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 | ||
06886a5a | 34 | void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj) |
b14f8ab2 | 35 | { |
06886a5a BH |
36 | osd_sec_init_nosec_doall_caps(cred_a, obj, false, true); |
37 | } | |
38 | ||
39 | int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj, | |
40 | u64 offset, void *p, unsigned length) | |
41 | { | |
42 | struct osd_request *or = osd_start_request(od, GFP_KERNEL); | |
43 | /* struct osd_sense_info osi = {.key = 0};*/ | |
44 | int ret; | |
45 | ||
46 | if (unlikely(!or)) { | |
47 | EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__); | |
48 | return -ENOMEM; | |
49 | } | |
50 | ret = osd_req_read_kern(or, obj, offset, p, length); | |
51 | if (unlikely(ret)) { | |
52 | EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__); | |
53 | goto out; | |
b14f8ab2 BH |
54 | } |
55 | ||
06886a5a BH |
56 | ret = osd_finalize_request(or, 0, cred, NULL); |
57 | if (unlikely(ret)) { | |
58 | EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n", ret); | |
59 | goto out; | |
60 | } | |
b14f8ab2 | 61 | |
06886a5a BH |
62 | ret = osd_execute_request(or); |
63 | if (unlikely(ret)) | |
64 | EXOFS_DBGMSG("osd_execute_request() => %d\n", ret); | |
65 | /* osd_req_decode_sense(or, ret); */ | |
b14f8ab2 | 66 | |
06886a5a BH |
67 | out: |
68 | osd_end_request(or); | |
b14f8ab2 BH |
69 | return ret; |
70 | } | |
71 | ||
45d3abcb BH |
72 | int exofs_get_io_state(struct exofs_layout *layout, |
73 | struct exofs_io_state **pios) | |
b14f8ab2 | 74 | { |
06886a5a BH |
75 | struct exofs_io_state *ios; |
76 | ||
77 | /*TODO: Maybe use kmem_cach per sbi of size | |
45d3abcb | 78 | * exofs_io_state_size(layout->s_numdevs) |
06886a5a | 79 | */ |
45d3abcb | 80 | ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL); |
06886a5a | 81 | if (unlikely(!ios)) { |
34ce4e7c | 82 | EXOFS_DBGMSG("Faild kzalloc bytes=%d\n", |
45d3abcb | 83 | exofs_io_state_size(layout->s_numdevs)); |
06886a5a BH |
84 | *pios = NULL; |
85 | return -ENOMEM; | |
86 | } | |
87 | ||
45d3abcb BH |
88 | ios->layout = layout; |
89 | ios->obj.partition = layout->s_pid; | |
06886a5a BH |
90 | *pios = ios; |
91 | return 0; | |
b14f8ab2 BH |
92 | } |
93 | ||
06886a5a | 94 | void exofs_put_io_state(struct exofs_io_state *ios) |
b14f8ab2 | 95 | { |
06886a5a BH |
96 | if (ios) { |
97 | unsigned i; | |
b14f8ab2 | 98 | |
06886a5a BH |
99 | for (i = 0; i < ios->numdevs; i++) { |
100 | struct exofs_per_dev_state *per_dev = &ios->per_dev[i]; | |
101 | ||
102 | if (per_dev->or) | |
103 | osd_end_request(per_dev->or); | |
104 | if (per_dev->bio) | |
105 | bio_put(per_dev->bio); | |
106 | } | |
107 | ||
108 | kfree(ios); | |
b14f8ab2 | 109 | } |
06886a5a | 110 | } |
b14f8ab2 | 111 | |
d9c740d2 BH |
112 | unsigned exofs_layout_od_id(struct exofs_layout *layout, |
113 | osd_id obj_no, unsigned layout_index) | |
114 | { | |
5d952b83 BH |
115 | /* switch (layout->lay_func) { |
116 | case LAYOUT_MOVING_WINDOW: | |
117 | {*/ | |
118 | unsigned dev_mod = obj_no; | |
119 | ||
120 | return (layout_index + dev_mod * layout->mirrors_p1) % | |
121 | layout->s_numdevs; | |
122 | /* } | |
123 | case LAYOUT_FUNC_IMPLICT: | |
124 | return layout->devs[layout_index]; | |
125 | }*/ | |
d9c740d2 BH |
126 | } |
127 | ||
128 | static inline struct osd_dev *exofs_ios_od(struct exofs_io_state *ios, | |
129 | unsigned layout_index) | |
130 | { | |
131 | return ios->layout->s_ods[ | |
132 | exofs_layout_od_id(ios->layout, ios->obj.id, layout_index)]; | |
133 | } | |
134 | ||
06886a5a BH |
135 | static void _sync_done(struct exofs_io_state *ios, void *p) |
136 | { | |
137 | struct completion *waiting = p; | |
b14f8ab2 | 138 | |
06886a5a BH |
139 | complete(waiting); |
140 | } | |
141 | ||
142 | static void _last_io(struct kref *kref) | |
143 | { | |
144 | struct exofs_io_state *ios = container_of( | |
145 | kref, struct exofs_io_state, kref); | |
146 | ||
147 | ios->done(ios, ios->private); | |
148 | } | |
149 | ||
150 | static void _done_io(struct osd_request *or, void *p) | |
151 | { | |
152 | struct exofs_io_state *ios = p; | |
153 | ||
154 | kref_put(&ios->kref, _last_io); | |
155 | } | |
156 | ||
157 | static int exofs_io_execute(struct exofs_io_state *ios) | |
158 | { | |
159 | DECLARE_COMPLETION_ONSTACK(wait); | |
160 | bool sync = (ios->done == NULL); | |
161 | int i, ret; | |
162 | ||
163 | if (sync) { | |
164 | ios->done = _sync_done; | |
165 | ios->private = &wait; | |
166 | } | |
167 | ||
168 | for (i = 0; i < ios->numdevs; i++) { | |
169 | struct osd_request *or = ios->per_dev[i].or; | |
170 | if (unlikely(!or)) | |
171 | continue; | |
172 | ||
173 | ret = osd_finalize_request(or, 0, ios->cred, NULL); | |
174 | if (unlikely(ret)) { | |
175 | EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n", | |
176 | ret); | |
177 | return ret; | |
178 | } | |
179 | } | |
180 | ||
181 | kref_init(&ios->kref); | |
182 | ||
183 | for (i = 0; i < ios->numdevs; i++) { | |
184 | struct osd_request *or = ios->per_dev[i].or; | |
185 | if (unlikely(!or)) | |
186 | continue; | |
187 | ||
188 | kref_get(&ios->kref); | |
189 | osd_execute_request_async(or, _done_io, ios); | |
190 | } | |
191 | ||
192 | kref_put(&ios->kref, _last_io); | |
193 | ret = 0; | |
194 | ||
195 | if (sync) { | |
196 | wait_for_completion(&wait); | |
197 | ret = exofs_check_io(ios, NULL); | |
198 | } | |
b14f8ab2 BH |
199 | return ret; |
200 | } | |
201 | ||
22ddc556 BH |
202 | static void _clear_bio(struct bio *bio) |
203 | { | |
204 | struct bio_vec *bv; | |
205 | unsigned i; | |
206 | ||
207 | __bio_for_each_segment(bv, bio, i, 0) { | |
208 | unsigned this_count = bv->bv_len; | |
209 | ||
210 | if (likely(PAGE_SIZE == this_count)) | |
211 | clear_highpage(bv->bv_page); | |
212 | else | |
213 | zero_user(bv->bv_page, bv->bv_offset, this_count); | |
214 | } | |
215 | } | |
216 | ||
06886a5a | 217 | int exofs_check_io(struct exofs_io_state *ios, u64 *resid) |
b14f8ab2 | 218 | { |
06886a5a BH |
219 | enum osd_err_priority acumulated_osd_err = 0; |
220 | int acumulated_lin_err = 0; | |
221 | int i; | |
b14f8ab2 | 222 | |
06886a5a BH |
223 | for (i = 0; i < ios->numdevs; i++) { |
224 | struct osd_sense_info osi; | |
22ddc556 BH |
225 | struct osd_request *or = ios->per_dev[i].or; |
226 | int ret; | |
227 | ||
228 | if (unlikely(!or)) | |
229 | continue; | |
06886a5a | 230 | |
22ddc556 | 231 | ret = osd_req_decode_sense(or, &osi); |
06886a5a BH |
232 | if (likely(!ret)) |
233 | continue; | |
234 | ||
22ddc556 BH |
235 | if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) { |
236 | /* start read offset passed endof file */ | |
237 | _clear_bio(ios->per_dev[i].bio); | |
238 | EXOFS_DBGMSG("start read offset passed end of file " | |
239 | "offset=0x%llx, length=0x%llx\n", | |
5d952b83 BH |
240 | _LLU(ios->per_dev[i].offset), |
241 | _LLU(ios->per_dev[i].length)); | |
22ddc556 BH |
242 | |
243 | continue; /* we recovered */ | |
06886a5a BH |
244 | } |
245 | ||
246 | if (osi.osd_err_pri >= acumulated_osd_err) { | |
247 | acumulated_osd_err = osi.osd_err_pri; | |
248 | acumulated_lin_err = ret; | |
249 | } | |
250 | } | |
251 | ||
252 | /* TODO: raid specific residual calculations */ | |
253 | if (resid) { | |
254 | if (likely(!acumulated_lin_err)) | |
255 | *resid = 0; | |
256 | else | |
257 | *resid = ios->length; | |
258 | } | |
259 | ||
260 | return acumulated_lin_err; | |
261 | } | |
262 | ||
b367e78b BH |
263 | /* |
264 | * L - logical offset into the file | |
265 | * | |
50a76fd3 | 266 | * U - The number of bytes in a stripe within a group |
b367e78b BH |
267 | * |
268 | * U = stripe_unit * group_width | |
269 | * | |
50a76fd3 BH |
270 | * T - The number of bytes striped within a group of component objects |
271 | * (before advancing to the next group) | |
b367e78b | 272 | * |
50a76fd3 BH |
273 | * T = stripe_unit * group_width * group_depth |
274 | * | |
275 | * S - The number of bytes striped across all component objects | |
276 | * before the pattern repeats | |
277 | * | |
278 | * S = stripe_unit * group_width * group_depth * group_count | |
279 | * | |
280 | * M - The "major" (i.e., across all components) stripe number | |
281 | * | |
282 | * M = L / S | |
283 | * | |
284 | * G - Counts the groups from the beginning of the major stripe | |
285 | * | |
286 | * G = (L - (M * S)) / T [or (L % S) / T] | |
287 | * | |
288 | * H - The byte offset within the group | |
289 | * | |
290 | * H = (L - (M * S)) % T [or (L % S) % T] | |
291 | * | |
292 | * N - The "minor" (i.e., across the group) stripe number | |
293 | * | |
294 | * N = H / U | |
b367e78b BH |
295 | * |
296 | * C - The component index coresponding to L | |
297 | * | |
50a76fd3 BH |
298 | * C = (H - (N * U)) / stripe_unit + G * group_width |
299 | * [or (L % U) / stripe_unit + G * group_width] | |
b367e78b BH |
300 | * |
301 | * O - The component offset coresponding to L | |
302 | * | |
50a76fd3 | 303 | * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit |
b367e78b | 304 | */ |
b367e78b BH |
305 | struct _striping_info { |
306 | u64 obj_offset; | |
50a76fd3 BH |
307 | u64 group_length; |
308 | u64 total_group_length; | |
309 | u64 Major; | |
b367e78b BH |
310 | unsigned dev; |
311 | unsigned unit_off; | |
312 | }; | |
313 | ||
314 | static void _calc_stripe_info(struct exofs_io_state *ios, u64 file_offset, | |
315 | struct _striping_info *si) | |
5d952b83 | 316 | { |
b367e78b BH |
317 | u32 stripe_unit = ios->layout->stripe_unit; |
318 | u32 group_width = ios->layout->group_width; | |
50a76fd3 BH |
319 | u64 group_depth = ios->layout->group_depth; |
320 | ||
b367e78b | 321 | u32 U = stripe_unit * group_width; |
50a76fd3 BH |
322 | u64 T = U * group_depth; |
323 | u64 S = T * ios->layout->group_count; | |
324 | u64 M = div64_u64(file_offset, S); | |
325 | ||
326 | /* | |
327 | G = (L - (M * S)) / T | |
328 | H = (L - (M * S)) % T | |
329 | */ | |
330 | u64 LmodS = file_offset - M * S; | |
331 | u32 G = div64_u64(LmodS, T); | |
332 | u64 H = LmodS - G * T; | |
333 | ||
334 | u32 N = div_u64(H, U); | |
335 | ||
336 | /* "H - (N * U)" is just "H % U" so it's bound to u32 */ | |
337 | si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width; | |
338 | si->dev *= ios->layout->mirrors_p1; | |
b367e78b | 339 | |
50a76fd3 | 340 | div_u64_rem(file_offset, stripe_unit, &si->unit_off); |
5d952b83 | 341 | |
50a76fd3 BH |
342 | si->obj_offset = si->unit_off + (N * stripe_unit) + |
343 | (M * group_depth * stripe_unit); | |
344 | ||
345 | si->group_length = T - H; | |
346 | si->total_group_length = T; | |
347 | si->Major = M; | |
5d952b83 BH |
348 | } |
349 | ||
86093aaf BH |
350 | static int _add_stripe_unit(struct exofs_io_state *ios, unsigned *cur_pg, |
351 | unsigned pgbase, struct exofs_per_dev_state *per_dev, | |
352 | int cur_len) | |
5d952b83 | 353 | { |
86093aaf | 354 | unsigned pg = *cur_pg; |
5d952b83 BH |
355 | struct request_queue *q = |
356 | osd_request_queue(exofs_ios_od(ios, per_dev->dev)); | |
357 | ||
358 | per_dev->length += cur_len; | |
359 | ||
360 | if (per_dev->bio == NULL) { | |
361 | unsigned pages_in_stripe = ios->layout->group_width * | |
362 | (ios->layout->stripe_unit / PAGE_SIZE); | |
86093aaf | 363 | unsigned bio_size = (ios->nr_pages + pages_in_stripe) / |
5d952b83 BH |
364 | ios->layout->group_width; |
365 | ||
366 | per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size); | |
367 | if (unlikely(!per_dev->bio)) { | |
368 | EXOFS_DBGMSG("Faild to allocate BIO size=%u\n", | |
369 | bio_size); | |
370 | return -ENOMEM; | |
371 | } | |
372 | } | |
373 | ||
374 | while (cur_len > 0) { | |
86093aaf BH |
375 | unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len); |
376 | unsigned added_len; | |
5d952b83 | 377 | |
86093aaf BH |
378 | BUG_ON(ios->nr_pages <= pg); |
379 | cur_len -= pglen; | |
5d952b83 | 380 | |
86093aaf BH |
381 | added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg], |
382 | pglen, pgbase); | |
383 | if (unlikely(pglen != added_len)) | |
5d952b83 | 384 | return -ENOMEM; |
86093aaf BH |
385 | pgbase = 0; |
386 | ++pg; | |
5d952b83 BH |
387 | } |
388 | BUG_ON(cur_len); | |
389 | ||
86093aaf | 390 | *cur_pg = pg; |
5d952b83 BH |
391 | return 0; |
392 | } | |
393 | ||
50a76fd3 | 394 | static int _prepare_one_group(struct exofs_io_state *ios, u64 length, |
6e31609b | 395 | struct _striping_info *si) |
5d952b83 | 396 | { |
5d952b83 | 397 | unsigned stripe_unit = ios->layout->stripe_unit; |
b367e78b | 398 | unsigned mirrors_p1 = ios->layout->mirrors_p1; |
50a76fd3 | 399 | unsigned devs_in_group = ios->layout->group_width * mirrors_p1; |
b367e78b | 400 | unsigned dev = si->dev; |
50a76fd3 | 401 | unsigned first_dev = dev - (dev % devs_in_group); |
50a76fd3 BH |
402 | unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0; |
403 | unsigned cur_pg = ios->pages_consumed; | |
86093aaf | 404 | int ret = 0; |
5d952b83 | 405 | |
5d952b83 | 406 | while (length) { |
6e31609b | 407 | struct exofs_per_dev_state *per_dev = &ios->per_dev[dev]; |
b367e78b | 408 | unsigned cur_len, page_off = 0; |
5d952b83 BH |
409 | |
410 | if (!per_dev->length) { | |
b367e78b BH |
411 | 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 | } | |
5d952b83 | 425 | |
6e31609b BH |
426 | if (max_comp < dev) |
427 | max_comp = dev; | |
5d952b83 | 428 | } else { |
b367e78b | 429 | cur_len = stripe_unit; |
5d952b83 | 430 | } |
b367e78b BH |
431 | if (cur_len >= length) |
432 | cur_len = length; | |
5d952b83 | 433 | |
86093aaf BH |
434 | ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev, |
435 | cur_len); | |
5d952b83 BH |
436 | if (unlikely(ret)) |
437 | goto out; | |
438 | ||
6e31609b BH |
439 | dev += mirrors_p1; |
440 | dev = (dev % devs_in_group) + first_dev; | |
5d952b83 BH |
441 | |
442 | length -= cur_len; | |
443 | } | |
444 | out: | |
50a76fd3 BH |
445 | ios->numdevs = max_comp + mirrors_p1; |
446 | ios->pages_consumed = cur_pg; | |
5d952b83 BH |
447 | return ret; |
448 | } | |
449 | ||
b367e78b BH |
450 | static int _prepare_for_striping(struct exofs_io_state *ios) |
451 | { | |
50a76fd3 | 452 | u64 length = ios->length; |
b367e78b | 453 | struct _striping_info si; |
50a76fd3 BH |
454 | unsigned devs_in_group = ios->layout->group_width * |
455 | ios->layout->mirrors_p1; | |
50a76fd3 | 456 | int ret = 0; |
b367e78b BH |
457 | |
458 | _calc_stripe_info(ios, ios->offset, &si); | |
459 | ||
460 | if (!ios->pages) { | |
461 | if (ios->kern_buff) { | |
462 | struct exofs_per_dev_state *per_dev = &ios->per_dev[0]; | |
463 | ||
464 | per_dev->offset = si.obj_offset; | |
465 | per_dev->dev = si.dev; | |
466 | ||
467 | /* no cross device without page array */ | |
468 | BUG_ON((ios->layout->group_width > 1) && | |
469 | (si.unit_off + ios->length > | |
470 | ios->layout->stripe_unit)); | |
471 | } | |
472 | ios->numdevs = ios->layout->mirrors_p1; | |
473 | return 0; | |
474 | } | |
475 | ||
50a76fd3 BH |
476 | while (length) { |
477 | if (length < si.group_length) | |
478 | si.group_length = length; | |
479 | ||
6e31609b | 480 | ret = _prepare_one_group(ios, si.group_length, &si); |
50a76fd3 BH |
481 | if (unlikely(ret)) |
482 | goto out; | |
483 | ||
484 | length -= si.group_length; | |
485 | ||
486 | si.group_length = si.total_group_length; | |
487 | si.unit_off = 0; | |
488 | ++si.Major; | |
489 | si.obj_offset = si.Major * ios->layout->stripe_unit * | |
490 | ios->layout->group_depth; | |
491 | ||
492 | si.dev = (si.dev - (si.dev % devs_in_group)) + devs_in_group; | |
493 | si.dev %= ios->layout->s_numdevs; | |
50a76fd3 BH |
494 | } |
495 | ||
496 | out: | |
497 | return ret; | |
b367e78b BH |
498 | } |
499 | ||
06886a5a BH |
500 | int exofs_sbi_create(struct exofs_io_state *ios) |
501 | { | |
502 | int i, ret; | |
503 | ||
45d3abcb | 504 | for (i = 0; i < ios->layout->s_numdevs; i++) { |
06886a5a BH |
505 | struct osd_request *or; |
506 | ||
d9c740d2 | 507 | or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL); |
06886a5a BH |
508 | if (unlikely(!or)) { |
509 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
510 | ret = -ENOMEM; | |
511 | goto out; | |
512 | } | |
513 | ios->per_dev[i].or = or; | |
514 | ios->numdevs++; | |
515 | ||
516 | osd_req_create_object(or, &ios->obj); | |
517 | } | |
518 | ret = exofs_io_execute(ios); | |
519 | ||
520 | out: | |
521 | return ret; | |
522 | } | |
523 | ||
524 | int exofs_sbi_remove(struct exofs_io_state *ios) | |
525 | { | |
526 | int i, ret; | |
527 | ||
45d3abcb | 528 | for (i = 0; i < ios->layout->s_numdevs; i++) { |
06886a5a BH |
529 | struct osd_request *or; |
530 | ||
d9c740d2 | 531 | or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL); |
06886a5a BH |
532 | if (unlikely(!or)) { |
533 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
534 | ret = -ENOMEM; | |
535 | goto out; | |
536 | } | |
537 | ios->per_dev[i].or = or; | |
538 | ios->numdevs++; | |
539 | ||
540 | osd_req_remove_object(or, &ios->obj); | |
541 | } | |
542 | ret = exofs_io_execute(ios); | |
543 | ||
544 | out: | |
545 | return ret; | |
546 | } | |
547 | ||
5d952b83 | 548 | static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp) |
06886a5a | 549 | { |
5d952b83 BH |
550 | struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp]; |
551 | unsigned dev = ios->per_dev[cur_comp].dev; | |
552 | unsigned last_comp = cur_comp + ios->layout->mirrors_p1; | |
553 | int ret = 0; | |
06886a5a | 554 | |
50a76fd3 BH |
555 | if (ios->pages && !master_dev->length) |
556 | return 0; /* Just an empty slot */ | |
557 | ||
5d952b83 BH |
558 | for (; cur_comp < last_comp; ++cur_comp, ++dev) { |
559 | struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp]; | |
06886a5a BH |
560 | struct osd_request *or; |
561 | ||
5d952b83 | 562 | or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL); |
06886a5a BH |
563 | if (unlikely(!or)) { |
564 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
565 | ret = -ENOMEM; | |
566 | goto out; | |
567 | } | |
5d952b83 BH |
568 | per_dev->or = or; |
569 | per_dev->offset = master_dev->offset; | |
06886a5a | 570 | |
86093aaf | 571 | if (ios->pages) { |
06886a5a BH |
572 | struct bio *bio; |
573 | ||
5d952b83 | 574 | if (per_dev != master_dev) { |
04dc1e88 | 575 | bio = bio_kmalloc(GFP_KERNEL, |
5d952b83 | 576 | master_dev->bio->bi_max_vecs); |
04dc1e88 | 577 | if (unlikely(!bio)) { |
34ce4e7c BH |
578 | EXOFS_DBGMSG( |
579 | "Faild to allocate BIO size=%u\n", | |
5d952b83 | 580 | master_dev->bio->bi_max_vecs); |
04dc1e88 BH |
581 | ret = -ENOMEM; |
582 | goto out; | |
583 | } | |
584 | ||
5d952b83 | 585 | __bio_clone(bio, master_dev->bio); |
04dc1e88 BH |
586 | bio->bi_bdev = NULL; |
587 | bio->bi_next = NULL; | |
5d952b83 BH |
588 | per_dev->length = master_dev->length; |
589 | per_dev->bio = bio; | |
590 | per_dev->dev = dev; | |
04dc1e88 | 591 | } else { |
5d952b83 BH |
592 | bio = master_dev->bio; |
593 | /* FIXME: bio_set_dir() */ | |
594 | bio->bi_rw |= (1 << BIO_RW); | |
04dc1e88 | 595 | } |
06886a5a | 596 | |
5d952b83 BH |
597 | osd_req_write(or, &ios->obj, per_dev->offset, bio, |
598 | per_dev->length); | |
34ce4e7c BH |
599 | EXOFS_DBGMSG("write(0x%llx) offset=0x%llx " |
600 | "length=0x%llx dev=%d\n", | |
5d952b83 BH |
601 | _LLU(ios->obj.id), _LLU(per_dev->offset), |
602 | _LLU(per_dev->length), dev); | |
06886a5a | 603 | } else if (ios->kern_buff) { |
5d952b83 | 604 | ret = osd_req_write_kern(or, &ios->obj, per_dev->offset, |
06886a5a | 605 | ios->kern_buff, ios->length); |
5d952b83 BH |
606 | if (unlikely(ret)) |
607 | goto out; | |
34ce4e7c BH |
608 | EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx " |
609 | "length=0x%llx dev=%d\n", | |
5d952b83 BH |
610 | _LLU(ios->obj.id), _LLU(per_dev->offset), |
611 | _LLU(ios->length), dev); | |
06886a5a BH |
612 | } else { |
613 | osd_req_set_attributes(or, &ios->obj); | |
34ce4e7c | 614 | EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n", |
5d952b83 | 615 | _LLU(ios->obj.id), ios->out_attr_len, dev); |
06886a5a BH |
616 | } |
617 | ||
618 | if (ios->out_attr) | |
619 | osd_req_add_set_attr_list(or, ios->out_attr, | |
620 | ios->out_attr_len); | |
621 | ||
622 | if (ios->in_attr) | |
623 | osd_req_add_get_attr_list(or, ios->in_attr, | |
624 | ios->in_attr_len); | |
b14f8ab2 | 625 | } |
06886a5a BH |
626 | |
627 | out: | |
628 | return ret; | |
629 | } | |
630 | ||
5d952b83 BH |
631 | int exofs_sbi_write(struct exofs_io_state *ios) |
632 | { | |
633 | int i; | |
634 | int ret; | |
635 | ||
636 | ret = _prepare_for_striping(ios); | |
637 | if (unlikely(ret)) | |
638 | return ret; | |
639 | ||
640 | for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) { | |
641 | ret = _sbi_write_mirror(ios, i); | |
642 | if (unlikely(ret)) | |
643 | return ret; | |
644 | } | |
645 | ||
646 | ret = exofs_io_execute(ios); | |
647 | return ret; | |
648 | } | |
649 | ||
650 | static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp) | |
06886a5a | 651 | { |
46f4d973 | 652 | struct osd_request *or; |
5d952b83 | 653 | struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp]; |
46f4d973 | 654 | unsigned first_dev = (unsigned)ios->obj.id; |
06886a5a | 655 | |
50a76fd3 BH |
656 | if (ios->pages && !per_dev->length) |
657 | return 0; /* Just an empty slot */ | |
658 | ||
5d952b83 | 659 | first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1; |
d9c740d2 | 660 | or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL); |
46f4d973 BH |
661 | if (unlikely(!or)) { |
662 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
663 | return -ENOMEM; | |
664 | } | |
665 | per_dev->or = or; | |
46f4d973 | 666 | |
86093aaf | 667 | if (ios->pages) { |
5d952b83 BH |
668 | osd_req_read(or, &ios->obj, per_dev->offset, |
669 | per_dev->bio, per_dev->length); | |
46f4d973 BH |
670 | EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx" |
671 | " dev=%d\n", _LLU(ios->obj.id), | |
5d952b83 | 672 | _LLU(per_dev->offset), _LLU(per_dev->length), |
46f4d973 BH |
673 | first_dev); |
674 | } else if (ios->kern_buff) { | |
5d952b83 | 675 | int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset, |
46f4d973 | 676 | ios->kern_buff, ios->length); |
46f4d973 BH |
677 | EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx " |
678 | "length=0x%llx dev=%d ret=>%d\n", | |
5d952b83 | 679 | _LLU(ios->obj.id), _LLU(per_dev->offset), |
46f4d973 BH |
680 | _LLU(ios->length), first_dev, ret); |
681 | if (unlikely(ret)) | |
682 | return ret; | |
683 | } else { | |
684 | osd_req_get_attributes(or, &ios->obj); | |
685 | EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n", | |
686 | _LLU(ios->obj.id), ios->in_attr_len, first_dev); | |
687 | } | |
46f4d973 BH |
688 | if (ios->out_attr) |
689 | osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len); | |
b14f8ab2 | 690 | |
46f4d973 BH |
691 | if (ios->in_attr) |
692 | osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len); | |
b14f8ab2 | 693 | |
5d952b83 BH |
694 | return 0; |
695 | } | |
696 | ||
697 | int exofs_sbi_read(struct exofs_io_state *ios) | |
698 | { | |
699 | int i; | |
700 | int ret; | |
701 | ||
702 | ret = _prepare_for_striping(ios); | |
703 | if (unlikely(ret)) | |
704 | return ret; | |
705 | ||
706 | for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) { | |
707 | ret = _sbi_read_mirror(ios, i); | |
708 | if (unlikely(ret)) | |
709 | return ret; | |
710 | } | |
711 | ||
712 | ret = exofs_io_execute(ios); | |
713 | return ret; | |
b14f8ab2 BH |
714 | } |
715 | ||
06886a5a | 716 | int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr) |
b14f8ab2 BH |
717 | { |
718 | struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */ | |
719 | void *iter = NULL; | |
720 | int nelem; | |
721 | ||
722 | do { | |
723 | nelem = 1; | |
06886a5a BH |
724 | osd_req_decode_get_attr_list(ios->per_dev[0].or, |
725 | &cur_attr, &nelem, &iter); | |
b14f8ab2 BH |
726 | if ((cur_attr.attr_page == attr->attr_page) && |
727 | (cur_attr.attr_id == attr->attr_id)) { | |
728 | attr->len = cur_attr.len; | |
729 | attr->val_ptr = cur_attr.val_ptr; | |
730 | return 0; | |
731 | } | |
732 | } while (iter); | |
733 | ||
734 | return -EIO; | |
735 | } | |
06886a5a | 736 | |
5d952b83 BH |
737 | static int _truncate_mirrors(struct exofs_io_state *ios, unsigned cur_comp, |
738 | struct osd_attr *attr) | |
739 | { | |
740 | int last_comp = cur_comp + ios->layout->mirrors_p1; | |
741 | ||
742 | for (; cur_comp < last_comp; ++cur_comp) { | |
743 | struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp]; | |
744 | struct osd_request *or; | |
745 | ||
746 | or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL); | |
747 | if (unlikely(!or)) { | |
748 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | |
749 | return -ENOMEM; | |
750 | } | |
751 | per_dev->or = or; | |
752 | ||
753 | osd_req_set_attributes(or, &ios->obj); | |
754 | osd_req_add_set_attr_list(or, attr, 1); | |
755 | } | |
756 | ||
757 | return 0; | |
758 | } | |
759 | ||
06886a5a BH |
760 | int exofs_oi_truncate(struct exofs_i_info *oi, u64 size) |
761 | { | |
762 | struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info; | |
763 | struct exofs_io_state *ios; | |
5d952b83 BH |
764 | struct exofs_trunc_attr { |
765 | struct osd_attr attr; | |
766 | __be64 newsize; | |
767 | } *size_attrs; | |
b367e78b | 768 | struct _striping_info si; |
06886a5a BH |
769 | int i, ret; |
770 | ||
5d952b83 BH |
771 | ret = exofs_get_io_state(&sbi->layout, &ios); |
772 | if (unlikely(ret)) | |
773 | return ret; | |
774 | ||
775 | size_attrs = kcalloc(ios->layout->group_width, sizeof(*size_attrs), | |
776 | GFP_KERNEL); | |
777 | if (unlikely(!size_attrs)) { | |
778 | ret = -ENOMEM; | |
779 | goto out; | |
780 | } | |
06886a5a BH |
781 | |
782 | ios->obj.id = exofs_oi_objno(oi); | |
783 | ios->cred = oi->i_cred; | |
784 | ||
5d952b83 | 785 | ios->numdevs = ios->layout->s_numdevs; |
b367e78b | 786 | _calc_stripe_info(ios, size, &si); |
06886a5a | 787 | |
5d952b83 BH |
788 | for (i = 0; i < ios->layout->group_width; ++i) { |
789 | struct exofs_trunc_attr *size_attr = &size_attrs[i]; | |
790 | u64 obj_size; | |
06886a5a | 791 | |
b367e78b BH |
792 | if (i < si.dev) |
793 | obj_size = si.obj_offset + | |
794 | ios->layout->stripe_unit - si.unit_off; | |
795 | else if (i == si.dev) | |
796 | obj_size = si.obj_offset; | |
797 | else /* i > si.dev */ | |
798 | obj_size = si.obj_offset - si.unit_off; | |
06886a5a | 799 | |
5d952b83 BH |
800 | size_attr->newsize = cpu_to_be64(obj_size); |
801 | size_attr->attr = g_attr_logical_length; | |
802 | size_attr->attr.val_ptr = &size_attr->newsize; | |
803 | ||
804 | ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1, | |
805 | &size_attr->attr); | |
806 | if (unlikely(ret)) | |
807 | goto out; | |
06886a5a BH |
808 | } |
809 | ret = exofs_io_execute(ios); | |
810 | ||
811 | out: | |
5d952b83 | 812 | kfree(size_attrs); |
06886a5a BH |
813 | exofs_put_io_state(ios); |
814 | return ret; | |
815 | } |