blob: 5293bc411d17735fcfed8b685e7d3ab9d0e80da4 [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
25#include <scsi/scsi_device.h>
Boaz Harrosh5d952b82010-02-01 13:35:51 +020026#include <asm/div64.h>
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020027
28#include "exofs.h"
29
Boaz Harrosh34ce4e72009-12-15 19:34:17 +020030#define EXOFS_DBGMSG2(M...) do {} while (0)
31/* #define EXOFS_DBGMSG2 EXOFS_DBGMSG */
32
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020033void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
34{
35 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
36}
37
Boaz Harrosh06886a52009-11-08 14:54:08 +020038int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
39 u64 offset, void *p, unsigned length)
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020040{
Boaz Harrosh06886a52009-11-08 14:54:08 +020041 struct osd_request *or = osd_start_request(od, GFP_KERNEL);
42/* struct osd_sense_info osi = {.key = 0};*/
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020043 int ret;
44
Boaz Harrosh06886a52009-11-08 14:54:08 +020045 if (unlikely(!or)) {
46 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
47 return -ENOMEM;
48 }
49 ret = osd_req_read_kern(or, obj, offset, p, length);
50 if (unlikely(ret)) {
51 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
52 goto out;
53 }
54
55 ret = osd_finalize_request(or, 0, cred, NULL);
56 if (unlikely(ret)) {
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020057 EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n", ret);
Boaz Harrosh06886a52009-11-08 14:54:08 +020058 goto out;
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020059 }
60
61 ret = osd_execute_request(or);
Boaz Harrosh06886a52009-11-08 14:54:08 +020062 if (unlikely(ret))
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020063 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
64 /* osd_req_decode_sense(or, ret); */
Boaz Harrosh06886a52009-11-08 14:54:08 +020065
66out:
67 osd_end_request(or);
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020068 return ret;
69}
70
Boaz Harrosh45d3abc2010-01-28 11:46:16 +020071int exofs_get_io_state(struct exofs_layout *layout,
72 struct exofs_io_state **pios)
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020073{
Boaz Harrosh06886a52009-11-08 14:54:08 +020074 struct exofs_io_state *ios;
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020075
Boaz Harrosh06886a52009-11-08 14:54:08 +020076 /*TODO: Maybe use kmem_cach per sbi of size
Boaz Harrosh45d3abc2010-01-28 11:46:16 +020077 * exofs_io_state_size(layout->s_numdevs)
Boaz Harrosh06886a52009-11-08 14:54:08 +020078 */
Boaz Harrosh45d3abc2010-01-28 11:46:16 +020079 ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL);
Boaz Harrosh06886a52009-11-08 14:54:08 +020080 if (unlikely(!ios)) {
Boaz Harrosh34ce4e72009-12-15 19:34:17 +020081 EXOFS_DBGMSG("Faild kzalloc bytes=%d\n",
Boaz Harrosh45d3abc2010-01-28 11:46:16 +020082 exofs_io_state_size(layout->s_numdevs));
Boaz Harrosh06886a52009-11-08 14:54:08 +020083 *pios = NULL;
84 return -ENOMEM;
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020085 }
86
Boaz Harrosh45d3abc2010-01-28 11:46:16 +020087 ios->layout = layout;
88 ios->obj.partition = layout->s_pid;
Boaz Harrosh06886a52009-11-08 14:54:08 +020089 *pios = ios;
90 return 0;
91}
Boaz Harroshb14f8ab2008-10-27 18:27:55 +020092
Boaz Harrosh06886a52009-11-08 14:54:08 +020093void exofs_put_io_state(struct exofs_io_state *ios)
94{
95 if (ios) {
96 unsigned i;
97
98 for (i = 0; i < ios->numdevs; i++) {
99 struct exofs_per_dev_state *per_dev = &ios->per_dev[i];
100
101 if (per_dev->or)
102 osd_end_request(per_dev->or);
103 if (per_dev->bio)
104 bio_put(per_dev->bio);
105 }
106
107 kfree(ios);
108 }
109}
110
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200111unsigned exofs_layout_od_id(struct exofs_layout *layout,
112 osd_id obj_no, unsigned layout_index)
113{
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200114/* switch (layout->lay_func) {
115 case LAYOUT_MOVING_WINDOW:
116 {*/
117 unsigned dev_mod = obj_no;
118
119 return (layout_index + dev_mod * layout->mirrors_p1) %
120 layout->s_numdevs;
121/* }
122 case LAYOUT_FUNC_IMPLICT:
123 return layout->devs[layout_index];
124 }*/
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200125}
126
127static inline struct osd_dev *exofs_ios_od(struct exofs_io_state *ios,
128 unsigned layout_index)
129{
130 return ios->layout->s_ods[
131 exofs_layout_od_id(ios->layout, ios->obj.id, layout_index)];
132}
133
Boaz Harrosh06886a52009-11-08 14:54:08 +0200134static void _sync_done(struct exofs_io_state *ios, void *p)
135{
136 struct completion *waiting = p;
137
138 complete(waiting);
139}
140
141static void _last_io(struct kref *kref)
142{
143 struct exofs_io_state *ios = container_of(
144 kref, struct exofs_io_state, kref);
145
146 ios->done(ios, ios->private);
147}
148
149static void _done_io(struct osd_request *or, void *p)
150{
151 struct exofs_io_state *ios = p;
152
153 kref_put(&ios->kref, _last_io);
154}
155
156static int exofs_io_execute(struct exofs_io_state *ios)
157{
158 DECLARE_COMPLETION_ONSTACK(wait);
159 bool sync = (ios->done == NULL);
160 int i, ret;
161
162 if (sync) {
163 ios->done = _sync_done;
164 ios->private = &wait;
165 }
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 ret = osd_finalize_request(or, 0, ios->cred, NULL);
173 if (unlikely(ret)) {
174 EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n",
175 ret);
176 return ret;
177 }
178 }
179
180 kref_init(&ios->kref);
181
182 for (i = 0; i < ios->numdevs; i++) {
183 struct osd_request *or = ios->per_dev[i].or;
184 if (unlikely(!or))
185 continue;
186
187 kref_get(&ios->kref);
188 osd_execute_request_async(or, _done_io, ios);
189 }
190
191 kref_put(&ios->kref, _last_io);
192 ret = 0;
193
194 if (sync) {
195 wait_for_completion(&wait);
196 ret = exofs_check_io(ios, NULL);
197 }
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200198 return ret;
199}
200
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200201static void _clear_bio(struct bio *bio)
202{
203 struct bio_vec *bv;
204 unsigned i;
205
206 __bio_for_each_segment(bv, bio, i, 0) {
207 unsigned this_count = bv->bv_len;
208
209 if (likely(PAGE_SIZE == this_count))
210 clear_highpage(bv->bv_page);
211 else
212 zero_user(bv->bv_page, bv->bv_offset, this_count);
213 }
214}
215
Boaz Harrosh06886a52009-11-08 14:54:08 +0200216int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
217{
218 enum osd_err_priority acumulated_osd_err = 0;
219 int acumulated_lin_err = 0;
220 int i;
221
222 for (i = 0; i < ios->numdevs; i++) {
223 struct osd_sense_info osi;
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200224 struct osd_request *or = ios->per_dev[i].or;
225 int ret;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200226
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200227 if (unlikely(!or))
228 continue;
229
230 ret = osd_req_decode_sense(or, &osi);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200231 if (likely(!ret))
232 continue;
233
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200234 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
235 /* start read offset passed endof file */
236 _clear_bio(ios->per_dev[i].bio);
237 EXOFS_DBGMSG("start read offset passed end of file "
238 "offset=0x%llx, length=0x%llx\n",
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200239 _LLU(ios->per_dev[i].offset),
240 _LLU(ios->per_dev[i].length));
Boaz Harrosh22ddc552010-01-19 19:24:45 +0200241
242 continue; /* we recovered */
Boaz Harrosh06886a52009-11-08 14:54:08 +0200243 }
244
245 if (osi.osd_err_pri >= acumulated_osd_err) {
246 acumulated_osd_err = osi.osd_err_pri;
247 acumulated_lin_err = ret;
248 }
249 }
250
251 /* TODO: raid specific residual calculations */
252 if (resid) {
253 if (likely(!acumulated_lin_err))
254 *resid = 0;
255 else
256 *resid = ios->length;
257 }
258
259 return acumulated_lin_err;
260}
261
Boaz Harroshb367e782010-02-07 19:18:58 +0200262/*
263 * L - logical offset into the file
264 *
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200265 * U - The number of bytes in a stripe within a group
Boaz Harroshb367e782010-02-07 19:18:58 +0200266 *
267 * U = stripe_unit * group_width
268 *
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200269 * T - The number of bytes striped within a group of component objects
270 * (before advancing to the next group)
Boaz Harroshb367e782010-02-07 19:18:58 +0200271 *
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200272 * T = stripe_unit * group_width * group_depth
273 *
274 * S - The number of bytes striped across all component objects
275 * before the pattern repeats
276 *
277 * S = stripe_unit * group_width * group_depth * group_count
278 *
279 * M - The "major" (i.e., across all components) stripe number
280 *
281 * M = L / S
282 *
283 * G - Counts the groups from the beginning of the major stripe
284 *
285 * G = (L - (M * S)) / T [or (L % S) / T]
286 *
287 * H - The byte offset within the group
288 *
289 * H = (L - (M * S)) % T [or (L % S) % T]
290 *
291 * N - The "minor" (i.e., across the group) stripe number
292 *
293 * N = H / U
Boaz Harroshb367e782010-02-07 19:18:58 +0200294 *
295 * C - The component index coresponding to L
296 *
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200297 * C = (H - (N * U)) / stripe_unit + G * group_width
298 * [or (L % U) / stripe_unit + G * group_width]
Boaz Harroshb367e782010-02-07 19:18:58 +0200299 *
300 * O - The component offset coresponding to L
301 *
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200302 * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
Boaz Harroshb367e782010-02-07 19:18:58 +0200303 */
Boaz Harroshb367e782010-02-07 19:18:58 +0200304struct _striping_info {
305 u64 obj_offset;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200306 u64 group_length;
307 u64 total_group_length;
308 u64 Major;
Boaz Harroshb367e782010-02-07 19:18:58 +0200309 unsigned dev;
310 unsigned unit_off;
311};
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200312
Boaz Harroshb367e782010-02-07 19:18:58 +0200313static void _calc_stripe_info(struct exofs_io_state *ios, u64 file_offset,
314 struct _striping_info *si)
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200315{
Boaz Harroshb367e782010-02-07 19:18:58 +0200316 u32 stripe_unit = ios->layout->stripe_unit;
317 u32 group_width = ios->layout->group_width;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200318 u64 group_depth = ios->layout->group_depth;
319
Boaz Harroshb367e782010-02-07 19:18:58 +0200320 u32 U = stripe_unit * group_width;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200321 u64 T = U * group_depth;
322 u64 S = T * ios->layout->group_count;
323 u64 M = div64_u64(file_offset, S);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200324
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200325 /*
326 G = (L - (M * S)) / T
327 H = (L - (M * S)) % T
328 */
329 u64 LmodS = file_offset - M * S;
330 u32 G = div64_u64(LmodS, T);
331 u64 H = LmodS - G * T;
Boaz Harroshb367e782010-02-07 19:18:58 +0200332
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200333 u32 N = div_u64(H, U);
334
335 /* "H - (N * U)" is just "H % U" so it's bound to u32 */
336 si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
Boaz Harroshb367e782010-02-07 19:18:58 +0200337 si->dev *= ios->layout->mirrors_p1;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200338
339 div_u64_rem(file_offset, stripe_unit, &si->unit_off);
340
341 si->obj_offset = si->unit_off + (N * stripe_unit) +
342 (M * group_depth * stripe_unit);
343
344 si->group_length = T - H;
345 si->total_group_length = T;
346 si->Major = M;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200347}
348
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200349static int _add_stripe_unit(struct exofs_io_state *ios, unsigned *cur_pg,
350 unsigned pgbase, struct exofs_per_dev_state *per_dev,
351 int cur_len)
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200352{
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200353 unsigned pg = *cur_pg;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200354 struct request_queue *q =
355 osd_request_queue(exofs_ios_od(ios, per_dev->dev));
356
357 per_dev->length += cur_len;
358
359 if (per_dev->bio == NULL) {
360 unsigned pages_in_stripe = ios->layout->group_width *
361 (ios->layout->stripe_unit / PAGE_SIZE);
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200362 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200363 ios->layout->group_width;
364
365 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
366 if (unlikely(!per_dev->bio)) {
367 EXOFS_DBGMSG("Faild to allocate BIO size=%u\n",
368 bio_size);
369 return -ENOMEM;
370 }
371 }
372
373 while (cur_len > 0) {
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200374 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
375 unsigned added_len;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200376
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200377 BUG_ON(ios->nr_pages <= pg);
378 cur_len -= pglen;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200379
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200380 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
381 pglen, pgbase);
382 if (unlikely(pglen != added_len))
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200383 return -ENOMEM;
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200384 pgbase = 0;
385 ++pg;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200386 }
387 BUG_ON(cur_len);
388
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200389 *cur_pg = pg;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200390 return 0;
391}
392
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200393static int _prepare_one_group(struct exofs_io_state *ios, u64 length,
394 struct _striping_info *si, unsigned first_comp)
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200395{
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200396 unsigned stripe_unit = ios->layout->stripe_unit;
Boaz Harroshb367e782010-02-07 19:18:58 +0200397 unsigned mirrors_p1 = ios->layout->mirrors_p1;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200398 unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
Boaz Harroshb367e782010-02-07 19:18:58 +0200399 unsigned dev = si->dev;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200400 unsigned first_dev = dev - (dev % devs_in_group);
401 unsigned comp = first_comp + (dev - first_dev);
402 unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
403 unsigned cur_pg = ios->pages_consumed;
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200404 int ret = 0;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200405
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200406 while (length) {
407 struct exofs_per_dev_state *per_dev = &ios->per_dev[comp];
Boaz Harroshb367e782010-02-07 19:18:58 +0200408 unsigned cur_len, page_off = 0;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200409
410 if (!per_dev->length) {
Boaz Harroshb367e782010-02-07 19:18:58 +0200411 per_dev->dev = dev;
412 if (dev < si->dev) {
413 per_dev->offset = si->obj_offset + stripe_unit -
414 si->unit_off;
415 cur_len = stripe_unit;
416 } else if (dev == si->dev) {
417 per_dev->offset = si->obj_offset;
418 cur_len = stripe_unit - si->unit_off;
419 page_off = si->unit_off & ~PAGE_MASK;
420 BUG_ON(page_off && (page_off != ios->pgbase));
421 } else { /* dev > si->dev */
422 per_dev->offset = si->obj_offset - si->unit_off;
423 cur_len = stripe_unit;
424 }
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200425
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200426 if (max_comp < comp)
427 max_comp = comp;
Boaz Harroshb367e782010-02-07 19:18:58 +0200428
429 dev += mirrors_p1;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200430 dev = (dev % devs_in_group) + first_dev;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200431 } else {
Boaz Harroshb367e782010-02-07 19:18:58 +0200432 cur_len = stripe_unit;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200433 }
Boaz Harroshb367e782010-02-07 19:18:58 +0200434 if (cur_len >= length)
435 cur_len = length;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200436
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200437 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
438 cur_len);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200439 if (unlikely(ret))
440 goto out;
441
Boaz Harroshb367e782010-02-07 19:18:58 +0200442 comp += mirrors_p1;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200443 comp = (comp % devs_in_group) + first_comp;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200444
445 length -= cur_len;
446 }
447out:
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200448 ios->numdevs = max_comp + mirrors_p1;
449 ios->pages_consumed = cur_pg;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200450 return ret;
451}
452
Boaz Harroshb367e782010-02-07 19:18:58 +0200453static int _prepare_for_striping(struct exofs_io_state *ios)
454{
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200455 u64 length = ios->length;
Boaz Harroshb367e782010-02-07 19:18:58 +0200456 struct _striping_info si;
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200457 unsigned devs_in_group = ios->layout->group_width *
458 ios->layout->mirrors_p1;
459 unsigned first_comp = 0;
460 int ret = 0;
Boaz Harroshb367e782010-02-07 19:18:58 +0200461
462 _calc_stripe_info(ios, ios->offset, &si);
463
464 if (!ios->pages) {
465 if (ios->kern_buff) {
466 struct exofs_per_dev_state *per_dev = &ios->per_dev[0];
467
468 per_dev->offset = si.obj_offset;
469 per_dev->dev = si.dev;
470
471 /* no cross device without page array */
472 BUG_ON((ios->layout->group_width > 1) &&
473 (si.unit_off + ios->length >
474 ios->layout->stripe_unit));
475 }
476 ios->numdevs = ios->layout->mirrors_p1;
477 return 0;
478 }
479
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200480 while (length) {
481 if (length < si.group_length)
482 si.group_length = length;
483
484 ret = _prepare_one_group(ios, si.group_length, &si, first_comp);
485 if (unlikely(ret))
486 goto out;
487
488 length -= si.group_length;
489
490 si.group_length = si.total_group_length;
491 si.unit_off = 0;
492 ++si.Major;
493 si.obj_offset = si.Major * ios->layout->stripe_unit *
494 ios->layout->group_depth;
495
496 si.dev = (si.dev - (si.dev % devs_in_group)) + devs_in_group;
497 si.dev %= ios->layout->s_numdevs;
498
499 first_comp += devs_in_group;
500 first_comp %= ios->layout->s_numdevs;
501 }
502
503out:
504 return ret;
Boaz Harroshb367e782010-02-07 19:18:58 +0200505}
506
Boaz Harrosh06886a52009-11-08 14:54:08 +0200507int exofs_sbi_create(struct exofs_io_state *ios)
508{
509 int i, ret;
510
Boaz Harrosh45d3abc2010-01-28 11:46:16 +0200511 for (i = 0; i < ios->layout->s_numdevs; i++) {
Boaz Harrosh06886a52009-11-08 14:54:08 +0200512 struct osd_request *or;
513
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200514 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200515 if (unlikely(!or)) {
516 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
517 ret = -ENOMEM;
518 goto out;
519 }
520 ios->per_dev[i].or = or;
521 ios->numdevs++;
522
523 osd_req_create_object(or, &ios->obj);
524 }
525 ret = exofs_io_execute(ios);
526
527out:
528 return ret;
529}
530
531int exofs_sbi_remove(struct exofs_io_state *ios)
532{
533 int i, ret;
534
Boaz Harrosh45d3abc2010-01-28 11:46:16 +0200535 for (i = 0; i < ios->layout->s_numdevs; i++) {
Boaz Harrosh06886a52009-11-08 14:54:08 +0200536 struct osd_request *or;
537
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200538 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200539 if (unlikely(!or)) {
540 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
541 ret = -ENOMEM;
542 goto out;
543 }
544 ios->per_dev[i].or = or;
545 ios->numdevs++;
546
547 osd_req_remove_object(or, &ios->obj);
548 }
549 ret = exofs_io_execute(ios);
550
551out:
552 return ret;
553}
554
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200555static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200556{
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200557 struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp];
558 unsigned dev = ios->per_dev[cur_comp].dev;
559 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
560 int ret = 0;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200561
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200562 if (ios->pages && !master_dev->length)
563 return 0; /* Just an empty slot */
564
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200565 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
566 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
Boaz Harrosh06886a52009-11-08 14:54:08 +0200567 struct osd_request *or;
568
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200569 or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200570 if (unlikely(!or)) {
571 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
572 ret = -ENOMEM;
573 goto out;
574 }
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200575 per_dev->or = or;
576 per_dev->offset = master_dev->offset;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200577
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200578 if (ios->pages) {
Boaz Harrosh06886a52009-11-08 14:54:08 +0200579 struct bio *bio;
580
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200581 if (per_dev != master_dev) {
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200582 bio = bio_kmalloc(GFP_KERNEL,
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200583 master_dev->bio->bi_max_vecs);
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200584 if (unlikely(!bio)) {
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200585 EXOFS_DBGMSG(
586 "Faild to allocate BIO size=%u\n",
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200587 master_dev->bio->bi_max_vecs);
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200588 ret = -ENOMEM;
589 goto out;
590 }
591
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200592 __bio_clone(bio, master_dev->bio);
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200593 bio->bi_bdev = NULL;
594 bio->bi_next = NULL;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200595 per_dev->length = master_dev->length;
596 per_dev->bio = bio;
597 per_dev->dev = dev;
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200598 } else {
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200599 bio = master_dev->bio;
600 /* FIXME: bio_set_dir() */
601 bio->bi_rw |= (1 << BIO_RW);
Boaz Harrosh04dc1e82009-11-16 16:03:05 +0200602 }
Boaz Harrosh06886a52009-11-08 14:54:08 +0200603
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200604 osd_req_write(or, &ios->obj, per_dev->offset, bio,
605 per_dev->length);
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200606 EXOFS_DBGMSG("write(0x%llx) offset=0x%llx "
607 "length=0x%llx dev=%d\n",
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200608 _LLU(ios->obj.id), _LLU(per_dev->offset),
609 _LLU(per_dev->length), dev);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200610 } else if (ios->kern_buff) {
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200611 ret = osd_req_write_kern(or, &ios->obj, per_dev->offset,
Boaz Harrosh06886a52009-11-08 14:54:08 +0200612 ios->kern_buff, ios->length);
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200613 if (unlikely(ret))
614 goto out;
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200615 EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
616 "length=0x%llx dev=%d\n",
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200617 _LLU(ios->obj.id), _LLU(per_dev->offset),
618 _LLU(ios->length), dev);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200619 } else {
620 osd_req_set_attributes(or, &ios->obj);
Boaz Harrosh34ce4e72009-12-15 19:34:17 +0200621 EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200622 _LLU(ios->obj.id), ios->out_attr_len, dev);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200623 }
624
625 if (ios->out_attr)
626 osd_req_add_set_attr_list(or, ios->out_attr,
627 ios->out_attr_len);
628
629 if (ios->in_attr)
630 osd_req_add_get_attr_list(or, ios->in_attr,
631 ios->in_attr_len);
632 }
Boaz Harrosh06886a52009-11-08 14:54:08 +0200633
634out:
635 return ret;
636}
637
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200638int exofs_sbi_write(struct exofs_io_state *ios)
639{
640 int i;
641 int ret;
642
643 ret = _prepare_for_striping(ios);
644 if (unlikely(ret))
645 return ret;
646
647 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
648 ret = _sbi_write_mirror(ios, i);
649 if (unlikely(ret))
650 return ret;
651 }
652
653 ret = exofs_io_execute(ios);
654 return ret;
655}
656
657static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp)
Boaz Harrosh06886a52009-11-08 14:54:08 +0200658{
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200659 struct osd_request *or;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200660 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200661 unsigned first_dev = (unsigned)ios->obj.id;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200662
Boaz Harrosh50a76fd2010-02-11 13:01:39 +0200663 if (ios->pages && !per_dev->length)
664 return 0; /* Just an empty slot */
665
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200666 first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
Boaz Harroshd9c740d2010-01-28 11:58:08 +0200667 or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL);
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200668 if (unlikely(!or)) {
669 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
670 return -ENOMEM;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200671 }
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200672 per_dev->or = or;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200673
Boaz Harrosh86093aa2010-01-28 18:24:06 +0200674 if (ios->pages) {
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200675 osd_req_read(or, &ios->obj, per_dev->offset,
676 per_dev->bio, per_dev->length);
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200677 EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
678 " dev=%d\n", _LLU(ios->obj.id),
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200679 _LLU(per_dev->offset), _LLU(per_dev->length),
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200680 first_dev);
681 } else if (ios->kern_buff) {
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200682 int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset,
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200683 ios->kern_buff, ios->length);
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200684 EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
685 "length=0x%llx dev=%d ret=>%d\n",
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200686 _LLU(ios->obj.id), _LLU(per_dev->offset),
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200687 _LLU(ios->length), first_dev, ret);
688 if (unlikely(ret))
689 return ret;
690 } else {
691 osd_req_get_attributes(or, &ios->obj);
692 EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
693 _LLU(ios->obj.id), ios->in_attr_len, first_dev);
694 }
Boaz Harrosh46f4d972010-02-01 11:37:30 +0200695 if (ios->out_attr)
696 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
697
698 if (ios->in_attr)
699 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
700
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200701 return 0;
702}
703
704int exofs_sbi_read(struct exofs_io_state *ios)
705{
706 int i;
707 int ret;
708
709 ret = _prepare_for_striping(ios);
710 if (unlikely(ret))
711 return ret;
712
713 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
714 ret = _sbi_read_mirror(ios, i);
715 if (unlikely(ret))
716 return ret;
717 }
718
719 ret = exofs_io_execute(ios);
720 return ret;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200721}
722
723int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr)
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200724{
725 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
726 void *iter = NULL;
727 int nelem;
728
729 do {
730 nelem = 1;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200731 osd_req_decode_get_attr_list(ios->per_dev[0].or,
732 &cur_attr, &nelem, &iter);
Boaz Harroshb14f8ab2008-10-27 18:27:55 +0200733 if ((cur_attr.attr_page == attr->attr_page) &&
734 (cur_attr.attr_id == attr->attr_id)) {
735 attr->len = cur_attr.len;
736 attr->val_ptr = cur_attr.val_ptr;
737 return 0;
738 }
739 } while (iter);
740
741 return -EIO;
742}
Boaz Harrosh06886a52009-11-08 14:54:08 +0200743
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200744static int _truncate_mirrors(struct exofs_io_state *ios, unsigned cur_comp,
745 struct osd_attr *attr)
746{
747 int last_comp = cur_comp + ios->layout->mirrors_p1;
748
749 for (; cur_comp < last_comp; ++cur_comp) {
750 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
751 struct osd_request *or;
752
753 or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL);
754 if (unlikely(!or)) {
755 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
756 return -ENOMEM;
757 }
758 per_dev->or = or;
759
760 osd_req_set_attributes(or, &ios->obj);
761 osd_req_add_set_attr_list(or, attr, 1);
762 }
763
764 return 0;
765}
766
Boaz Harrosh06886a52009-11-08 14:54:08 +0200767int exofs_oi_truncate(struct exofs_i_info *oi, u64 size)
768{
769 struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info;
770 struct exofs_io_state *ios;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200771 struct exofs_trunc_attr {
772 struct osd_attr attr;
773 __be64 newsize;
774 } *size_attrs;
Boaz Harroshb367e782010-02-07 19:18:58 +0200775 struct _striping_info si;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200776 int i, ret;
777
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200778 ret = exofs_get_io_state(&sbi->layout, &ios);
779 if (unlikely(ret))
780 return ret;
781
782 size_attrs = kcalloc(ios->layout->group_width, sizeof(*size_attrs),
783 GFP_KERNEL);
784 if (unlikely(!size_attrs)) {
785 ret = -ENOMEM;
786 goto out;
787 }
Boaz Harrosh06886a52009-11-08 14:54:08 +0200788
789 ios->obj.id = exofs_oi_objno(oi);
790 ios->cred = oi->i_cred;
791
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200792 ios->numdevs = ios->layout->s_numdevs;
Boaz Harroshb367e782010-02-07 19:18:58 +0200793 _calc_stripe_info(ios, size, &si);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200794
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200795 for (i = 0; i < ios->layout->group_width; ++i) {
796 struct exofs_trunc_attr *size_attr = &size_attrs[i];
797 u64 obj_size;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200798
Boaz Harroshb367e782010-02-07 19:18:58 +0200799 if (i < si.dev)
800 obj_size = si.obj_offset +
801 ios->layout->stripe_unit - si.unit_off;
802 else if (i == si.dev)
803 obj_size = si.obj_offset;
804 else /* i > si.dev */
805 obj_size = si.obj_offset - si.unit_off;
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200806
807 size_attr->newsize = cpu_to_be64(obj_size);
808 size_attr->attr = g_attr_logical_length;
809 size_attr->attr.val_ptr = &size_attr->newsize;
810
811 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
812 &size_attr->attr);
813 if (unlikely(ret))
Boaz Harrosh06886a52009-11-08 14:54:08 +0200814 goto out;
Boaz Harrosh06886a52009-11-08 14:54:08 +0200815 }
816 ret = exofs_io_execute(ios);
817
818out:
Boaz Harrosh5d952b82010-02-01 13:35:51 +0200819 kfree(size_attrs);
Boaz Harrosh06886a52009-11-08 14:54:08 +0200820 exofs_put_io_state(ios);
821 return ret;
822}