cur_pg. This means that we might have a bigger
620 * bio than the CDB requested length (per_dev->length). That's fine
621 * only the oposite is fatal.
622 */
623 return ret;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200624}
625
Boaz Harrosh98260752011-10-02 15:32:50 +0200626static int _prepare_for_striping(struct ore_io_state *ios)
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200627{
Boaz Harrosh98260752011-10-02 15:32:50 +0200628 struct ore_striping_info *si = &ios->si;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200629 unsigned stripe_unit = ios->layout->stripe_unit;
Boaz Harroshb367e782010-02-07 19:18:58 +0200630 unsigned mirrors_p1 = ios->layout->mirrors_p1;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200631 unsigned group_width = ios->layout->group_width;
632 unsigned devs_in_group = group_width * mirrors_p1;
Boaz Harroshb367e782010-02-07 19:18:58 +0200633 unsigned dev = si->dev;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200634 unsigned first_dev = dev - (dev % devs_in_group);
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200635 unsigned dev_order;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200636 unsigned cur_pg = ios->pages_consumed;
Boaz Harrosh98260752011-10-02 15:32:50 +0200637 u64 length = ios->length;
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200638 int ret = 0;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200639
Boaz Harrosh98260752011-10-02 15:32:50 +0200640 if (!ios->pages) {
Boaz Harrosh98260752011-10-02 15:32:50 +0200641 ios->numdevs = ios->layout->mirrors_p1;
642 return 0;
643 }
644
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200645 BUG_ON(length > si->length);
646
647 dev_order = _dev_order(devs_in_group, mirrors_p1, si->par_dev, dev);
648 si->cur_comp = dev_order;
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200649 si->cur_pg = si->unit_off / PAGE_SIZE;
Boaz Harrosh98260752011-10-02 15:32:50 +0200650
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200651 while (length) {
Boaz Harroshb916c5c2011-09-28 11:55:51 +0300652 unsigned comp = dev - first_dev;
653 struct ore_per_dev_state *per_dev = &ios->per_dev[comp];
Boaz Harroshb367e782010-02-07 19:18:58 +0200654 unsigned cur_len, page_off = 0;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200655
656 if (!per_dev->length) {
Boaz Harroshb367e782010-02-07 19:18:58 +0200657 per_dev->dev = dev;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200658 if (dev == si->dev) {
659 WARN_ON(dev == si->par_dev);
Boaz Harroshb367e782010-02-07 19:18:58 +0200660 per_dev->offset = si->obj_offset;
661 cur_len = stripe_unit - si->unit_off;
662 page_off = si->unit_off & ~PAGE_MASK;
663 BUG_ON(page_off && (page_off != ios->pgbase));
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200664 } else {
665 if (si->cur_comp > dev_order)
666 per_dev->offset =
667 si->obj_offset - si->unit_off;
668 else /* si->cur_comp < dev_order */
669 per_dev->offset =
670 si->obj_offset + stripe_unit -
671 si->unit_off;
Boaz Harroshb367e782010-02-07 19:18:58 +0200672 cur_len = stripe_unit;
673 }
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200674 } else {
Boaz Harroshb367e782010-02-07 19:18:58 +0200675 cur_len = stripe_unit;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200676 }
Boaz Harroshb367e782010-02-07 19:18:58 +0200677 if (cur_len >= length)
678 cur_len = length;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200679
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200680 ret = _ore_add_stripe_unit(ios, &cur_pg, page_off, ios->pages,
681 per_dev, cur_len);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200682 if (unlikely(ret))
683 goto out;
684
Boaz Harrosh6e316092010-07-29 17:08:13 +0300685 dev += mirrors_p1;
686 dev = (dev % devs_in_group) + first_dev;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200687
688 length -= cur_len;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200689
690 si->cur_comp = (si->cur_comp + 1) % group_width;
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200691 if (unlikely((dev == si->par_dev) || (!length && ios->sp2d))) {
692 if (!length && ios->sp2d) {
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200693 /* If we are writing and this is the very last
694 * stripe. then operate on parity dev.
695 */
696 dev = si->par_dev;
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200697 }
698 if (ios->sp2d)
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200699 /* In writes cur_len just means if it's the
700 * last one. See _ore_add_parity_unit.
701 */
702 cur_len = length;
703 per_dev = &ios->per_dev[dev - first_dev];
704 if (!per_dev->length) {
705 /* Only/always the parity unit of the first
706 * stripe will be empty. So this is a chance to
707 * initialize the per_dev info.
708 */
709 per_dev->dev = dev;
710 per_dev->offset = si->obj_offset - si->unit_off;
711 }
712
713 ret = _ore_add_parity_unit(ios, si, per_dev, cur_len);
714 if (unlikely(ret))
715 goto out;
716
717 /* Rotate next par_dev backwards with wraping */
718 si->par_dev = (devs_in_group + si->par_dev -
719 ios->layout->parity * mirrors_p1) %
720 devs_in_group + first_dev;
721 /* Next stripe, start fresh */
722 si->cur_comp = 0;
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200723 si->cur_pg = 0;
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200724 }
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200725 }
726out:
Boaz Harroshb916c5c2011-09-28 11:55:51 +0300727 ios->numdevs = devs_in_group;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200728 ios->pages_consumed = cur_pg;
Boaz Harroshbbf9a312011-08-26 21:04:52 -0700729 if (unlikely(ret)) {
730 if (length == ios->length)
731 return ret;
732 else
733 ios->length -= length;
734 }
735 return 0;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200736}
737
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700738int ore_create(struct ore_io_state *ios)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200739{
740 int i, ret;
741
Boaz Harrosh5bf696d2011-09-28 11:39:59 +0300742 for (i = 0; i < ios->oc->numdevs; i++) {
Boaz Harrosh06886a52009-11-08 14:54:08 +0200743 struct osd_request *or;
744
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700745 or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200746 if (unlikely(!or)) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700747 ORE_ERR("%s: osd_start_request failed\n", __func__);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200748 ret = -ENOMEM;
749 goto out;
750 }
751 ios->per_dev[i].or = or;
752 ios->numdevs++;
753
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700754 osd_req_create_object(or, _ios_obj(ios, i));
Boaz Harrosh06886a52009-11-08 14:54:08 +0200755 }
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700756 ret = ore_io_execute(ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200757
758out:
759 return ret;
760}
Boaz Harroshcf283ad2011-08-06 19:22:06 -0700761EXPORT_SYMBOL(ore_create);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200762
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700763int ore_remove(struct ore_io_state *ios)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200764{
765 int i, ret;
766
Boaz Harrosh5bf696d2011-09-28 11:39:59 +0300767 for (i = 0; i < ios->oc->numdevs; i++) {
Boaz Harrosh06886a52009-11-08 14:54:08 +0200768 struct osd_request *or;
769
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700770 or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200771 if (unlikely(!or)) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700772 ORE_ERR("%s: osd_start_request failed\n", __func__);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200773 ret = -ENOMEM;
774 goto out;
775 }
776 ios->per_dev[i].or = or;
777 ios->numdevs++;
778
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700779 osd_req_remove_object(or, _ios_obj(ios, i));
Boaz Harrosh06886a52009-11-08 14:54:08 +0200780 }
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700781 ret = ore_io_execute(ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200782
783out:
784 return ret;
785}
Boaz Harroshcf283ad2011-08-06 19:22:06 -0700786EXPORT_SYMBOL(ore_remove);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200787
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700788static int _write_mirror(struct ore_io_state *ios, int cur_comp)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200789{
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700790 struct ore_per_dev_state *master_dev = &ios->per_dev[cur_comp];
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200791 unsigned dev = ios->per_dev[cur_comp].dev;
792 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
793 int ret = 0;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200794
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200795 if (ios->pages && !master_dev->length)
796 return 0; /* Just an empty slot */
797
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200798 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700799 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
Boaz Harrosh06886a52009-11-08 14:54:08 +0200800 struct osd_request *or;
801
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700802 or = osd_start_request(_ios_od(ios, dev), GFP_KERNEL);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200803 if (unlikely(!or)) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700804 ORE_ERR("%s: osd_start_request failed\n", __func__);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200805 ret = -ENOMEM;
806 goto out;
807 }
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200808 per_dev->or = or;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200809
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200810 if (ios->pages) {
Boaz Harrosh06886a52009-11-08 14:54:08 +0200811 struct bio *bio;
812
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200813 if (per_dev != master_dev) {
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200814 bio = bio_kmalloc(GFP_KERNEL,
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200815 master_dev->bio->bi_max_vecs);
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200816 if (unlikely(!bio)) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700817 ORE_DBGMSG(
Paul Bolle426d3102010-08-07 12:30:03 +0200818 "Failed to allocate BIO size=%u\n",
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200819 master_dev->bio->bi_max_vecs);
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200820 ret = -ENOMEM;
821 goto out;
822 }
823
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200824 __bio_clone(bio, master_dev->bio);
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200825 bio->bi_bdev = NULL;
826 bio->bi_next = NULL;
Boaz Harrosh6851a5e2011-08-24 18:10:49 -0700827 per_dev->offset = master_dev->offset;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200828 per_dev->length = master_dev->length;
829 per_dev->bio = bio;
830 per_dev->dev = dev;
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200831 } else {
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200832 bio = master_dev->bio;
833 /* FIXME: bio_set_dir() */
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200834 bio->bi_rw |= REQ_WRITE;
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200835 }
Boaz Harrosh06886a52009-11-08 14:54:08 +0200836
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700837 osd_req_write(or, _ios_obj(ios, dev), per_dev->offset,
838 bio, per_dev->length);
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700839 ORE_DBGMSG("write(0x%llx) offset=0x%llx "
Boaz Harrosh34ce4e7c2009-12-15 19:34:17 +0200840 "length=0x%llx dev=%d\n",
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700841 _LLU(_ios_obj(ios, dev)->id),
842 _LLU(per_dev->offset),
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200843 _LLU(per_dev->length), dev);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200844 } else if (ios->kern_buff) {
Boaz Harrosh6851a5e2011-08-24 18:10:49 -0700845 per_dev->offset = ios->si.obj_offset;
846 per_dev->dev = ios->si.dev + dev;
847
848 /* no cross device without page array */
849 BUG_ON((ios->layout->group_width > 1) &&
850 (ios->si.unit_off + ios->length >
851 ios->layout->stripe_unit));
852
853 ret = osd_req_write_kern(or, _ios_obj(ios, per_dev->dev),
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700854 per_dev->offset,
855 ios->kern_buff, ios->length);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200856 if (unlikely(ret))
857 goto out;
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700858 ORE_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
Boaz Harrosh34ce4e7c2009-12-15 19:34:17 +0200859 "length=0x%llx dev=%d\n",
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700860 _LLU(_ios_obj(ios, dev)->id),
861 _LLU(per_dev->offset),
Boaz Harrosh6851a5e2011-08-24 18:10:49 -0700862 _LLU(ios->length), per_dev->dev);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200863 } else {
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700864 osd_req_set_attributes(or, _ios_obj(ios, dev));
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700865 ORE_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700866 _LLU(_ios_obj(ios, dev)->id),
867 ios->out_attr_len, dev);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200868 }
869
870 if (ios->out_attr)
871 osd_req_add_set_attr_list(or, ios->out_attr,
872 ios->out_attr_len);
873
874 if (ios->in_attr)
875 osd_req_add_get_attr_list(or, ios->in_attr,
876 ios->in_attr_len);
877 }
Boaz Harrosh06886a52009-11-08 14:54:08 +0200878
879out:
880 return ret;
881}
882
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700883int ore_write(struct ore_io_state *ios)
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200884{
885 int i;
886 int ret;
887
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200888 if (unlikely(ios->sp2d && !ios->r4w)) {
889 /* A library is attempting a RAID-write without providing
890 * a pages lock interface.
891 */
892 WARN_ON_ONCE(1);
893 return -ENOTSUPP;
894 }
895
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200896 ret = _prepare_for_striping(ios);
897 if (unlikely(ret))
898 return ret;
899
900 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700901 ret = _write_mirror(ios, i);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200902 if (unlikely(ret))
903 return ret;
904 }
905
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700906 ret = ore_io_execute(ios);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200907 return ret;
908}
Boaz Harroshcf283ad2011-08-06 19:22:06 -0700909EXPORT_SYMBOL(ore_write);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200910
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200911int _ore_read_mirror(struct ore_io_state *ios, unsigned cur_comp)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200912{
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200913 struct osd_request *or;
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700914 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700915 struct osd_obj_id *obj = _ios_obj(ios, cur_comp);
916 unsigned first_dev = (unsigned)obj->id;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200917
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200918 if (ios->pages && !per_dev->length)
919 return 0; /* Just an empty slot */
920
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200921 first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700922 or = osd_start_request(_ios_od(ios, first_dev), GFP_KERNEL);
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200923 if (unlikely(!or)) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700924 ORE_ERR("%s: osd_start_request failed\n", __func__);
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200925 return -ENOMEM;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200926 }
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200927 per_dev->or = or;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200928
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200929 if (ios->pages) {
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200930 if (per_dev->cur_sg) {
931 /* finalize the last sg_entry */
932 _ore_add_sg_seg(per_dev, 0, false);
933 if (unlikely(!per_dev->cur_sg))
934 return 0; /* Skip parity only device */
935
936 osd_req_read_sg(or, obj, per_dev->bio,
937 per_dev->sglist, per_dev->cur_sg);
938 } else {
939 /* The no raid case */
940 osd_req_read(or, obj, per_dev->offset,
941 per_dev->bio, per_dev->length);
942 }
943
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700944 ORE_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200945 " dev=%d sg_len=%d\n", _LLU(obj->id),
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200946 _LLU(per_dev->offset), _LLU(per_dev->length),
Boaz Harrosha1fec1d2011-10-12 18:42:22 +0200947 first_dev, per_dev->cur_sg);
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200948 } else {
Boaz Harrosh6851a5e2011-08-24 18:10:49 -0700949 BUG_ON(ios->kern_buff);
950
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700951 osd_req_get_attributes(or, obj);
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700952 ORE_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
Boaz Harrosh9e9db452011-08-05 15:06:04 -0700953 _LLU(obj->id),
954 ios->in_attr_len, first_dev);
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200955 }
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200956 if (ios->out_attr)
957 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
958
959 if (ios->in_attr)
960 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
961
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200962 return 0;
963}
964
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700965int ore_read(struct ore_io_state *ios)
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200966{
967 int i;
968 int ret;
969
970 ret = _prepare_for_striping(ios);
971 if (unlikely(ret))
972 return ret;
973
974 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
Boaz Harrosh769ba8d2011-10-14 15:33:51 +0200975 ret = _ore_read_mirror(ios, i);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200976 if (unlikely(ret))
977 return ret;
978 }
979
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700980 ret = ore_io_execute(ios);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200981 return ret;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200982}
Boaz Harroshcf283ad2011-08-06 19:22:06 -0700983EXPORT_SYMBOL(ore_read);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200984
Boaz Harrosh8ff660a2011-08-06 19:26:31 -0700985int extract_attr_from_ios(struct ore_io_state *ios, struct osd_attr *attr)
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200986{
987 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
988 void *iter = NULL;
989 int nelem;
990
991 do {
992 nelem = 1;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200993 osd_req_decode_get_attr_list(ios->per_dev[0].or,
994 &cur_attr, &nelem, &iter);
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200995 if ((cur_attr.attr_page == attr->attr_page) &&
996 (cur_attr.attr_id == attr->attr_id)) {
997 attr->len = cur_attr.len;
998 attr->val_ptr = cur_attr.val_ptr;
999 return 0;
1000 }
1001 } while (iter);
1002
1003 return -EIO;
1004}
Boaz Harroshcf283ad2011-08-06 19:22:06 -07001005EXPORT_SYMBOL(extract_attr_from_ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +02001006
Boaz Harrosh8ff660a2011-08-06 19:26:31 -07001007static int _truncate_mirrors(struct ore_io_state *ios, unsigned cur_comp,
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001008 struct osd_attr *attr)
1009{
1010 int last_comp = cur_comp + ios->layout->mirrors_p1;
1011
1012 for (; cur_comp < last_comp; ++cur_comp) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -07001013 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001014 struct osd_request *or;
1015
Boaz Harrosh9e9db452011-08-05 15:06:04 -07001016 or = osd_start_request(_ios_od(ios, cur_comp), GFP_KERNEL);
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001017 if (unlikely(!or)) {
Boaz Harrosh8ff660a2011-08-06 19:26:31 -07001018 ORE_ERR("%s: osd_start_request failed\n", __func__);
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001019 return -ENOMEM;
1020 }
1021 per_dev->or = or;
1022
Boaz Harrosh9e9db452011-08-05 15:06:04 -07001023 osd_req_set_attributes(or, _ios_obj(ios, cur_comp));
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001024 osd_req_add_set_attr_list(or, attr, 1);
1025 }
1026
1027 return 0;
1028}
1029
Boaz Harrosh16f75bb2011-08-03 20:44:16 -07001030struct _trunc_info {
Boaz Harrosheb507bc2011-08-10 14:17:28 -07001031 struct ore_striping_info si;
Boaz Harrosh16f75bb2011-08-03 20:44:16 -07001032 u64 prev_group_obj_off;
1033 u64 next_group_obj_off;
1034
1035 unsigned first_group_dev;
1036 unsigned nex_group_dev;
Boaz Harrosh16f75bb2011-08-03 20:44:16 -07001037};
1038
H Hartley Sweeten1958c7c22011-09-23 13:42:43 -07001039static void _calc_trunk_info(struct ore_layout *layout, u64 file_offset,
1040 struct _trunc_info *ti)
Boaz Harrosh16f75bb2011-08-03 20:44:16 -07001041{
1042 unsigned stripe_unit = layout->stripe_unit;
1043
Boaz Harrosha1fec1d2011-10-12 18:42:22 +02001044 ore_calc_stripe_info(layout, file_offset, 0, &ti->si);
Boaz Harrosh16f75bb2011-08-03 20:44:16 -07001045
1046 ti->prev_group_obj_off = ti->si.M * stripe_unit;
1047 ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
1048
1049 ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
1050 ti->nex_group_dev = ti->first_group_dev + layout->group_width;
Boaz Harrosh16f75bb2011-08-03 20:44:16 -07001051}
1052
Boaz Harrosh5bf696d2011-09-28 11:39:59 +03001053int ore_truncate(struct ore_layout *layout, struct ore_components *oc,
Boaz Harrosh9e9db452011-08-05 15:06:04 -07001054 u64 size)
Boaz Harrosh06886a52009-11-08 14:54:08 +02001055{
Boaz Harrosh8ff660a2011-08-06 19:26:31 -07001056 struct ore_io_state *ios;
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001057 struct exofs_trunc_attr {
1058 struct osd_attr attr;
1059 __be64 newsize;
1060 } *size_attrs;
Boaz Harrosh16f75bb2011-08-03 20:44:16 -07001061 struct _trunc_info ti;
Boaz Harrosh06886a52009-11-08 14:54:08 +02001062 int i, ret;
1063
Boaz Harrosh5bf696d2011-09-28 11:39:59 +03001064 ret = ore_get_io_state(layout, oc, &ios);
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001065 if (unlikely(ret))
1066 return ret;
1067
Boaz Harrosh16f75bb2011-08-03 20:44:16 -07001068 _calc_trunk_info(ios->layout, size, &ti);
1069
Boaz Harroshb916c5c2011-09-28 11:55:51 +03001070 size_attrs = kcalloc(ios->oc->numdevs, sizeof(*size_attrs),
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001071 GFP_KERNEL);
1072 if (unlikely(!size_attrs)) {
1073 ret = -ENOMEM;
1074 goto out;
1075 }
Boaz Harrosh06886a52009-11-08 14:54:08 +02001076
Boaz Harrosh5bf696d2011-09-28 11:39:59 +03001077 ios->numdevs = ios->oc->numdevs;
Boaz Harrosh06886a52009-11-08 14:54:08 +02001078
Boaz Harroshb916c5c2011-09-28 11:55:51 +03001079 for (i = 0; i < ios->numdevs; ++i) {
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001080 struct exofs_trunc_attr *size_attr = &size_attrs[i];
1081 u64 obj_size;
Boaz Harrosh06886a52009-11-08 14:54:08 +02001082
Boaz Harrosh16f75bb2011-08-03 20:44:16 -07001083 if (i < ti.first_group_dev)
1084 obj_size = ti.prev_group_obj_off;
1085 else if (i >= ti.nex_group_dev)
1086 obj_size = ti.next_group_obj_off;
1087 else if (i < ti.si.dev) /* dev within this group */
1088 obj_size = ti.si.obj_offset +
1089 ios->layout->stripe_unit - ti.si.unit_off;
1090 else if (i == ti.si.dev)
1091 obj_size = ti.si.obj_offset;
1092 else /* i > ti.dev */
1093 obj_size = ti.si.obj_offset - ti.si.unit_off;
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001094
1095 size_attr->newsize = cpu_to_be64(obj_size);
1096 size_attr->attr = g_attr_logical_length;
1097 size_attr->attr.val_ptr = &size_attr->newsize;
1098
Boaz Harrosh8ff660a2011-08-06 19:26:31 -07001099 ORE_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
Boaz Harrosh5bf696d2011-09-28 11:39:59 +03001100 _LLU(oc->comps->obj.id), _LLU(obj_size), i);
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001101 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
1102 &size_attr->attr);
1103 if (unlikely(ret))
Boaz Harrosh06886a52009-11-08 14:54:08 +02001104 goto out;
Boaz Harrosh06886a52009-11-08 14:54:08 +02001105 }
Boaz Harrosh8ff660a2011-08-06 19:26:31 -07001106 ret = ore_io_execute(ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +02001107
1108out:
Boaz Harrosh5d952b82010-02-01 13:35:51 +02001109 kfree(size_attrs);
Boaz Harrosh8ff660a2011-08-06 19:26:31 -07001110 ore_put_io_state(ios);
Boaz Harrosh06886a52009-11-08 14:54:08 +02001111 return ret;
1112}
Boaz Harroshcf283ad2011-08-06 19:22:06 -07001113EXPORT_SYMBOL(ore_truncate);
Boaz Harrosh85e44df2011-05-16 15:26:47 +03001114
1115const struct osd_attr g_attr_logical_length = ATTR_DEF(
1116 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
Boaz Harroshcf283ad2011-08-06 19:22:06 -07001117EXPORT_SYMBOL(g_attr_logical_length);