Commit | Line | Data |
---|---|---|
b14f8ab2 BH |
1 | /* |
2 | * Copyright (C) 2005, 2006 | |
27d2e149 | 3 | * Avishay Traeger (avishay@gmail.com) |
b14f8ab2 BH |
4 | * Copyright (C) 2008, 2009 |
5 | * Boaz Harrosh <bharrosh@panasas.com> | |
6 | * | |
7 | * This file is part of exofs. | |
8 | * | |
9 | * exofs is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation. Since it is based on ext2, and the only | |
12 | * valid version of GPL for the Linux kernel is version 2, the only valid | |
13 | * version of GPL for exofs is version 2. | |
14 | * | |
15 | * exofs is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | * GNU General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public License | |
21 | * along with exofs; if not, write to the Free Software | |
22 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
23 | */ | |
24 | ||
5a0e3ad6 | 25 | #include <linux/slab.h> |
b14f8ab2 | 26 | #include <scsi/scsi_device.h> |
5d952b83 | 27 | #include <asm/div64.h> |
b14f8ab2 BH |
28 | |
29 | #include "exofs.h" | |
30 | ||
34ce4e7c BH |
31 | #define EXOFS_DBGMSG2(M...) do {} while (0) |
32 | /* #define EXOFS_DBGMSG2 EXOFS_DBGMSG */ | |
33 | ||
e1042ba0 BH |
34 | int exofs_get_rw_state(struct exofs_layout *layout, bool is_reading, |
35 | u64 offset, u64 length, struct exofs_io_state **pios) | |
b14f8ab2 | 36 | { |
06886a5a BH |
37 | struct exofs_io_state *ios; |
38 | ||
39 | /*TODO: Maybe use kmem_cach per sbi of size | |
45d3abcb | 40 | * exofs_io_state_size(layout->s_numdevs) |
06886a5a | 41 | */ |
45d3abcb | 42 | ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL); |
06886a5a | 43 | if (unlikely(!ios)) { |
426d3107 | 44 | EXOFS_DBGMSG("Failed kzalloc bytes=%d\n", |
45d3abcb | 45 | exofs_io_state_size(layout->s_numdevs)); |
06886a5a BH |
46 | *pios = NULL; |
47 | return -ENOMEM; | |
48 | } | |
49 | ||
45d3abcb BH |
50 | ios->layout = layout; |
51 | ios->obj.partition = layout->s_pid; | |
e1042ba0 BH |
52 | ios->offset = offset; |
53 | ios->length = length; | |
54 | ios->reading = is_reading; | |
55 | ||
06886a5a BH |
56 | *pios = ios; |
57 | return 0; | |
b14f8ab2 BH |
58 | } |
59 | ||
e1042ba0 BH |
60 | int exofs_get_io_state(struct exofs_layout *layout, |
61 | struct exofs_io_state **ios) | |
62 | { | |
63 | return exofs_get_rw_state(layout, true, 0, 0, ios); | |
64 | } | |
65 | ||
06886a5a | 66 | void exofs_put_io_state(struct exofs_io_state *ios) |
b14f8ab2 | 67 | { |
06886a5a BH |
68 | if (ios) { |
69 | unsigned i; | |
b14f8ab2 | 70 | |
06886a5a BH |
71 | for (i = 0; i < ios->numdevs; i++) { |
72 | struct exofs_per_dev_state *per_dev = &ios->per_dev[i]; | |
73 | ||
74 | if (per_dev->or) | |
75 | osd_end_request(per_dev->or); | |
76 | if (per_dev->bio) | |
77 | bio_put(per_dev->bio); | |
78 | } | |
79 | ||
80 | kfree(ios); | |
b14f8ab2 | 81 | } |
06886a5a | 82 | } |
b14f8ab2 | 83 | |
06886a5a BH |
84 | static void _sync_done(struct exofs_io_state *ios, void *p) |
85 | { | |
86 | struct completion *waiting = p; | |
b14f8ab2 | 87 | |
06886a5a BH |
88 | complete(waiting); |
89 | } | |
90 | ||
91 | static void _last_io(struct kref *kref) | |
92 | { | |
93 | struct exofs_io_state *ios = container_of( | |
94 | kref, struct exofs_io_state, kref); | |
95 | ||
96 | ios->done(ios, ios->private); | |
97 | } | |
98 | ||
99 | static void _done_io(struct osd_request *or, void *p) | |
100 | { | |
101 | struct exofs_io_state *ios = p; | |
102 | ||
103 | kref_put(&ios->kref, _last_io); | |
104 | } | |
105 | ||
106 | static int exofs_io_execute(struct exofs_io_state *ios) | |
107 | { | |
108 | DECLARE_COMPLETION_ONSTACK(wait); | |
109 | bool sync = (ios->done == NULL); | |
110 | int i, ret; | |
111 | ||
112 | if (sync) { | |
113 | ios->done = _sync_done; | |
114 | ios->private = &wait; | |
115 | } | |
116 | ||
117 | for (i = 0; i < ios->numdevs; i++) { | |
118 | struct osd_request *or = ios->per_dev[i].or; | |
119 | if (unlikely(!or)) | |
120 | continue; | |
121 | ||
122 | ret = osd_finalize_request(or, 0, ios->cred, NULL); | |
123 | if (unlikely(ret)) { | |
426d3107 | 124 | EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", |
06886a5a BH |
125 | ret); |
126 | return ret; | |
127 | } | |
128 | } | |
129 | ||
130 | kref_init(&ios->kref); | |
131 | ||
132 | for (i = 0; i < ios->numdevs; i++) { | |
133 | struct osd_request *or = ios->per_dev[i].or; | |
134 | if (unlikely(!or)) | |
135 | continue; | |
136 | ||
137 | kref_get(&ios->kref); | |
138 | osd_execute_request_async(or, _done_io, ios); | |
139 | } | |
140 | ||
141 | kref_put(&ios->kref, _last_io); | |
142 | ret = 0; | |
143 | ||
144 | if (sync) { | |
145 | wait_for_completion(&wait); | |
146 | ret = exofs_check_io(ios, NULL); | |
147 | } | |
b14f8ab2 BH |
148 | return ret; |
149 | } | |
150 | ||
22ddc556 BH |
151 | static void _clear_bio(struct bio *bio) |
152 | { | |
153 | struct bio_vec *bv; | |
154 | unsigned i; | |
155 | ||
156 | __bio_for_each_segment(bv, bio, i, 0) { | |
157 | unsigned this_count = bv->bv_len; | |
158 | ||
159 | if (likely(PAGE_SIZE == this_count)) | |
160 | clear_highpage(bv->bv_page); | |
161 | else | |
162 | zero_user(bv->bv_page, bv->bv_offset, this_count); | |
163 | } | |
164 | } | |
165 | ||
06886a5a | 166 | int exofs_check_io(struct exofs_io_state *ios, u64 *resid) |
b14f8ab2 | 167 | { |
06886a5a BH |
168 | enum osd_err_priority acumulated_osd_err = 0; |
169 | int acumulated_lin_err = 0; | |
170 | int i; | |
b14f8ab2 | 171 | |
06886a5a BH |
172 | for (i = 0; i < ios->numdevs; i++) { |
173 | struct osd_sense_info osi; | |
22ddc556 BH |
174 | struct osd_request *or = ios->per_dev[i].or; |
175 | int ret; | |
176 | ||
177 | if (unlikely(!or)) | |
178 | continue; | |
06886a5a | 179 | |
22ddc556 | 180 | ret = osd_req_decode_sense(or, &osi); |
06886a5a BH |
181 | if (likely(!ret)) |
182 | continue; | |
183 | ||
22ddc556 BH |
184 | if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) { |
185 | /* start read offset passed endof file */ | |
186 | _clear_bio(ios->per_dev[i].bio); | |
187 | EXOFS_DBGMSG("start read offset passed end of file " | |
188 | "offset=0x%llx, length=0x%llx\n", | |
5d952b83 BH |
189 | _LLU(ios->per_dev[i].offset), |
190 | _LLU(ios->per_dev[i].length)); | |
22ddc556 BH |
191 | |
192 | continue; /* we recovered */ | |
06886a5a BH |
193 | } |
194 | ||
195 | if (osi.osd_err_pri >= acumulated_osd_err) { | |
196 | acumulated_osd_err = osi.osd_err_pri; | |
197 | acumulated_lin_err = ret; | |
198 | } | |
199 | } | |
200 | ||
201 | /* TODO: raid specific residual calculations */ | |
202 | if (resid) { | |
203 | if (likely(!acumulated_lin_err)) | |
204 | *resid = 0; | |
205 | else | |
206 | *resid = ios->length; | |
207 | } | |
208 | ||
209 | return acumulated_lin_err; | |
210 | } | |
211 | ||
b367e78b BH |
212 | /* |
213 | * L - logical offset into the file | |
214 | * | |
50a76fd3 | 215 | * U - The number of bytes in a stripe within a group |
b367e78b BH |
216 | * |
217 | * U = stripe_unit * group_width | |
218 | * | |
50a76fd3 BH |
219 | * T - The number of bytes striped within a group of component objects |
220 | * (before advancing to the next group) | |
b367e78b | 221 | * |
50a76fd3 BH |
222 | * T = stripe_unit * group_width * group_depth |
223 | * | |
224 | * S - The number of bytes striped across all component objects | |
225 | * before the pattern repeats | |
226 | * | |
227 | * S = stripe_unit * group_width * group_depth * group_count | |
228 | * | |
229 | * M - The "major" (i.e., across all components) stripe number | |
230 | * | |
231 | * M = L / S | |
232 | * | |
233 | * G - Counts the groups from the beginning of the major stripe | |
234 | * | |
235 | * G = (L - (M * S)) / T [or (L % S) / T] | |
236 | * | |
237 | * H - The byte offset within the group | |
238 | * | |
239 | * H = (L - (M * S)) % T [or (L % S) % T] | |
240 | * | |
241 | * N - The "minor" (i.e., across the group) stripe number | |
242 | * | |
243 | * N = H / U | |
b367e78b BH |
244 | * |
245 | * C - The component index coresponding to L | |
246 | * | |
50a76fd3 BH |
247 | * C = (H - (N * U)) / stripe_unit + G * group_width |
248 | * [or (L % U) / stripe_unit + G * group_width] | |
b367e78b BH |
249 | * |
250 | * O - The component offset coresponding to L | |
251 | * | |
50a76fd3 | 252 | * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit |
b367e78b | 253 | */ |
b367e78b BH |
254 | struct _striping_info { |
255 | u64 obj_offset; | |
50a76fd3 | 256 | u64 group_length; |
16f75bb3 | 257 | u64 M; /* for truncate */ |
b367e78b BH |
258 | unsigned dev; |
259 | unsigned unit_off; | |
260 | }; | |
261 | ||
16f75bb3 | 262 | static void _calc_stripe_info(struct exofs_layout *layout, u64 file_offset, |
b367e78b | 263 | struct _striping_info *si) |
5d952b83 | 264 | { |
16f75bb3 BH |
265 | u32 stripe_unit = layout->stripe_unit; |
266 | u32 group_width = layout->group_width; | |
267 | u64 group_depth = layout->group_depth; | |
50a76fd3 | 268 | |
b367e78b | 269 | u32 U = stripe_unit * group_width; |
50a76fd3 | 270 | u64 T = U * group_depth; |
16f75bb3 | 271 | u64 S = T * layout->group_count; |
50a76fd3 BH |
272 | u64 M = div64_u64(file_offset, S); |
273 | ||
274 | /* | |
275 | G = (L - (M * S)) / T | |
276 | H = (L - (M * S)) % T | |
277 | */ | |
278 | u64 LmodS = file_offset - M * S; | |
279 | u32 G = div64_u64(LmodS, T); | |
280 | u64 H = LmodS - G * T; | |
281 | ||
282 | u32 N = div_u64(H, U); | |
283 | ||
284 | /* "H - (N * U)" is just "H % U" so it's bound to u32 */ | |
285 | si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width; | |
16f75bb3 | 286 | si->dev *= layout->mirrors_p1; |
b367e78b | 287 | |
50a76fd3 | 288 | div_u64_rem(file_offset, stripe_unit, &si->unit_off); |
5d952b83 | 289 | |
50a76fd3 BH |
290 | si->obj_offset = si->unit_off + (N * stripe_unit) + |
291 | (M * group_depth * stripe_unit); | |
292 | ||
293 | si->group_length = T - H; | |
16f75bb3 | 294 | si->M = M; |
5d952b83 BH |
295 | } |
296 | ||
86093aaf BH |
297 | static int _add_stripe_unit(struct exofs_io_state *ios, unsigned *cur_pg, |
298 | unsigned pgbase, struct exofs_per_dev_state *per_dev, | |
299 | int cur_len) | |
5d952b83 | 300 | { |
86093aaf | 301 | unsigned pg = *cur_pg; |
5d952b83 BH |
302 | struct request_queue *q = |
303 | osd_request_queue(exofs_ios_od(ios, per_dev->dev)); | |
304 | ||
305 | per_dev->length += cur_len; | |
306 | ||
307 | if (per_dev->bio == NULL) { | |
308 | unsigned pages_in_stripe = ios->layout->group_width * | |
309 | (ios->layout->stripe_unit / PAGE_SIZE); | |
86093aaf | 310 | unsigned bio_size = (ios->nr_pages + pages_in_stripe) / |
5d952b83 BH |
311 | ios->layout->group_width; |
312 | ||
313 | per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size); | |
314 | if (unlikely(!per_dev->bio)) { | |
426d3107 | 315 | EXOFS_DBGMSG("Failed to allocate BIO size=%u\n", |
5d952b83 BH |
316 | bio_size); |
317 | return -ENOMEM; | |
318 | } | |
319 | } | |
320 | ||
321 | while (cur_len > 0) { | |
86093aaf BH |
322 | unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len); |
323 | unsigned added_len; | |
5d952b83 | 324 | |
86093aaf BH |
325 | BUG_ON(ios->nr_pages <= pg); |
326 | cur_len -= pglen; | |
5d952b83 | 327 | |
86093aaf BH |
328 | added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg], |
329 | pglen, pgbase); | |
330 | if (unlikely(pglen != added_len)) | |
5d952b83 | 331 | return -ENOMEM; |
86093aaf BH |
332 | pgbase = 0; |
333 | ++pg; | |
5d952b83 BH |
334 | } |
335 | BUG_ON(cur_len); | |
336 | ||
86093aaf | 337 | *cur_pg = pg; |
5d952b83 BH |
338 | return 0; |
339 | } | |
340 | ||
50a76fd3 | 341 | static int _prepare_one_group(struct exofs_io_state *ios, u64 length, |
6e31609b | 342 | struct _striping_info *si) |
5d952b83 | 343 | { |
5d952b83 | 344 | unsigned stripe_unit = ios->layout->stripe_unit; |
b367e78b | 345 | unsigned mirrors_p1 = ios->layout->mirrors_p1; |
50a76fd3 | 346 | unsigned devs_in_group = ios->layout->group_width * mirrors_p1; |
b367e78b | 347 | unsigned dev = si->dev; |
50a76fd3 | 348 | unsigned first_dev = dev - (dev % devs_in_group); |
50a76fd3 BH |
349 | unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0; |
350 | unsigned cur_pg = ios->pages_consumed; | |
86093aaf | 351 | int ret = 0; |
5d952b83 | 352 | |
5d952b83 | 353 | while (length) { |
6e31609b | 354 | struct exofs_per_dev_state *per_dev = &ios->per_dev[dev]; |
b367e78b | 355 | unsigned cur_len, page_off = 0; |
5d952b83 BH |
356 | |
357 | if (!per_dev->length) { | |
b367e78b BH |
358 | per_dev->dev = dev; |
359 | if (dev < si->dev) { | |
360 | per_dev->offset = si->obj_offset + stripe_unit - | |
361 | si->unit_off; | |
362 | cur_len = stripe_unit; | |
363 | } else if (dev == si->dev) { | |
364 | per_dev->offset = si->obj_offset; | |
365 | cur_len = stripe_unit - si->unit_off; | |
366 | page_off = si->unit_off & ~PAGE_MASK; | |
367 | BUG_ON(page_off && (page_off != ios->pgbase)); | |
368 | } else { /* dev > si->dev */ | |
369 | per_dev->offset = si->obj_offset - si->unit_off; | |
370 | cur_len = stripe_unit; | |
371 | } | |
5d952b83 | 372 | |
6e31609b BH |
373 | if (max_comp < dev) |
374 | max_comp = dev; | |
5d952b83 | 375 | } else { |
b367e78b | 376 | cur_len = stripe_unit; |
5d952b83 | 377 | } |
b367e78b BH |
378 | if (cur_len >= length) |
379 | cur_len = length; | |
5d952b83 | 380 | |
86093aaf BH |
381 | ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev, |
382 | cur_len); | |
5d952b83 BH |
383 | if (unlikely(ret)) |
384 | goto out; | |
385 | ||
6e31609b BH |
386 | dev += mirrors_p1; |
387 | dev = (dev % devs_in_group) + first_dev; | |
5d952b83 BH |
388 | |
389 | length -= cur_len; | |
390 | } | |
391 | out: | |
50a76fd3 BH |
392 | ios->numdevs = max_comp + mirrors_p1; |
393 | ios->pages_consumed = cur_pg; | |
5d952b83 BH |
394 | return ret; |
395 | } | |
396 | ||
b367e78b BH |
397 | static int _prepare_for_striping(struct exofs_io_state *ios) |
398 | { | |
50a76fd3 | 399 | u64 length = ios->length; |
5002dd18 | 400 | u64 offset = ios->offset; |
b367e78b | 401 | struct _striping_info si; |
50a76fd3 | 402 | int ret = 0; |
b367e78b | 403 | |
b367e78b BH |
404 | if (!ios->pages) { |
405 | if (ios->kern_buff) { | |
406 | struct exofs_per_dev_state *per_dev = &ios->per_dev[0]; | |
407 | ||
16f75bb3 | 408 | _calc_stripe_info(ios->layout, ios->offset, &si); |
b367e78b BH |
409 | per_dev->offset = si.obj_offset; |
410 | per_dev->dev = si.dev; | |
411 | ||
412 | /* no cross device without page array */ | |
413 | BUG_ON((ios->layout->group_width > 1) && | |
414 | (si.unit_off + ios->length > | |
415 | ios->layout->stripe_unit)); | |
416 | } | |
417 | ios->numdevs = ios->layout->mirrors_p1; | |
418 | return 0; | |
419 | } | |
420 | ||
50a76fd3 | 421 | while (length) { |
16f75bb3 | 422 | _calc_stripe_info(ios->layout, offset, &si); |
5002dd18 | 423 | |
50a76fd3 BH |
424 | if (length < si.group_length) |
425 | si.group_length = length; | |
426 | ||
6e31609b | 427 | ret = _prepare_one_group(ios, si.group_length, &si); |
50a76fd3 BH |
428 | if (unlikely(ret)) |
429 | goto out; | |
430 | ||
5002dd18 | 431 | offset += si.group_length; |
50a76fd3 | 432 | length -= si.group_length; |
50a76fd3 BH |
433 | } |
434 | ||
435 | out: | |
436 | return ret; | |
b367e78b BH |
437 | } |
438 | ||
06886a5a BH |
439 | int exofs_sbi_create(struct exofs_io_state *ios) |
440 | { | |
441 | int i, ret; | |
442 | ||
45d3abcb | 443 | for (i = 0; i < ios->layout->s_numdevs; i++) { |
06886a5a BH |
444 | struct osd_request *or; |
445 | ||
d9c740d2 | 446 | or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL); |
06886a5a BH |
447 | if (unlikely(!or)) { |
448 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
449 | ret = -ENOMEM; | |
450 | goto out; | |
451 | } | |
452 | ios->per_dev[i].or = or; | |
453 | ios->numdevs++; | |
454 | ||
455 | osd_req_create_object(or, &ios->obj); | |
456 | } | |
457 | ret = exofs_io_execute(ios); | |
458 | ||
459 | out: | |
460 | return ret; | |
461 | } | |
462 | ||
463 | int exofs_sbi_remove(struct exofs_io_state *ios) | |
464 | { | |
465 | int i, ret; | |
466 | ||
45d3abcb | 467 | for (i = 0; i < ios->layout->s_numdevs; i++) { |
06886a5a BH |
468 | struct osd_request *or; |
469 | ||
d9c740d2 | 470 | or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL); |
06886a5a BH |
471 | if (unlikely(!or)) { |
472 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
473 | ret = -ENOMEM; | |
474 | goto out; | |
475 | } | |
476 | ios->per_dev[i].or = or; | |
477 | ios->numdevs++; | |
478 | ||
479 | osd_req_remove_object(or, &ios->obj); | |
480 | } | |
481 | ret = exofs_io_execute(ios); | |
482 | ||
483 | out: | |
484 | return ret; | |
485 | } | |
486 | ||
5d952b83 | 487 | static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp) |
06886a5a | 488 | { |
5d952b83 BH |
489 | struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp]; |
490 | unsigned dev = ios->per_dev[cur_comp].dev; | |
491 | unsigned last_comp = cur_comp + ios->layout->mirrors_p1; | |
492 | int ret = 0; | |
06886a5a | 493 | |
50a76fd3 BH |
494 | if (ios->pages && !master_dev->length) |
495 | return 0; /* Just an empty slot */ | |
496 | ||
5d952b83 BH |
497 | for (; cur_comp < last_comp; ++cur_comp, ++dev) { |
498 | struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp]; | |
06886a5a BH |
499 | struct osd_request *or; |
500 | ||
5d952b83 | 501 | or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL); |
06886a5a BH |
502 | if (unlikely(!or)) { |
503 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
504 | ret = -ENOMEM; | |
505 | goto out; | |
506 | } | |
5d952b83 BH |
507 | per_dev->or = or; |
508 | per_dev->offset = master_dev->offset; | |
06886a5a | 509 | |
86093aaf | 510 | if (ios->pages) { |
06886a5a BH |
511 | struct bio *bio; |
512 | ||
5d952b83 | 513 | if (per_dev != master_dev) { |
04dc1e88 | 514 | bio = bio_kmalloc(GFP_KERNEL, |
5d952b83 | 515 | master_dev->bio->bi_max_vecs); |
04dc1e88 | 516 | if (unlikely(!bio)) { |
34ce4e7c | 517 | EXOFS_DBGMSG( |
426d3107 | 518 | "Failed to allocate BIO size=%u\n", |
5d952b83 | 519 | master_dev->bio->bi_max_vecs); |
04dc1e88 BH |
520 | ret = -ENOMEM; |
521 | goto out; | |
522 | } | |
523 | ||
5d952b83 | 524 | __bio_clone(bio, master_dev->bio); |
04dc1e88 BH |
525 | bio->bi_bdev = NULL; |
526 | bio->bi_next = NULL; | |
5d952b83 BH |
527 | per_dev->length = master_dev->length; |
528 | per_dev->bio = bio; | |
529 | per_dev->dev = dev; | |
04dc1e88 | 530 | } else { |
5d952b83 BH |
531 | bio = master_dev->bio; |
532 | /* FIXME: bio_set_dir() */ | |
7b6d91da | 533 | bio->bi_rw |= REQ_WRITE; |
04dc1e88 | 534 | } |
06886a5a | 535 | |
5d952b83 BH |
536 | osd_req_write(or, &ios->obj, per_dev->offset, bio, |
537 | per_dev->length); | |
34ce4e7c BH |
538 | EXOFS_DBGMSG("write(0x%llx) offset=0x%llx " |
539 | "length=0x%llx dev=%d\n", | |
5d952b83 BH |
540 | _LLU(ios->obj.id), _LLU(per_dev->offset), |
541 | _LLU(per_dev->length), dev); | |
06886a5a | 542 | } else if (ios->kern_buff) { |
5d952b83 | 543 | ret = osd_req_write_kern(or, &ios->obj, per_dev->offset, |
06886a5a | 544 | ios->kern_buff, ios->length); |
5d952b83 BH |
545 | if (unlikely(ret)) |
546 | goto out; | |
34ce4e7c BH |
547 | EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx " |
548 | "length=0x%llx dev=%d\n", | |
5d952b83 BH |
549 | _LLU(ios->obj.id), _LLU(per_dev->offset), |
550 | _LLU(ios->length), dev); | |
06886a5a BH |
551 | } else { |
552 | osd_req_set_attributes(or, &ios->obj); | |
34ce4e7c | 553 | EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n", |
5d952b83 | 554 | _LLU(ios->obj.id), ios->out_attr_len, dev); |
06886a5a BH |
555 | } |
556 | ||
557 | if (ios->out_attr) | |
558 | osd_req_add_set_attr_list(or, ios->out_attr, | |
559 | ios->out_attr_len); | |
560 | ||
561 | if (ios->in_attr) | |
562 | osd_req_add_get_attr_list(or, ios->in_attr, | |
563 | ios->in_attr_len); | |
b14f8ab2 | 564 | } |
06886a5a BH |
565 | |
566 | out: | |
567 | return ret; | |
568 | } | |
569 | ||
5d952b83 BH |
570 | int exofs_sbi_write(struct exofs_io_state *ios) |
571 | { | |
572 | int i; | |
573 | int ret; | |
574 | ||
575 | ret = _prepare_for_striping(ios); | |
576 | if (unlikely(ret)) | |
577 | return ret; | |
578 | ||
579 | for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) { | |
580 | ret = _sbi_write_mirror(ios, i); | |
581 | if (unlikely(ret)) | |
582 | return ret; | |
583 | } | |
584 | ||
585 | ret = exofs_io_execute(ios); | |
586 | return ret; | |
587 | } | |
588 | ||
589 | static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp) | |
06886a5a | 590 | { |
46f4d973 | 591 | struct osd_request *or; |
5d952b83 | 592 | struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp]; |
46f4d973 | 593 | unsigned first_dev = (unsigned)ios->obj.id; |
06886a5a | 594 | |
50a76fd3 BH |
595 | if (ios->pages && !per_dev->length) |
596 | return 0; /* Just an empty slot */ | |
597 | ||
5d952b83 | 598 | first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1; |
d9c740d2 | 599 | or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL); |
46f4d973 BH |
600 | if (unlikely(!or)) { |
601 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
602 | return -ENOMEM; | |
603 | } | |
604 | per_dev->or = or; | |
46f4d973 | 605 | |
86093aaf | 606 | if (ios->pages) { |
5d952b83 BH |
607 | osd_req_read(or, &ios->obj, per_dev->offset, |
608 | per_dev->bio, per_dev->length); | |
46f4d973 BH |
609 | EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx" |
610 | " dev=%d\n", _LLU(ios->obj.id), | |
5d952b83 | 611 | _LLU(per_dev->offset), _LLU(per_dev->length), |
46f4d973 BH |
612 | first_dev); |
613 | } else if (ios->kern_buff) { | |
5d952b83 | 614 | int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset, |
46f4d973 | 615 | ios->kern_buff, ios->length); |
46f4d973 BH |
616 | EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx " |
617 | "length=0x%llx dev=%d ret=>%d\n", | |
5d952b83 | 618 | _LLU(ios->obj.id), _LLU(per_dev->offset), |
46f4d973 BH |
619 | _LLU(ios->length), first_dev, ret); |
620 | if (unlikely(ret)) | |
621 | return ret; | |
622 | } else { | |
623 | osd_req_get_attributes(or, &ios->obj); | |
624 | EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n", | |
625 | _LLU(ios->obj.id), ios->in_attr_len, first_dev); | |
626 | } | |
46f4d973 BH |
627 | if (ios->out_attr) |
628 | osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len); | |
b14f8ab2 | 629 | |
46f4d973 BH |
630 | if (ios->in_attr) |
631 | osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len); | |
b14f8ab2 | 632 | |
5d952b83 BH |
633 | return 0; |
634 | } | |
635 | ||
636 | int exofs_sbi_read(struct exofs_io_state *ios) | |
637 | { | |
638 | int i; | |
639 | int ret; | |
640 | ||
641 | ret = _prepare_for_striping(ios); | |
642 | if (unlikely(ret)) | |
643 | return ret; | |
644 | ||
645 | for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) { | |
646 | ret = _sbi_read_mirror(ios, i); | |
647 | if (unlikely(ret)) | |
648 | return ret; | |
649 | } | |
650 | ||
651 | ret = exofs_io_execute(ios); | |
652 | return ret; | |
b14f8ab2 BH |
653 | } |
654 | ||
06886a5a | 655 | int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr) |
b14f8ab2 BH |
656 | { |
657 | struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */ | |
658 | void *iter = NULL; | |
659 | int nelem; | |
660 | ||
661 | do { | |
662 | nelem = 1; | |
06886a5a BH |
663 | osd_req_decode_get_attr_list(ios->per_dev[0].or, |
664 | &cur_attr, &nelem, &iter); | |
b14f8ab2 BH |
665 | if ((cur_attr.attr_page == attr->attr_page) && |
666 | (cur_attr.attr_id == attr->attr_id)) { | |
667 | attr->len = cur_attr.len; | |
668 | attr->val_ptr = cur_attr.val_ptr; | |
669 | return 0; | |
670 | } | |
671 | } while (iter); | |
672 | ||
673 | return -EIO; | |
674 | } | |
06886a5a | 675 | |
5d952b83 BH |
676 | static int _truncate_mirrors(struct exofs_io_state *ios, unsigned cur_comp, |
677 | struct osd_attr *attr) | |
678 | { | |
679 | int last_comp = cur_comp + ios->layout->mirrors_p1; | |
680 | ||
681 | for (; cur_comp < last_comp; ++cur_comp) { | |
682 | struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp]; | |
683 | struct osd_request *or; | |
684 | ||
685 | or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL); | |
686 | if (unlikely(!or)) { | |
687 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
688 | return -ENOMEM; | |
689 | } | |
690 | per_dev->or = or; | |
691 | ||
692 | osd_req_set_attributes(or, &ios->obj); | |
693 | osd_req_add_set_attr_list(or, attr, 1); | |
694 | } | |
695 | ||
696 | return 0; | |
697 | } | |
698 | ||
16f75bb3 BH |
699 | struct _trunc_info { |
700 | struct _striping_info si; | |
701 | u64 prev_group_obj_off; | |
702 | u64 next_group_obj_off; | |
703 | ||
704 | unsigned first_group_dev; | |
705 | unsigned nex_group_dev; | |
706 | unsigned max_devs; | |
707 | }; | |
708 | ||
709 | void _calc_trunk_info(struct exofs_layout *layout, u64 file_offset, | |
710 | struct _trunc_info *ti) | |
711 | { | |
712 | unsigned stripe_unit = layout->stripe_unit; | |
713 | ||
714 | _calc_stripe_info(layout, file_offset, &ti->si); | |
715 | ||
716 | ti->prev_group_obj_off = ti->si.M * stripe_unit; | |
717 | ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0; | |
718 | ||
719 | ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width); | |
720 | ti->nex_group_dev = ti->first_group_dev + layout->group_width; | |
721 | ti->max_devs = layout->group_width * layout->group_count; | |
722 | } | |
723 | ||
06886a5a BH |
724 | int exofs_oi_truncate(struct exofs_i_info *oi, u64 size) |
725 | { | |
726 | struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info; | |
727 | struct exofs_io_state *ios; | |
5d952b83 BH |
728 | struct exofs_trunc_attr { |
729 | struct osd_attr attr; | |
730 | __be64 newsize; | |
731 | } *size_attrs; | |
16f75bb3 | 732 | struct _trunc_info ti; |
06886a5a BH |
733 | int i, ret; |
734 | ||
5d952b83 BH |
735 | ret = exofs_get_io_state(&sbi->layout, &ios); |
736 | if (unlikely(ret)) | |
737 | return ret; | |
738 | ||
16f75bb3 BH |
739 | _calc_trunk_info(ios->layout, size, &ti); |
740 | ||
741 | size_attrs = kcalloc(ti.max_devs, sizeof(*size_attrs), | |
5d952b83 BH |
742 | GFP_KERNEL); |
743 | if (unlikely(!size_attrs)) { | |
744 | ret = -ENOMEM; | |
745 | goto out; | |
746 | } | |
06886a5a BH |
747 | |
748 | ios->obj.id = exofs_oi_objno(oi); | |
749 | ios->cred = oi->i_cred; | |
5d952b83 | 750 | ios->numdevs = ios->layout->s_numdevs; |
06886a5a | 751 | |
16f75bb3 | 752 | for (i = 0; i < ti.max_devs; ++i) { |
5d952b83 BH |
753 | struct exofs_trunc_attr *size_attr = &size_attrs[i]; |
754 | u64 obj_size; | |
06886a5a | 755 | |
16f75bb3 BH |
756 | if (i < ti.first_group_dev) |
757 | obj_size = ti.prev_group_obj_off; | |
758 | else if (i >= ti.nex_group_dev) | |
759 | obj_size = ti.next_group_obj_off; | |
760 | else if (i < ti.si.dev) /* dev within this group */ | |
761 | obj_size = ti.si.obj_offset + | |
762 | ios->layout->stripe_unit - ti.si.unit_off; | |
763 | else if (i == ti.si.dev) | |
764 | obj_size = ti.si.obj_offset; | |
765 | else /* i > ti.dev */ | |
766 | obj_size = ti.si.obj_offset - ti.si.unit_off; | |
06886a5a | 767 | |
5d952b83 BH |
768 | size_attr->newsize = cpu_to_be64(obj_size); |
769 | size_attr->attr = g_attr_logical_length; | |
770 | size_attr->attr.val_ptr = &size_attr->newsize; | |
771 | ||
16f75bb3 BH |
772 | EXOFS_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n", |
773 | _LLU(ios->obj.id), _LLU(obj_size), i); | |
5d952b83 BH |
774 | ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1, |
775 | &size_attr->attr); | |
776 | if (unlikely(ret)) | |
777 | goto out; | |
06886a5a BH |
778 | } |
779 | ret = exofs_io_execute(ios); | |
780 | ||
781 | out: | |
5d952b83 | 782 | kfree(size_attrs); |
06886a5a BH |
783 | exofs_put_io_state(ios); |
784 | return ret; | |
785 | } | |
85e44df4 BH |
786 | |
787 | const struct osd_attr g_attr_logical_length = ATTR_DEF( | |
788 | OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8); |