diff options
Diffstat (limited to 'fs/exofs/ios.c')
-rw-r--r-- | fs/exofs/ios.c | 823 |
1 files changed, 823 insertions, 0 deletions
diff --git a/fs/exofs/ios.c b/fs/exofs/ios.c new file mode 100644 index 000000000000..4337cad7777b --- /dev/null +++ b/fs/exofs/ios.c | |||
@@ -0,0 +1,823 @@ | |||
1 | /* | ||
2 | * Copyright (C) 2005, 2006 | ||
3 | * Avishay Traeger (avishay@gmail.com) | ||
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 | |||
25 | #include <linux/slab.h> | ||
26 | #include <scsi/scsi_device.h> | ||
27 | #include <asm/div64.h> | ||
28 | |||
29 | #include "exofs.h" | ||
30 | |||
31 | #define EXOFS_DBGMSG2(M...) do {} while (0) | ||
32 | /* #define EXOFS_DBGMSG2 EXOFS_DBGMSG */ | ||
33 | |||
34 | void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj) | ||
35 | { | ||
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; | ||
54 | } | ||
55 | |||
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 | } | ||
61 | |||
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); */ | ||
66 | |||
67 | out: | ||
68 | osd_end_request(or); | ||
69 | return ret; | ||
70 | } | ||
71 | |||
72 | int exofs_get_io_state(struct exofs_layout *layout, | ||
73 | struct exofs_io_state **pios) | ||
74 | { | ||
75 | struct exofs_io_state *ios; | ||
76 | |||
77 | /*TODO: Maybe use kmem_cach per sbi of size | ||
78 | * exofs_io_state_size(layout->s_numdevs) | ||
79 | */ | ||
80 | ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL); | ||
81 | if (unlikely(!ios)) { | ||
82 | EXOFS_DBGMSG("Faild kzalloc bytes=%d\n", | ||
83 | exofs_io_state_size(layout->s_numdevs)); | ||
84 | *pios = NULL; | ||
85 | return -ENOMEM; | ||
86 | } | ||
87 | |||
88 | ios->layout = layout; | ||
89 | ios->obj.partition = layout->s_pid; | ||
90 | *pios = ios; | ||
91 | return 0; | ||
92 | } | ||
93 | |||
94 | void exofs_put_io_state(struct exofs_io_state *ios) | ||
95 | { | ||
96 | if (ios) { | ||
97 | unsigned i; | ||
98 | |||
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); | ||
109 | } | ||
110 | } | ||
111 | |||
112 | unsigned exofs_layout_od_id(struct exofs_layout *layout, | ||
113 | osd_id obj_no, unsigned layout_index) | ||
114 | { | ||
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 | }*/ | ||
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 | |||
135 | static void _sync_done(struct exofs_io_state *ios, void *p) | ||
136 | { | ||
137 | struct completion *waiting = p; | ||
138 | |||
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 | } | ||
199 | return ret; | ||
200 | } | ||
201 | |||
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 | |||
217 | int exofs_check_io(struct exofs_io_state *ios, u64 *resid) | ||
218 | { | ||
219 | enum osd_err_priority acumulated_osd_err = 0; | ||
220 | int acumulated_lin_err = 0; | ||
221 | int i; | ||
222 | |||
223 | for (i = 0; i < ios->numdevs; i++) { | ||
224 | struct osd_sense_info osi; | ||
225 | struct osd_request *or = ios->per_dev[i].or; | ||
226 | int ret; | ||
227 | |||
228 | if (unlikely(!or)) | ||
229 | continue; | ||
230 | |||
231 | ret = osd_req_decode_sense(or, &osi); | ||
232 | if (likely(!ret)) | ||
233 | continue; | ||
234 | |||
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", | ||
240 | _LLU(ios->per_dev[i].offset), | ||
241 | _LLU(ios->per_dev[i].length)); | ||
242 | |||
243 | continue; /* we recovered */ | ||
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 | |||
263 | /* | ||
264 | * L - logical offset into the file | ||
265 | * | ||
266 | * U - The number of bytes in a stripe within a group | ||
267 | * | ||
268 | * U = stripe_unit * group_width | ||
269 | * | ||
270 | * T - The number of bytes striped within a group of component objects | ||
271 | * (before advancing to the next group) | ||
272 | * | ||
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 | ||
295 | * | ||
296 | * C - The component index coresponding to L | ||
297 | * | ||
298 | * C = (H - (N * U)) / stripe_unit + G * group_width | ||
299 | * [or (L % U) / stripe_unit + G * group_width] | ||
300 | * | ||
301 | * O - The component offset coresponding to L | ||
302 | * | ||
303 | * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit | ||
304 | */ | ||
305 | struct _striping_info { | ||
306 | u64 obj_offset; | ||
307 | u64 group_length; | ||
308 | u64 total_group_length; | ||
309 | u64 Major; | ||
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) | ||
316 | { | ||
317 | u32 stripe_unit = ios->layout->stripe_unit; | ||
318 | u32 group_width = ios->layout->group_width; | ||
319 | u64 group_depth = ios->layout->group_depth; | ||
320 | |||
321 | u32 U = stripe_unit * group_width; | ||
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; | ||
339 | |||
340 | div_u64_rem(file_offset, stripe_unit, &si->unit_off); | ||
341 | |||
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; | ||
348 | } | ||
349 | |||
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) | ||
353 | { | ||
354 | unsigned pg = *cur_pg; | ||
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); | ||
363 | unsigned bio_size = (ios->nr_pages + pages_in_stripe) / | ||
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) { | ||
375 | unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len); | ||
376 | unsigned added_len; | ||
377 | |||
378 | BUG_ON(ios->nr_pages <= pg); | ||
379 | cur_len -= pglen; | ||
380 | |||
381 | added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg], | ||
382 | pglen, pgbase); | ||
383 | if (unlikely(pglen != added_len)) | ||
384 | return -ENOMEM; | ||
385 | pgbase = 0; | ||
386 | ++pg; | ||
387 | } | ||
388 | BUG_ON(cur_len); | ||
389 | |||
390 | *cur_pg = pg; | ||
391 | return 0; | ||
392 | } | ||
393 | |||
394 | static int _prepare_one_group(struct exofs_io_state *ios, u64 length, | ||
395 | struct _striping_info *si, unsigned first_comp) | ||
396 | { | ||
397 | unsigned stripe_unit = ios->layout->stripe_unit; | ||
398 | unsigned mirrors_p1 = ios->layout->mirrors_p1; | ||
399 | unsigned devs_in_group = ios->layout->group_width * mirrors_p1; | ||
400 | unsigned dev = si->dev; | ||
401 | unsigned first_dev = dev - (dev % devs_in_group); | ||
402 | unsigned comp = first_comp + (dev - first_dev); | ||
403 | unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0; | ||
404 | unsigned cur_pg = ios->pages_consumed; | ||
405 | int ret = 0; | ||
406 | |||
407 | while (length) { | ||
408 | struct exofs_per_dev_state *per_dev = &ios->per_dev[comp]; | ||
409 | unsigned cur_len, page_off = 0; | ||
410 | |||
411 | if (!per_dev->length) { | ||
412 | per_dev->dev = dev; | ||
413 | if (dev < si->dev) { | ||
414 | per_dev->offset = si->obj_offset + stripe_unit - | ||
415 | si->unit_off; | ||
416 | cur_len = stripe_unit; | ||
417 | } else if (dev == si->dev) { | ||
418 | per_dev->offset = si->obj_offset; | ||
419 | cur_len = stripe_unit - si->unit_off; | ||
420 | page_off = si->unit_off & ~PAGE_MASK; | ||
421 | BUG_ON(page_off && (page_off != ios->pgbase)); | ||
422 | } else { /* dev > si->dev */ | ||
423 | per_dev->offset = si->obj_offset - si->unit_off; | ||
424 | cur_len = stripe_unit; | ||
425 | } | ||
426 | |||
427 | if (max_comp < comp) | ||
428 | max_comp = comp; | ||
429 | |||
430 | dev += mirrors_p1; | ||
431 | dev = (dev % devs_in_group) + first_dev; | ||
432 | } else { | ||
433 | cur_len = stripe_unit; | ||
434 | } | ||
435 | if (cur_len >= length) | ||
436 | cur_len = length; | ||
437 | |||
438 | ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev, | ||
439 | cur_len); | ||
440 | if (unlikely(ret)) | ||
441 | goto out; | ||
442 | |||
443 | comp += mirrors_p1; | ||
444 | comp = (comp % devs_in_group) + first_comp; | ||
445 | |||
446 | length -= cur_len; | ||
447 | } | ||
448 | out: | ||
449 | ios->numdevs = max_comp + mirrors_p1; | ||
450 | ios->pages_consumed = cur_pg; | ||
451 | return ret; | ||
452 | } | ||
453 | |||
454 | static int _prepare_for_striping(struct exofs_io_state *ios) | ||
455 | { | ||
456 | u64 length = ios->length; | ||
457 | struct _striping_info si; | ||
458 | unsigned devs_in_group = ios->layout->group_width * | ||
459 | ios->layout->mirrors_p1; | ||
460 | unsigned first_comp = 0; | ||
461 | int ret = 0; | ||
462 | |||
463 | _calc_stripe_info(ios, ios->offset, &si); | ||
464 | |||
465 | if (!ios->pages) { | ||
466 | if (ios->kern_buff) { | ||
467 | struct exofs_per_dev_state *per_dev = &ios->per_dev[0]; | ||
468 | |||
469 | per_dev->offset = si.obj_offset; | ||
470 | per_dev->dev = si.dev; | ||
471 | |||
472 | /* no cross device without page array */ | ||
473 | BUG_ON((ios->layout->group_width > 1) && | ||
474 | (si.unit_off + ios->length > | ||
475 | ios->layout->stripe_unit)); | ||
476 | } | ||
477 | ios->numdevs = ios->layout->mirrors_p1; | ||
478 | return 0; | ||
479 | } | ||
480 | |||
481 | while (length) { | ||
482 | if (length < si.group_length) | ||
483 | si.group_length = length; | ||
484 | |||
485 | ret = _prepare_one_group(ios, si.group_length, &si, first_comp); | ||
486 | if (unlikely(ret)) | ||
487 | goto out; | ||
488 | |||
489 | length -= si.group_length; | ||
490 | |||
491 | si.group_length = si.total_group_length; | ||
492 | si.unit_off = 0; | ||
493 | ++si.Major; | ||
494 | si.obj_offset = si.Major * ios->layout->stripe_unit * | ||
495 | ios->layout->group_depth; | ||
496 | |||
497 | si.dev = (si.dev - (si.dev % devs_in_group)) + devs_in_group; | ||
498 | si.dev %= ios->layout->s_numdevs; | ||
499 | |||
500 | first_comp += devs_in_group; | ||
501 | first_comp %= ios->layout->s_numdevs; | ||
502 | } | ||
503 | |||
504 | out: | ||
505 | return ret; | ||
506 | } | ||
507 | |||
508 | int exofs_sbi_create(struct exofs_io_state *ios) | ||
509 | { | ||
510 | int i, ret; | ||
511 | |||
512 | for (i = 0; i < ios->layout->s_numdevs; i++) { | ||
513 | struct osd_request *or; | ||
514 | |||
515 | or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL); | ||
516 | if (unlikely(!or)) { | ||
517 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | ||
518 | ret = -ENOMEM; | ||
519 | goto out; | ||
520 | } | ||
521 | ios->per_dev[i].or = or; | ||
522 | ios->numdevs++; | ||
523 | |||
524 | osd_req_create_object(or, &ios->obj); | ||
525 | } | ||
526 | ret = exofs_io_execute(ios); | ||
527 | |||
528 | out: | ||
529 | return ret; | ||
530 | } | ||
531 | |||
532 | int exofs_sbi_remove(struct exofs_io_state *ios) | ||
533 | { | ||
534 | int i, ret; | ||
535 | |||
536 | for (i = 0; i < ios->layout->s_numdevs; i++) { | ||
537 | struct osd_request *or; | ||
538 | |||
539 | or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL); | ||
540 | if (unlikely(!or)) { | ||
541 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | ||
542 | ret = -ENOMEM; | ||
543 | goto out; | ||
544 | } | ||
545 | ios->per_dev[i].or = or; | ||
546 | ios->numdevs++; | ||
547 | |||
548 | osd_req_remove_object(or, &ios->obj); | ||
549 | } | ||
550 | ret = exofs_io_execute(ios); | ||
551 | |||
552 | out: | ||
553 | return ret; | ||
554 | } | ||
555 | |||
556 | static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp) | ||
557 | { | ||
558 | struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp]; | ||
559 | unsigned dev = ios->per_dev[cur_comp].dev; | ||
560 | unsigned last_comp = cur_comp + ios->layout->mirrors_p1; | ||
561 | int ret = 0; | ||
562 | |||
563 | if (ios->pages && !master_dev->length) | ||
564 | return 0; /* Just an empty slot */ | ||
565 | |||
566 | for (; cur_comp < last_comp; ++cur_comp, ++dev) { | ||
567 | struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp]; | ||
568 | struct osd_request *or; | ||
569 | |||
570 | or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL); | ||
571 | if (unlikely(!or)) { | ||
572 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | ||
573 | ret = -ENOMEM; | ||
574 | goto out; | ||
575 | } | ||
576 | per_dev->or = or; | ||
577 | per_dev->offset = master_dev->offset; | ||
578 | |||
579 | if (ios->pages) { | ||
580 | struct bio *bio; | ||
581 | |||
582 | if (per_dev != master_dev) { | ||
583 | bio = bio_kmalloc(GFP_KERNEL, | ||
584 | master_dev->bio->bi_max_vecs); | ||
585 | if (unlikely(!bio)) { | ||
586 | EXOFS_DBGMSG( | ||
587 | "Faild to allocate BIO size=%u\n", | ||
588 | master_dev->bio->bi_max_vecs); | ||
589 | ret = -ENOMEM; | ||
590 | goto out; | ||
591 | } | ||
592 | |||
593 | __bio_clone(bio, master_dev->bio); | ||
594 | bio->bi_bdev = NULL; | ||
595 | bio->bi_next = NULL; | ||
596 | per_dev->length = master_dev->length; | ||
597 | per_dev->bio = bio; | ||
598 | per_dev->dev = dev; | ||
599 | } else { | ||
600 | bio = master_dev->bio; | ||
601 | /* FIXME: bio_set_dir() */ | ||
602 | bio->bi_rw |= (1 << BIO_RW); | ||
603 | } | ||
604 | |||
605 | osd_req_write(or, &ios->obj, per_dev->offset, bio, | ||
606 | per_dev->length); | ||
607 | EXOFS_DBGMSG("write(0x%llx) offset=0x%llx " | ||
608 | "length=0x%llx dev=%d\n", | ||
609 | _LLU(ios->obj.id), _LLU(per_dev->offset), | ||
610 | _LLU(per_dev->length), dev); | ||
611 | } else if (ios->kern_buff) { | ||
612 | ret = osd_req_write_kern(or, &ios->obj, per_dev->offset, | ||
613 | ios->kern_buff, ios->length); | ||
614 | if (unlikely(ret)) | ||
615 | goto out; | ||
616 | EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx " | ||
617 | "length=0x%llx dev=%d\n", | ||
618 | _LLU(ios->obj.id), _LLU(per_dev->offset), | ||
619 | _LLU(ios->length), dev); | ||
620 | } else { | ||
621 | osd_req_set_attributes(or, &ios->obj); | ||
622 | EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n", | ||
623 | _LLU(ios->obj.id), ios->out_attr_len, dev); | ||
624 | } | ||
625 | |||
626 | if (ios->out_attr) | ||
627 | osd_req_add_set_attr_list(or, ios->out_attr, | ||
628 | ios->out_attr_len); | ||
629 | |||
630 | if (ios->in_attr) | ||
631 | osd_req_add_get_attr_list(or, ios->in_attr, | ||
632 | ios->in_attr_len); | ||
633 | } | ||
634 | |||
635 | out: | ||
636 | return ret; | ||
637 | } | ||
638 | |||
639 | int exofs_sbi_write(struct exofs_io_state *ios) | ||
640 | { | ||
641 | int i; | ||
642 | int ret; | ||
643 | |||
644 | ret = _prepare_for_striping(ios); | ||
645 | if (unlikely(ret)) | ||
646 | return ret; | ||
647 | |||
648 | for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) { | ||
649 | ret = _sbi_write_mirror(ios, i); | ||
650 | if (unlikely(ret)) | ||
651 | return ret; | ||
652 | } | ||
653 | |||
654 | ret = exofs_io_execute(ios); | ||
655 | return ret; | ||
656 | } | ||
657 | |||
658 | static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp) | ||
659 | { | ||
660 | struct osd_request *or; | ||
661 | struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp]; | ||
662 | unsigned first_dev = (unsigned)ios->obj.id; | ||
663 | |||
664 | if (ios->pages && !per_dev->length) | ||
665 | return 0; /* Just an empty slot */ | ||
666 | |||
667 | first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1; | ||
668 | or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL); | ||
669 | if (unlikely(!or)) { | ||
670 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | ||
671 | return -ENOMEM; | ||
672 | } | ||
673 | per_dev->or = or; | ||
674 | |||
675 | if (ios->pages) { | ||
676 | osd_req_read(or, &ios->obj, per_dev->offset, | ||
677 | per_dev->bio, per_dev->length); | ||
678 | EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx" | ||
679 | " dev=%d\n", _LLU(ios->obj.id), | ||
680 | _LLU(per_dev->offset), _LLU(per_dev->length), | ||
681 | first_dev); | ||
682 | } else if (ios->kern_buff) { | ||
683 | int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset, | ||
684 | ios->kern_buff, ios->length); | ||
685 | EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx " | ||
686 | "length=0x%llx dev=%d ret=>%d\n", | ||
687 | _LLU(ios->obj.id), _LLU(per_dev->offset), | ||
688 | _LLU(ios->length), first_dev, ret); | ||
689 | if (unlikely(ret)) | ||
690 | return ret; | ||
691 | } else { | ||
692 | osd_req_get_attributes(or, &ios->obj); | ||
693 | EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n", | ||
694 | _LLU(ios->obj.id), ios->in_attr_len, first_dev); | ||
695 | } | ||
696 | if (ios->out_attr) | ||
697 | osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len); | ||
698 | |||
699 | if (ios->in_attr) | ||
700 | osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len); | ||
701 | |||
702 | return 0; | ||
703 | } | ||
704 | |||
705 | int exofs_sbi_read(struct exofs_io_state *ios) | ||
706 | { | ||
707 | int i; | ||
708 | int ret; | ||
709 | |||
710 | ret = _prepare_for_striping(ios); | ||
711 | if (unlikely(ret)) | ||
712 | return ret; | ||
713 | |||
714 | for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) { | ||
715 | ret = _sbi_read_mirror(ios, i); | ||
716 | if (unlikely(ret)) | ||
717 | return ret; | ||
718 | } | ||
719 | |||
720 | ret = exofs_io_execute(ios); | ||
721 | return ret; | ||
722 | } | ||
723 | |||
724 | int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr) | ||
725 | { | ||
726 | struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */ | ||
727 | void *iter = NULL; | ||
728 | int nelem; | ||
729 | |||
730 | do { | ||
731 | nelem = 1; | ||
732 | osd_req_decode_get_attr_list(ios->per_dev[0].or, | ||
733 | &cur_attr, &nelem, &iter); | ||
734 | if ((cur_attr.attr_page == attr->attr_page) && | ||
735 | (cur_attr.attr_id == attr->attr_id)) { | ||
736 | attr->len = cur_attr.len; | ||
737 | attr->val_ptr = cur_attr.val_ptr; | ||
738 | return 0; | ||
739 | } | ||
740 | } while (iter); | ||
741 | |||
742 | return -EIO; | ||
743 | } | ||
744 | |||
745 | static int _truncate_mirrors(struct exofs_io_state *ios, unsigned cur_comp, | ||
746 | struct osd_attr *attr) | ||
747 | { | ||
748 | int last_comp = cur_comp + ios->layout->mirrors_p1; | ||
749 | |||
750 | for (; cur_comp < last_comp; ++cur_comp) { | ||
751 | struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp]; | ||
752 | struct osd_request *or; | ||
753 | |||
754 | or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL); | ||
755 | if (unlikely(!or)) { | ||
756 | EXOFS_ERR("%s: osd_start_request failed\n", __func__); | ||
757 | return -ENOMEM; | ||
758 | } | ||
759 | per_dev->or = or; | ||
760 | |||
761 | osd_req_set_attributes(or, &ios->obj); | ||
762 | osd_req_add_set_attr_list(or, attr, 1); | ||
763 | } | ||
764 | |||
765 | return 0; | ||
766 | } | ||
767 | |||
768 | int exofs_oi_truncate(struct exofs_i_info *oi, u64 size) | ||
769 | { | ||
770 | struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info; | ||
771 | struct exofs_io_state *ios; | ||
772 | struct exofs_trunc_attr { | ||
773 | struct osd_attr attr; | ||
774 | __be64 newsize; | ||
775 | } *size_attrs; | ||
776 | struct _striping_info si; | ||
777 | int i, ret; | ||
778 | |||
779 | ret = exofs_get_io_state(&sbi->layout, &ios); | ||
780 | if (unlikely(ret)) | ||
781 | return ret; | ||
782 | |||
783 | size_attrs = kcalloc(ios->layout->group_width, sizeof(*size_attrs), | ||
784 | GFP_KERNEL); | ||
785 | if (unlikely(!size_attrs)) { | ||
786 | ret = -ENOMEM; | ||
787 | goto out; | ||
788 | } | ||
789 | |||
790 | ios->obj.id = exofs_oi_objno(oi); | ||
791 | ios->cred = oi->i_cred; | ||
792 | |||
793 | ios->numdevs = ios->layout->s_numdevs; | ||
794 | _calc_stripe_info(ios, size, &si); | ||
795 | |||
796 | for (i = 0; i < ios->layout->group_width; ++i) { | ||
797 | struct exofs_trunc_attr *size_attr = &size_attrs[i]; | ||
798 | u64 obj_size; | ||
799 | |||
800 | if (i < si.dev) | ||
801 | obj_size = si.obj_offset + | ||
802 | ios->layout->stripe_unit - si.unit_off; | ||
803 | else if (i == si.dev) | ||
804 | obj_size = si.obj_offset; | ||
805 | else /* i > si.dev */ | ||
806 | obj_size = si.obj_offset - si.unit_off; | ||
807 | |||
808 | size_attr->newsize = cpu_to_be64(obj_size); | ||
809 | size_attr->attr = g_attr_logical_length; | ||
810 | size_attr->attr.val_ptr = &size_attr->newsize; | ||
811 | |||
812 | ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1, | ||
813 | &size_attr->attr); | ||
814 | if (unlikely(ret)) | ||
815 | goto out; | ||
816 | } | ||
817 | ret = exofs_io_execute(ios); | ||
818 | |||
819 | out: | ||
820 | kfree(size_attrs); | ||
821 | exofs_put_io_state(ios); | ||
822 | return ret; | ||
823 | } | ||