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> |
143cb494 | 26 | #include <linux/module.h> |
5d952b83 | 27 | #include <asm/div64.h> |
a1fec1db | 28 | #include <linux/lcm.h> |
b14f8ab2 | 29 | |
a1fec1db | 30 | #include "ore_raid.h" |
8ff660ab | 31 | |
cf283ade BH |
32 | MODULE_AUTHOR("Boaz Harrosh <bharrosh@panasas.com>"); |
33 | MODULE_DESCRIPTION("Objects Raid Engine ore.ko"); | |
34 | MODULE_LICENSE("GPL"); | |
35 | ||
5a51c0c7 BH |
36 | /* ore_verify_layout does a couple of things: |
37 | * 1. Given a minimum number of needed parameters fixes up the rest of the | |
38 | * members to be operatonals for the ore. The needed parameters are those | |
39 | * that are defined by the pnfs-objects layout STD. | |
40 | * 2. Check to see if the current ore code actually supports these parameters | |
41 | * for example stripe_unit must be a multple of the system PAGE_SIZE, | |
42 | * and etc... | |
43 | * 3. Cache some havily used calculations that will be needed by users. | |
44 | */ | |
45 | ||
5a51c0c7 BH |
46 | enum { BIO_MAX_PAGES_KMALLOC = |
47 | (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),}; | |
48 | ||
49 | int ore_verify_layout(unsigned total_comps, struct ore_layout *layout) | |
50 | { | |
51 | u64 stripe_length; | |
52 | ||
44231e68 BH |
53 | switch (layout->raid_algorithm) { |
54 | case PNFS_OSD_RAID_0: | |
55 | layout->parity = 0; | |
56 | break; | |
57 | case PNFS_OSD_RAID_5: | |
58 | layout->parity = 1; | |
59 | break; | |
60 | case PNFS_OSD_RAID_PQ: | |
61 | case PNFS_OSD_RAID_4: | |
62 | default: | |
63 | ORE_ERR("Only RAID_0/5 for now\n"); | |
5a51c0c7 BH |
64 | return -EINVAL; |
65 | } | |
66 | if (0 != (layout->stripe_unit & ~PAGE_MASK)) { | |
67 | ORE_ERR("Stripe Unit(0x%llx)" | |
68 | " must be Multples of PAGE_SIZE(0x%lx)\n", | |
69 | _LLU(layout->stripe_unit), PAGE_SIZE); | |
70 | return -EINVAL; | |
71 | } | |
72 | if (layout->group_width) { | |
73 | if (!layout->group_depth) { | |
74 | ORE_ERR("group_depth == 0 && group_width != 0\n"); | |
75 | return -EINVAL; | |
76 | } | |
77 | if (total_comps < (layout->group_width * layout->mirrors_p1)) { | |
78 | ORE_ERR("Data Map wrong, " | |
79 | "numdevs=%d < group_width=%d * mirrors=%d\n", | |
80 | total_comps, layout->group_width, | |
81 | layout->mirrors_p1); | |
82 | return -EINVAL; | |
83 | } | |
84 | layout->group_count = total_comps / layout->mirrors_p1 / | |
85 | layout->group_width; | |
86 | } else { | |
87 | if (layout->group_depth) { | |
88 | printk(KERN_NOTICE "Warning: group_depth ignored " | |
89 | "group_width == 0 && group_depth == %lld\n", | |
90 | _LLU(layout->group_depth)); | |
91 | } | |
92 | layout->group_width = total_comps / layout->mirrors_p1; | |
93 | layout->group_depth = -1; | |
94 | layout->group_count = 1; | |
95 | } | |
96 | ||
97 | stripe_length = (u64)layout->group_width * layout->stripe_unit; | |
98 | if (stripe_length >= (1ULL << 32)) { | |
99 | ORE_ERR("Stripe_length(0x%llx) >= 32bit is not supported\n", | |
100 | _LLU(stripe_length)); | |
101 | return -EINVAL; | |
102 | } | |
103 | ||
104 | layout->max_io_length = | |
105 | (BIO_MAX_PAGES_KMALLOC * PAGE_SIZE - layout->stripe_unit) * | |
106 | layout->group_width; | |
769ba8d9 BH |
107 | if (layout->parity) { |
108 | unsigned stripe_length = | |
109 | (layout->group_width - layout->parity) * | |
110 | layout->stripe_unit; | |
111 | ||
112 | layout->max_io_length /= stripe_length; | |
113 | layout->max_io_length *= stripe_length; | |
114 | } | |
5a51c0c7 BH |
115 | return 0; |
116 | } | |
117 | EXPORT_SYMBOL(ore_verify_layout); | |
118 | ||
8ff660ab | 119 | static u8 *_ios_cred(struct ore_io_state *ios, unsigned index) |
9e9db456 | 120 | { |
5bf696da | 121 | return ios->oc->comps[index & ios->oc->single_comp].cred; |
9e9db456 BH |
122 | } |
123 | ||
8ff660ab | 124 | static struct osd_obj_id *_ios_obj(struct ore_io_state *ios, unsigned index) |
9e9db456 | 125 | { |
5bf696da | 126 | return &ios->oc->comps[index & ios->oc->single_comp].obj; |
9e9db456 BH |
127 | } |
128 | ||
8ff660ab | 129 | static struct osd_dev *_ios_od(struct ore_io_state *ios, unsigned index) |
9e9db456 | 130 | { |
3bd98568 BH |
131 | ORE_DBGMSG2("oc->first_dev=%d oc->numdevs=%d i=%d oc->ods=%p\n", |
132 | ios->oc->first_dev, ios->oc->numdevs, index, | |
133 | ios->oc->ods); | |
134 | ||
d866d875 | 135 | return ore_comp_dev(ios->oc, index); |
9e9db456 BH |
136 | } |
137 | ||
769ba8d9 | 138 | int _ore_get_io_state(struct ore_layout *layout, |
a1fec1db BH |
139 | struct ore_components *oc, unsigned numdevs, |
140 | unsigned sgs_per_dev, unsigned num_par_pages, | |
141 | struct ore_io_state **pios) | |
b14f8ab2 | 142 | { |
8ff660ab | 143 | struct ore_io_state *ios; |
a1fec1db BH |
144 | struct page **pages; |
145 | struct osd_sg_entry *sgilist; | |
146 | struct __alloc_all_io_state { | |
147 | struct ore_io_state ios; | |
148 | struct ore_per_dev_state per_dev[numdevs]; | |
149 | union { | |
150 | struct osd_sg_entry sglist[sgs_per_dev * numdevs]; | |
151 | struct page *pages[num_par_pages]; | |
152 | }; | |
153 | } *_aios; | |
154 | ||
155 | if (likely(sizeof(*_aios) <= PAGE_SIZE)) { | |
156 | _aios = kzalloc(sizeof(*_aios), GFP_KERNEL); | |
157 | if (unlikely(!_aios)) { | |
158 | ORE_DBGMSG("Failed kzalloc bytes=%zd\n", | |
159 | sizeof(*_aios)); | |
160 | *pios = NULL; | |
161 | return -ENOMEM; | |
162 | } | |
163 | pages = num_par_pages ? _aios->pages : NULL; | |
164 | sgilist = sgs_per_dev ? _aios->sglist : NULL; | |
165 | ios = &_aios->ios; | |
166 | } else { | |
167 | struct __alloc_small_io_state { | |
168 | struct ore_io_state ios; | |
169 | struct ore_per_dev_state per_dev[numdevs]; | |
170 | } *_aio_small; | |
171 | union __extra_part { | |
172 | struct osd_sg_entry sglist[sgs_per_dev * numdevs]; | |
173 | struct page *pages[num_par_pages]; | |
174 | } *extra_part; | |
175 | ||
176 | _aio_small = kzalloc(sizeof(*_aio_small), GFP_KERNEL); | |
177 | if (unlikely(!_aio_small)) { | |
178 | ORE_DBGMSG("Failed alloc first part bytes=%zd\n", | |
179 | sizeof(*_aio_small)); | |
180 | *pios = NULL; | |
181 | return -ENOMEM; | |
182 | } | |
183 | extra_part = kzalloc(sizeof(*extra_part), GFP_KERNEL); | |
184 | if (unlikely(!extra_part)) { | |
185 | ORE_DBGMSG("Failed alloc second part bytes=%zd\n", | |
186 | sizeof(*extra_part)); | |
187 | kfree(_aio_small); | |
188 | *pios = NULL; | |
189 | return -ENOMEM; | |
190 | } | |
06886a5a | 191 | |
a1fec1db BH |
192 | pages = num_par_pages ? extra_part->pages : NULL; |
193 | sgilist = sgs_per_dev ? extra_part->sglist : NULL; | |
194 | /* In this case the per_dev[0].sgilist holds the pointer to | |
195 | * be freed | |
196 | */ | |
197 | ios = &_aio_small->ios; | |
198 | ios->extra_part_alloc = true; | |
199 | } | |
200 | ||
201 | if (pages) { | |
202 | ios->parity_pages = pages; | |
203 | ios->max_par_pages = num_par_pages; | |
204 | } | |
205 | if (sgilist) { | |
206 | unsigned d; | |
207 | ||
208 | for (d = 0; d < numdevs; ++d) { | |
209 | ios->per_dev[d].sglist = sgilist; | |
210 | sgilist += sgs_per_dev; | |
211 | } | |
212 | ios->sgs_per_dev = sgs_per_dev; | |
06886a5a BH |
213 | } |
214 | ||
45d3abcb | 215 | ios->layout = layout; |
5bf696da | 216 | ios->oc = oc; |
b916c5cd BH |
217 | *pios = ios; |
218 | return 0; | |
219 | } | |
220 | ||
221 | /* Allocate an io_state for only a single group of devices | |
222 | * | |
223 | * If a user needs to call ore_read/write() this version must be used becase it | |
224 | * allocates extra stuff for striping and raid. | |
225 | * The ore might decide to only IO less then @length bytes do to alignmets | |
226 | * and constrains as follows: | |
227 | * - The IO cannot cross group boundary. | |
228 | * - In raid5/6 The end of the IO must align at end of a stripe eg. | |
229 | * (@offset + @length) % strip_size == 0. Or the complete range is within a | |
230 | * single stripe. | |
231 | * - Memory condition only permitted a shorter IO. (A user can use @length=~0 | |
232 | * And check the returned ios->length for max_io_size.) | |
233 | * | |
234 | * The caller must check returned ios->length (and/or ios->nr_pages) and | |
235 | * re-issue these pages that fall outside of ios->length | |
236 | */ | |
237 | int ore_get_rw_state(struct ore_layout *layout, struct ore_components *oc, | |
238 | bool is_reading, u64 offset, u64 length, | |
239 | struct ore_io_state **pios) | |
240 | { | |
241 | struct ore_io_state *ios; | |
242 | unsigned numdevs = layout->group_width * layout->mirrors_p1; | |
a1fec1db | 243 | unsigned sgs_per_dev = 0, max_par_pages = 0; |
b916c5cd BH |
244 | int ret; |
245 | ||
a1fec1db BH |
246 | if (layout->parity && length) { |
247 | unsigned data_devs = layout->group_width - layout->parity; | |
248 | unsigned stripe_size = layout->stripe_unit * data_devs; | |
249 | unsigned pages_in_unit = layout->stripe_unit / PAGE_SIZE; | |
250 | u32 remainder; | |
251 | u64 num_stripes; | |
252 | u64 num_raid_units; | |
253 | ||
254 | num_stripes = div_u64_rem(length, stripe_size, &remainder); | |
255 | if (remainder) | |
256 | ++num_stripes; | |
257 | ||
258 | num_raid_units = num_stripes * layout->parity; | |
259 | ||
260 | if (is_reading) { | |
261 | /* For reads add per_dev sglist array */ | |
262 | /* TODO: Raid 6 we need twice more. Actually: | |
263 | * num_stripes / LCMdP(W,P); | |
264 | * if (W%P != 0) num_stripes *= parity; | |
265 | */ | |
266 | ||
267 | /* first/last seg is split */ | |
268 | num_raid_units += layout->group_width; | |
269 | sgs_per_dev = div_u64(num_raid_units, data_devs); | |
270 | } else { | |
271 | /* For Writes add parity pages array. */ | |
272 | max_par_pages = num_raid_units * pages_in_unit * | |
273 | sizeof(struct page *); | |
274 | } | |
275 | } | |
276 | ||
277 | ret = _ore_get_io_state(layout, oc, numdevs, sgs_per_dev, max_par_pages, | |
278 | pios); | |
b916c5cd BH |
279 | if (unlikely(ret)) |
280 | return ret; | |
281 | ||
282 | ios = *pios; | |
e1042ba0 | 283 | ios->reading = is_reading; |
b916c5cd BH |
284 | ios->offset = offset; |
285 | ||
286 | if (length) { | |
a1fec1db BH |
287 | ore_calc_stripe_info(layout, offset, length, &ios->si); |
288 | ios->length = ios->si.length; | |
b916c5cd | 289 | ios->nr_pages = (ios->length + PAGE_SIZE - 1) / PAGE_SIZE; |
a1fec1db BH |
290 | if (layout->parity) |
291 | _ore_post_alloc_raid_stuff(ios); | |
b916c5cd | 292 | } |
e1042ba0 | 293 | |
06886a5a | 294 | return 0; |
b14f8ab2 | 295 | } |
cf283ade | 296 | EXPORT_SYMBOL(ore_get_rw_state); |
b14f8ab2 | 297 | |
b916c5cd BH |
298 | /* Allocate an io_state for all the devices in the comps array |
299 | * | |
300 | * This version of io_state allocation is used mostly by create/remove | |
301 | * and trunc where we currently need all the devices. The only wastful | |
302 | * bit is the read/write_attributes with no IO. Those sites should | |
303 | * be converted to use ore_get_rw_state() with length=0 | |
304 | */ | |
5bf696da | 305 | int ore_get_io_state(struct ore_layout *layout, struct ore_components *oc, |
b916c5cd | 306 | struct ore_io_state **pios) |
e1042ba0 | 307 | { |
a1fec1db | 308 | return _ore_get_io_state(layout, oc, oc->numdevs, 0, 0, pios); |
e1042ba0 | 309 | } |
cf283ade | 310 | EXPORT_SYMBOL(ore_get_io_state); |
e1042ba0 | 311 | |
8ff660ab | 312 | void ore_put_io_state(struct ore_io_state *ios) |
b14f8ab2 | 313 | { |
06886a5a BH |
314 | if (ios) { |
315 | unsigned i; | |
b14f8ab2 | 316 | |
06886a5a | 317 | for (i = 0; i < ios->numdevs; i++) { |
8ff660ab | 318 | struct ore_per_dev_state *per_dev = &ios->per_dev[i]; |
06886a5a BH |
319 | |
320 | if (per_dev->or) | |
321 | osd_end_request(per_dev->or); | |
322 | if (per_dev->bio) | |
323 | bio_put(per_dev->bio); | |
324 | } | |
325 | ||
a1fec1db | 326 | _ore_free_raid_stuff(ios); |
06886a5a | 327 | kfree(ios); |
b14f8ab2 | 328 | } |
06886a5a | 329 | } |
cf283ade | 330 | EXPORT_SYMBOL(ore_put_io_state); |
b14f8ab2 | 331 | |
8ff660ab | 332 | static void _sync_done(struct ore_io_state *ios, void *p) |
06886a5a BH |
333 | { |
334 | struct completion *waiting = p; | |
b14f8ab2 | 335 | |
06886a5a BH |
336 | complete(waiting); |
337 | } | |
338 | ||
339 | static void _last_io(struct kref *kref) | |
340 | { | |
8ff660ab BH |
341 | struct ore_io_state *ios = container_of( |
342 | kref, struct ore_io_state, kref); | |
06886a5a BH |
343 | |
344 | ios->done(ios, ios->private); | |
345 | } | |
346 | ||
347 | static void _done_io(struct osd_request *or, void *p) | |
348 | { | |
8ff660ab | 349 | struct ore_io_state *ios = p; |
06886a5a BH |
350 | |
351 | kref_put(&ios->kref, _last_io); | |
352 | } | |
353 | ||
769ba8d9 | 354 | int ore_io_execute(struct ore_io_state *ios) |
06886a5a BH |
355 | { |
356 | DECLARE_COMPLETION_ONSTACK(wait); | |
357 | bool sync = (ios->done == NULL); | |
358 | int i, ret; | |
359 | ||
360 | if (sync) { | |
361 | ios->done = _sync_done; | |
362 | ios->private = &wait; | |
363 | } | |
364 | ||
365 | for (i = 0; i < ios->numdevs; i++) { | |
366 | struct osd_request *or = ios->per_dev[i].or; | |
367 | if (unlikely(!or)) | |
368 | continue; | |
369 | ||
9e9db456 | 370 | ret = osd_finalize_request(or, 0, _ios_cred(ios, i), NULL); |
06886a5a | 371 | if (unlikely(ret)) { |
8ff660ab | 372 | ORE_DBGMSG("Failed to osd_finalize_request() => %d\n", |
06886a5a BH |
373 | ret); |
374 | return ret; | |
375 | } | |
376 | } | |
377 | ||
378 | kref_init(&ios->kref); | |
379 | ||
380 | for (i = 0; i < ios->numdevs; i++) { | |
381 | struct osd_request *or = ios->per_dev[i].or; | |
382 | if (unlikely(!or)) | |
383 | continue; | |
384 | ||
385 | kref_get(&ios->kref); | |
386 | osd_execute_request_async(or, _done_io, ios); | |
387 | } | |
388 | ||
389 | kref_put(&ios->kref, _last_io); | |
390 | ret = 0; | |
391 | ||
392 | if (sync) { | |
393 | wait_for_completion(&wait); | |
8ff660ab | 394 | ret = ore_check_io(ios, NULL); |
06886a5a | 395 | } |
b14f8ab2 BH |
396 | return ret; |
397 | } | |
398 | ||
22ddc556 BH |
399 | static void _clear_bio(struct bio *bio) |
400 | { | |
401 | struct bio_vec *bv; | |
402 | unsigned i; | |
403 | ||
404 | __bio_for_each_segment(bv, bio, i, 0) { | |
405 | unsigned this_count = bv->bv_len; | |
406 | ||
407 | if (likely(PAGE_SIZE == this_count)) | |
408 | clear_highpage(bv->bv_page); | |
409 | else | |
410 | zero_user(bv->bv_page, bv->bv_offset, this_count); | |
411 | } | |
412 | } | |
413 | ||
4b46c9f5 | 414 | int ore_check_io(struct ore_io_state *ios, ore_on_dev_error on_dev_error) |
b14f8ab2 | 415 | { |
06886a5a BH |
416 | enum osd_err_priority acumulated_osd_err = 0; |
417 | int acumulated_lin_err = 0; | |
418 | int i; | |
b14f8ab2 | 419 | |
06886a5a BH |
420 | for (i = 0; i < ios->numdevs; i++) { |
421 | struct osd_sense_info osi; | |
4b46c9f5 BH |
422 | struct ore_per_dev_state *per_dev = &ios->per_dev[i]; |
423 | struct osd_request *or = per_dev->or; | |
22ddc556 BH |
424 | int ret; |
425 | ||
426 | if (unlikely(!or)) | |
427 | continue; | |
06886a5a | 428 | |
22ddc556 | 429 | ret = osd_req_decode_sense(or, &osi); |
06886a5a BH |
430 | if (likely(!ret)) |
431 | continue; | |
432 | ||
22ddc556 BH |
433 | if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) { |
434 | /* start read offset passed endof file */ | |
4b46c9f5 | 435 | _clear_bio(per_dev->bio); |
8ff660ab | 436 | ORE_DBGMSG("start read offset passed end of file " |
22ddc556 | 437 | "offset=0x%llx, length=0x%llx\n", |
4b46c9f5 BH |
438 | _LLU(per_dev->offset), |
439 | _LLU(per_dev->length)); | |
22ddc556 BH |
440 | |
441 | continue; /* we recovered */ | |
06886a5a BH |
442 | } |
443 | ||
4b46c9f5 BH |
444 | if (on_dev_error) { |
445 | u64 residual = ios->reading ? | |
446 | or->in.residual : or->out.residual; | |
447 | u64 offset = (ios->offset + ios->length) - residual; | |
448 | struct ore_dev *od = ios->oc->ods[ | |
449 | per_dev->dev - ios->oc->first_dev]; | |
450 | ||
451 | on_dev_error(ios, od, per_dev->dev, osi.osd_err_pri, | |
452 | offset, residual); | |
453 | } | |
06886a5a BH |
454 | if (osi.osd_err_pri >= acumulated_osd_err) { |
455 | acumulated_osd_err = osi.osd_err_pri; | |
456 | acumulated_lin_err = ret; | |
457 | } | |
458 | } | |
459 | ||
06886a5a BH |
460 | return acumulated_lin_err; |
461 | } | |
cf283ade | 462 | EXPORT_SYMBOL(ore_check_io); |
06886a5a | 463 | |
b367e78b BH |
464 | /* |
465 | * L - logical offset into the file | |
466 | * | |
a1fec1db BH |
467 | * D - number of Data devices |
468 | * D = group_width - parity | |
b367e78b | 469 | * |
a1fec1db BH |
470 | * U - The number of bytes in a stripe within a group |
471 | * U = stripe_unit * D | |
b367e78b | 472 | * |
50a76fd3 BH |
473 | * T - The number of bytes striped within a group of component objects |
474 | * (before advancing to the next group) | |
a1fec1db | 475 | * T = U * group_depth |
50a76fd3 BH |
476 | * |
477 | * S - The number of bytes striped across all component objects | |
478 | * before the pattern repeats | |
a1fec1db | 479 | * S = T * group_count |
50a76fd3 | 480 | * |
a1fec1db | 481 | * M - The "major" (i.e., across all components) cycle number |
50a76fd3 BH |
482 | * M = L / S |
483 | * | |
a1fec1db | 484 | * G - Counts the groups from the beginning of the major cycle |
50a76fd3 BH |
485 | * G = (L - (M * S)) / T [or (L % S) / T] |
486 | * | |
487 | * H - The byte offset within the group | |
50a76fd3 BH |
488 | * H = (L - (M * S)) % T [or (L % S) % T] |
489 | * | |
490 | * N - The "minor" (i.e., across the group) stripe number | |
50a76fd3 | 491 | * N = H / U |
b367e78b BH |
492 | * |
493 | * C - The component index coresponding to L | |
494 | * | |
a1fec1db BH |
495 | * C = (H - (N * U)) / stripe_unit + G * D |
496 | * [or (L % U) / stripe_unit + G * D] | |
b367e78b BH |
497 | * |
498 | * O - The component offset coresponding to L | |
50a76fd3 | 499 | * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit |
a1fec1db BH |
500 | * |
501 | * LCMdP – Parity cycle: Lowest Common Multiple of group_width, parity | |
502 | * divide by parity | |
503 | * LCMdP = lcm(group_width, parity) / parity | |
504 | * | |
505 | * R - The parity Rotation stripe | |
506 | * (Note parity cycle always starts at a group's boundary) | |
507 | * R = N % LCMdP | |
508 | * | |
509 | * I = the first parity device index | |
510 | * I = (group_width + group_width - R*parity - parity) % group_width | |
511 | * | |
512 | * Craid - The component index Rotated | |
513 | * Craid = (group_width + C - R*parity) % group_width | |
514 | * (We add the group_width to avoid negative numbers modulo math) | |
b367e78b | 515 | */ |
611d7a5d | 516 | void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset, |
a1fec1db | 517 | u64 length, struct ore_striping_info *si) |
5d952b83 | 518 | { |
16f75bb3 BH |
519 | u32 stripe_unit = layout->stripe_unit; |
520 | u32 group_width = layout->group_width; | |
521 | u64 group_depth = layout->group_depth; | |
a1fec1db | 522 | u32 parity = layout->parity; |
50a76fd3 | 523 | |
a1fec1db BH |
524 | u32 D = group_width - parity; |
525 | u32 U = D * stripe_unit; | |
50a76fd3 | 526 | u64 T = U * group_depth; |
16f75bb3 | 527 | u64 S = T * layout->group_count; |
50a76fd3 BH |
528 | u64 M = div64_u64(file_offset, S); |
529 | ||
530 | /* | |
531 | G = (L - (M * S)) / T | |
532 | H = (L - (M * S)) % T | |
533 | */ | |
534 | u64 LmodS = file_offset - M * S; | |
535 | u32 G = div64_u64(LmodS, T); | |
536 | u64 H = LmodS - G * T; | |
537 | ||
538 | u32 N = div_u64(H, U); | |
539 | ||
540 | /* "H - (N * U)" is just "H % U" so it's bound to u32 */ | |
a1fec1db | 541 | u32 C = (u32)(H - (N * U)) / stripe_unit + G * group_width; |
b367e78b | 542 | |
50a76fd3 | 543 | div_u64_rem(file_offset, stripe_unit, &si->unit_off); |
5d952b83 | 544 | |
50a76fd3 BH |
545 | si->obj_offset = si->unit_off + (N * stripe_unit) + |
546 | (M * group_depth * stripe_unit); | |
547 | ||
a1fec1db BH |
548 | if (parity) { |
549 | u32 LCMdP = lcm(group_width, parity) / parity; | |
550 | /* R = N % LCMdP; */ | |
551 | u32 RxP = (N % LCMdP) * parity; | |
552 | u32 first_dev = C - C % group_width; | |
553 | ||
554 | si->par_dev = (group_width + group_width - parity - RxP) % | |
555 | group_width + first_dev; | |
556 | si->dev = (group_width + C - RxP) % group_width + first_dev; | |
557 | si->bytes_in_stripe = U; | |
558 | si->first_stripe_start = M * S + G * T + N * U; | |
559 | } else { | |
560 | /* Make the math correct see _prepare_one_group */ | |
561 | si->par_dev = group_width; | |
562 | si->dev = C; | |
563 | } | |
564 | ||
565 | si->dev *= layout->mirrors_p1; | |
566 | si->par_dev *= layout->mirrors_p1; | |
567 | si->offset = file_offset; | |
568 | si->length = T - H; | |
569 | if (si->length > length) | |
570 | si->length = length; | |
16f75bb3 | 571 | si->M = M; |
5d952b83 | 572 | } |
611d7a5d | 573 | EXPORT_SYMBOL(ore_calc_stripe_info); |
5d952b83 | 574 | |
a1fec1db BH |
575 | int _ore_add_stripe_unit(struct ore_io_state *ios, unsigned *cur_pg, |
576 | unsigned pgbase, struct page **pages, | |
577 | struct ore_per_dev_state *per_dev, int cur_len) | |
5d952b83 | 578 | { |
86093aaf | 579 | unsigned pg = *cur_pg; |
5d952b83 | 580 | struct request_queue *q = |
9e9db456 | 581 | osd_request_queue(_ios_od(ios, per_dev->dev)); |
bbf9a31b BH |
582 | unsigned len = cur_len; |
583 | int ret; | |
5d952b83 BH |
584 | |
585 | if (per_dev->bio == NULL) { | |
586 | unsigned pages_in_stripe = ios->layout->group_width * | |
587 | (ios->layout->stripe_unit / PAGE_SIZE); | |
a1fec1db BH |
588 | unsigned nr_pages = ios->nr_pages * ios->layout->group_width / |
589 | (ios->layout->group_width - | |
590 | ios->layout->parity); | |
591 | unsigned bio_size = (nr_pages + pages_in_stripe) / | |
592 | ios->layout->group_width; | |
5d952b83 BH |
593 | |
594 | per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size); | |
595 | if (unlikely(!per_dev->bio)) { | |
8ff660ab | 596 | ORE_DBGMSG("Failed to allocate BIO size=%u\n", |
5d952b83 | 597 | bio_size); |
bbf9a31b BH |
598 | ret = -ENOMEM; |
599 | goto out; | |
5d952b83 BH |
600 | } |
601 | } | |
602 | ||
603 | while (cur_len > 0) { | |
86093aaf BH |
604 | unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len); |
605 | unsigned added_len; | |
5d952b83 | 606 | |
86093aaf | 607 | cur_len -= pglen; |
5d952b83 | 608 | |
a1fec1db | 609 | added_len = bio_add_pc_page(q, per_dev->bio, pages[pg], |
86093aaf | 610 | pglen, pgbase); |
bbf9a31b | 611 | if (unlikely(pglen != added_len)) { |
a1fec1db BH |
612 | ORE_DBGMSG("Failed bio_add_pc_page bi_vcnt=%u\n", |
613 | per_dev->bio->bi_vcnt); | |
bbf9a31b BH |
614 | ret = -ENOMEM; |
615 | goto out; | |
616 | } | |
769ba8d9 BH |
617 | _add_stripe_page(ios->sp2d, &ios->si, pages[pg]); |
618 | ||
86093aaf BH |
619 | pgbase = 0; |
620 | ++pg; | |
5d952b83 BH |
621 | } |
622 | BUG_ON(cur_len); | |
623 | ||
bbf9a31b | 624 | per_dev->length += len; |
86093aaf | 625 | *cur_pg = pg; |
bbf9a31b BH |
626 | ret = 0; |
627 | out: /* we fail the complete unit on an error eg don't advance | |
628 | * per_dev->length and cur_pg. This means that we might have a bigger | |
629 | * bio than the CDB requested length (per_dev->length). That's fine | |
630 | * only the oposite is fatal. | |
631 | */ | |
632 | return ret; | |
5d952b83 BH |
633 | } |
634 | ||
98260754 | 635 | static int _prepare_for_striping(struct ore_io_state *ios) |
5d952b83 | 636 | { |
98260754 | 637 | struct ore_striping_info *si = &ios->si; |
5d952b83 | 638 | unsigned stripe_unit = ios->layout->stripe_unit; |
b367e78b | 639 | unsigned mirrors_p1 = ios->layout->mirrors_p1; |
a1fec1db BH |
640 | unsigned group_width = ios->layout->group_width; |
641 | unsigned devs_in_group = group_width * mirrors_p1; | |
b367e78b | 642 | unsigned dev = si->dev; |
50a76fd3 | 643 | unsigned first_dev = dev - (dev % devs_in_group); |
a1fec1db | 644 | unsigned dev_order; |
50a76fd3 | 645 | unsigned cur_pg = ios->pages_consumed; |
98260754 | 646 | u64 length = ios->length; |
86093aaf | 647 | int ret = 0; |
5d952b83 | 648 | |
98260754 | 649 | if (!ios->pages) { |
98260754 BH |
650 | ios->numdevs = ios->layout->mirrors_p1; |
651 | return 0; | |
652 | } | |
653 | ||
a1fec1db BH |
654 | BUG_ON(length > si->length); |
655 | ||
656 | dev_order = _dev_order(devs_in_group, mirrors_p1, si->par_dev, dev); | |
657 | si->cur_comp = dev_order; | |
769ba8d9 | 658 | si->cur_pg = si->unit_off / PAGE_SIZE; |
98260754 | 659 | |
5d952b83 | 660 | while (length) { |
b916c5cd BH |
661 | unsigned comp = dev - first_dev; |
662 | struct ore_per_dev_state *per_dev = &ios->per_dev[comp]; | |
b367e78b | 663 | unsigned cur_len, page_off = 0; |
5d952b83 BH |
664 | |
665 | if (!per_dev->length) { | |
b367e78b | 666 | per_dev->dev = dev; |
a1fec1db BH |
667 | if (dev == si->dev) { |
668 | WARN_ON(dev == si->par_dev); | |
b367e78b BH |
669 | per_dev->offset = si->obj_offset; |
670 | cur_len = stripe_unit - si->unit_off; | |
671 | page_off = si->unit_off & ~PAGE_MASK; | |
672 | BUG_ON(page_off && (page_off != ios->pgbase)); | |
a1fec1db BH |
673 | } else { |
674 | if (si->cur_comp > dev_order) | |
675 | per_dev->offset = | |
676 | si->obj_offset - si->unit_off; | |
677 | else /* si->cur_comp < dev_order */ | |
678 | per_dev->offset = | |
679 | si->obj_offset + stripe_unit - | |
680 | si->unit_off; | |
b367e78b BH |
681 | cur_len = stripe_unit; |
682 | } | |
5d952b83 | 683 | } else { |
b367e78b | 684 | cur_len = stripe_unit; |
5d952b83 | 685 | } |
b367e78b BH |
686 | if (cur_len >= length) |
687 | cur_len = length; | |
5d952b83 | 688 | |
a1fec1db BH |
689 | ret = _ore_add_stripe_unit(ios, &cur_pg, page_off, ios->pages, |
690 | per_dev, cur_len); | |
5d952b83 BH |
691 | if (unlikely(ret)) |
692 | goto out; | |
693 | ||
6e31609b BH |
694 | dev += mirrors_p1; |
695 | dev = (dev % devs_in_group) + first_dev; | |
5d952b83 BH |
696 | |
697 | length -= cur_len; | |
a1fec1db BH |
698 | |
699 | si->cur_comp = (si->cur_comp + 1) % group_width; | |
769ba8d9 BH |
700 | if (unlikely((dev == si->par_dev) || (!length && ios->sp2d))) { |
701 | if (!length && ios->sp2d) { | |
a1fec1db BH |
702 | /* If we are writing and this is the very last |
703 | * stripe. then operate on parity dev. | |
704 | */ | |
705 | dev = si->par_dev; | |
769ba8d9 BH |
706 | } |
707 | if (ios->sp2d) | |
a1fec1db BH |
708 | /* In writes cur_len just means if it's the |
709 | * last one. See _ore_add_parity_unit. | |
710 | */ | |
711 | cur_len = length; | |
712 | per_dev = &ios->per_dev[dev - first_dev]; | |
713 | if (!per_dev->length) { | |
714 | /* Only/always the parity unit of the first | |
715 | * stripe will be empty. So this is a chance to | |
716 | * initialize the per_dev info. | |
717 | */ | |
718 | per_dev->dev = dev; | |
719 | per_dev->offset = si->obj_offset - si->unit_off; | |
720 | } | |
721 | ||
722 | ret = _ore_add_parity_unit(ios, si, per_dev, cur_len); | |
723 | if (unlikely(ret)) | |
724 | goto out; | |
725 | ||
726 | /* Rotate next par_dev backwards with wraping */ | |
727 | si->par_dev = (devs_in_group + si->par_dev - | |
728 | ios->layout->parity * mirrors_p1) % | |
729 | devs_in_group + first_dev; | |
730 | /* Next stripe, start fresh */ | |
731 | si->cur_comp = 0; | |
769ba8d9 | 732 | si->cur_pg = 0; |
a1fec1db | 733 | } |
5d952b83 BH |
734 | } |
735 | out: | |
b916c5cd | 736 | ios->numdevs = devs_in_group; |
50a76fd3 | 737 | ios->pages_consumed = cur_pg; |
bbf9a31b BH |
738 | if (unlikely(ret)) { |
739 | if (length == ios->length) | |
740 | return ret; | |
741 | else | |
742 | ios->length -= length; | |
743 | } | |
744 | return 0; | |
5d952b83 BH |
745 | } |
746 | ||
8ff660ab | 747 | int ore_create(struct ore_io_state *ios) |
06886a5a BH |
748 | { |
749 | int i, ret; | |
750 | ||
5bf696da | 751 | for (i = 0; i < ios->oc->numdevs; i++) { |
06886a5a BH |
752 | struct osd_request *or; |
753 | ||
9e9db456 | 754 | or = osd_start_request(_ios_od(ios, i), GFP_KERNEL); |
06886a5a | 755 | if (unlikely(!or)) { |
8ff660ab | 756 | ORE_ERR("%s: osd_start_request failed\n", __func__); |
06886a5a BH |
757 | ret = -ENOMEM; |
758 | goto out; | |
759 | } | |
760 | ios->per_dev[i].or = or; | |
761 | ios->numdevs++; | |
762 | ||
9e9db456 | 763 | osd_req_create_object(or, _ios_obj(ios, i)); |
06886a5a | 764 | } |
8ff660ab | 765 | ret = ore_io_execute(ios); |
06886a5a BH |
766 | |
767 | out: | |
768 | return ret; | |
769 | } | |
cf283ade | 770 | EXPORT_SYMBOL(ore_create); |
06886a5a | 771 | |
8ff660ab | 772 | int ore_remove(struct ore_io_state *ios) |
06886a5a BH |
773 | { |
774 | int i, ret; | |
775 | ||
5bf696da | 776 | for (i = 0; i < ios->oc->numdevs; i++) { |
06886a5a BH |
777 | struct osd_request *or; |
778 | ||
9e9db456 | 779 | or = osd_start_request(_ios_od(ios, i), GFP_KERNEL); |
06886a5a | 780 | if (unlikely(!or)) { |
8ff660ab | 781 | ORE_ERR("%s: osd_start_request failed\n", __func__); |
06886a5a BH |
782 | ret = -ENOMEM; |
783 | goto out; | |
784 | } | |
785 | ios->per_dev[i].or = or; | |
786 | ios->numdevs++; | |
787 | ||
9e9db456 | 788 | osd_req_remove_object(or, _ios_obj(ios, i)); |
06886a5a | 789 | } |
8ff660ab | 790 | ret = ore_io_execute(ios); |
06886a5a BH |
791 | |
792 | out: | |
793 | return ret; | |
794 | } | |
cf283ade | 795 | EXPORT_SYMBOL(ore_remove); |
06886a5a | 796 | |
8ff660ab | 797 | static int _write_mirror(struct ore_io_state *ios, int cur_comp) |
06886a5a | 798 | { |
8ff660ab | 799 | struct ore_per_dev_state *master_dev = &ios->per_dev[cur_comp]; |
5d952b83 BH |
800 | unsigned dev = ios->per_dev[cur_comp].dev; |
801 | unsigned last_comp = cur_comp + ios->layout->mirrors_p1; | |
802 | int ret = 0; | |
06886a5a | 803 | |
50a76fd3 BH |
804 | if (ios->pages && !master_dev->length) |
805 | return 0; /* Just an empty slot */ | |
806 | ||
5d952b83 | 807 | for (; cur_comp < last_comp; ++cur_comp, ++dev) { |
8ff660ab | 808 | struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp]; |
06886a5a BH |
809 | struct osd_request *or; |
810 | ||
9e9db456 | 811 | or = osd_start_request(_ios_od(ios, dev), GFP_KERNEL); |
06886a5a | 812 | if (unlikely(!or)) { |
8ff660ab | 813 | ORE_ERR("%s: osd_start_request failed\n", __func__); |
06886a5a BH |
814 | ret = -ENOMEM; |
815 | goto out; | |
816 | } | |
5d952b83 | 817 | per_dev->or = or; |
06886a5a | 818 | |
86093aaf | 819 | if (ios->pages) { |
06886a5a BH |
820 | struct bio *bio; |
821 | ||
5d952b83 | 822 | if (per_dev != master_dev) { |
04dc1e88 | 823 | bio = bio_kmalloc(GFP_KERNEL, |
5d952b83 | 824 | master_dev->bio->bi_max_vecs); |
04dc1e88 | 825 | if (unlikely(!bio)) { |
8ff660ab | 826 | ORE_DBGMSG( |
426d3107 | 827 | "Failed to allocate BIO size=%u\n", |
5d952b83 | 828 | master_dev->bio->bi_max_vecs); |
04dc1e88 BH |
829 | ret = -ENOMEM; |
830 | goto out; | |
831 | } | |
832 | ||
5d952b83 | 833 | __bio_clone(bio, master_dev->bio); |
04dc1e88 BH |
834 | bio->bi_bdev = NULL; |
835 | bio->bi_next = NULL; | |
6851a5e5 | 836 | per_dev->offset = master_dev->offset; |
5d952b83 BH |
837 | per_dev->length = master_dev->length; |
838 | per_dev->bio = bio; | |
839 | per_dev->dev = dev; | |
04dc1e88 | 840 | } else { |
5d952b83 BH |
841 | bio = master_dev->bio; |
842 | /* FIXME: bio_set_dir() */ | |
7b6d91da | 843 | bio->bi_rw |= REQ_WRITE; |
04dc1e88 | 844 | } |
06886a5a | 845 | |
9e9db456 BH |
846 | osd_req_write(or, _ios_obj(ios, dev), per_dev->offset, |
847 | bio, per_dev->length); | |
8ff660ab | 848 | ORE_DBGMSG("write(0x%llx) offset=0x%llx " |
34ce4e7c | 849 | "length=0x%llx dev=%d\n", |
9e9db456 BH |
850 | _LLU(_ios_obj(ios, dev)->id), |
851 | _LLU(per_dev->offset), | |
5d952b83 | 852 | _LLU(per_dev->length), dev); |
06886a5a | 853 | } else if (ios->kern_buff) { |
6851a5e5 BH |
854 | per_dev->offset = ios->si.obj_offset; |
855 | per_dev->dev = ios->si.dev + dev; | |
856 | ||
857 | /* no cross device without page array */ | |
858 | BUG_ON((ios->layout->group_width > 1) && | |
859 | (ios->si.unit_off + ios->length > | |
860 | ios->layout->stripe_unit)); | |
861 | ||
862 | ret = osd_req_write_kern(or, _ios_obj(ios, per_dev->dev), | |
9e9db456 BH |
863 | per_dev->offset, |
864 | ios->kern_buff, ios->length); | |
5d952b83 BH |
865 | if (unlikely(ret)) |
866 | goto out; | |
8ff660ab | 867 | ORE_DBGMSG2("write_kern(0x%llx) offset=0x%llx " |
34ce4e7c | 868 | "length=0x%llx dev=%d\n", |
9e9db456 BH |
869 | _LLU(_ios_obj(ios, dev)->id), |
870 | _LLU(per_dev->offset), | |
6851a5e5 | 871 | _LLU(ios->length), per_dev->dev); |
06886a5a | 872 | } else { |
9e9db456 | 873 | osd_req_set_attributes(or, _ios_obj(ios, dev)); |
8ff660ab | 874 | ORE_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n", |
9e9db456 BH |
875 | _LLU(_ios_obj(ios, dev)->id), |
876 | ios->out_attr_len, dev); | |
06886a5a BH |
877 | } |
878 | ||
879 | if (ios->out_attr) | |
880 | osd_req_add_set_attr_list(or, ios->out_attr, | |
881 | ios->out_attr_len); | |
882 | ||
883 | if (ios->in_attr) | |
884 | osd_req_add_get_attr_list(or, ios->in_attr, | |
885 | ios->in_attr_len); | |
b14f8ab2 | 886 | } |
06886a5a BH |
887 | |
888 | out: | |
889 | return ret; | |
890 | } | |
891 | ||
8ff660ab | 892 | int ore_write(struct ore_io_state *ios) |
5d952b83 BH |
893 | { |
894 | int i; | |
895 | int ret; | |
896 | ||
769ba8d9 BH |
897 | if (unlikely(ios->sp2d && !ios->r4w)) { |
898 | /* A library is attempting a RAID-write without providing | |
899 | * a pages lock interface. | |
900 | */ | |
901 | WARN_ON_ONCE(1); | |
902 | return -ENOTSUPP; | |
903 | } | |
904 | ||
5d952b83 BH |
905 | ret = _prepare_for_striping(ios); |
906 | if (unlikely(ret)) | |
907 | return ret; | |
908 | ||
909 | for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) { | |
8ff660ab | 910 | ret = _write_mirror(ios, i); |
5d952b83 BH |
911 | if (unlikely(ret)) |
912 | return ret; | |
913 | } | |
914 | ||
8ff660ab | 915 | ret = ore_io_execute(ios); |
5d952b83 BH |
916 | return ret; |
917 | } | |
cf283ade | 918 | EXPORT_SYMBOL(ore_write); |
5d952b83 | 919 | |
769ba8d9 | 920 | int _ore_read_mirror(struct ore_io_state *ios, unsigned cur_comp) |
06886a5a | 921 | { |
46f4d973 | 922 | struct osd_request *or; |
8ff660ab | 923 | struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp]; |
9e9db456 BH |
924 | struct osd_obj_id *obj = _ios_obj(ios, cur_comp); |
925 | unsigned first_dev = (unsigned)obj->id; | |
06886a5a | 926 | |
50a76fd3 BH |
927 | if (ios->pages && !per_dev->length) |
928 | return 0; /* Just an empty slot */ | |
929 | ||
5d952b83 | 930 | first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1; |
9e9db456 | 931 | or = osd_start_request(_ios_od(ios, first_dev), GFP_KERNEL); |
46f4d973 | 932 | if (unlikely(!or)) { |
8ff660ab | 933 | ORE_ERR("%s: osd_start_request failed\n", __func__); |
46f4d973 BH |
934 | return -ENOMEM; |
935 | } | |
936 | per_dev->or = or; | |
46f4d973 | 937 | |
86093aaf | 938 | if (ios->pages) { |
a1fec1db BH |
939 | if (per_dev->cur_sg) { |
940 | /* finalize the last sg_entry */ | |
941 | _ore_add_sg_seg(per_dev, 0, false); | |
942 | if (unlikely(!per_dev->cur_sg)) | |
943 | return 0; /* Skip parity only device */ | |
944 | ||
945 | osd_req_read_sg(or, obj, per_dev->bio, | |
946 | per_dev->sglist, per_dev->cur_sg); | |
947 | } else { | |
948 | /* The no raid case */ | |
949 | osd_req_read(or, obj, per_dev->offset, | |
950 | per_dev->bio, per_dev->length); | |
951 | } | |
952 | ||
8ff660ab | 953 | ORE_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx" |
a1fec1db | 954 | " dev=%d sg_len=%d\n", _LLU(obj->id), |
5d952b83 | 955 | _LLU(per_dev->offset), _LLU(per_dev->length), |
a1fec1db | 956 | first_dev, per_dev->cur_sg); |
46f4d973 | 957 | } else { |
6851a5e5 BH |
958 | BUG_ON(ios->kern_buff); |
959 | ||
9e9db456 | 960 | osd_req_get_attributes(or, obj); |
8ff660ab | 961 | ORE_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n", |
9e9db456 BH |
962 | _LLU(obj->id), |
963 | ios->in_attr_len, first_dev); | |
46f4d973 | 964 | } |
46f4d973 BH |
965 | if (ios->out_attr) |
966 | osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len); | |
b14f8ab2 | 967 | |
46f4d973 BH |
968 | if (ios->in_attr) |
969 | osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len); | |
b14f8ab2 | 970 | |
5d952b83 BH |
971 | return 0; |
972 | } | |
973 | ||
8ff660ab | 974 | int ore_read(struct ore_io_state *ios) |
5d952b83 BH |
975 | { |
976 | int i; | |
977 | int ret; | |
978 | ||
979 | ret = _prepare_for_striping(ios); | |
980 | if (unlikely(ret)) | |
981 | return ret; | |
982 | ||
983 | for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) { | |
769ba8d9 | 984 | ret = _ore_read_mirror(ios, i); |
5d952b83 BH |
985 | if (unlikely(ret)) |
986 | return ret; | |
987 | } | |
988 | ||
8ff660ab | 989 | ret = ore_io_execute(ios); |
5d952b83 | 990 | return ret; |
b14f8ab2 | 991 | } |
cf283ade | 992 | EXPORT_SYMBOL(ore_read); |
b14f8ab2 | 993 | |
8ff660ab | 994 | int extract_attr_from_ios(struct ore_io_state *ios, struct osd_attr *attr) |
b14f8ab2 BH |
995 | { |
996 | struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */ | |
997 | void *iter = NULL; | |
998 | int nelem; | |
999 | ||
1000 | do { | |
1001 | nelem = 1; | |
06886a5a BH |
1002 | osd_req_decode_get_attr_list(ios->per_dev[0].or, |
1003 | &cur_attr, &nelem, &iter); | |
b14f8ab2 BH |
1004 | if ((cur_attr.attr_page == attr->attr_page) && |
1005 | (cur_attr.attr_id == attr->attr_id)) { | |
1006 | attr->len = cur_attr.len; | |
1007 | attr->val_ptr = cur_attr.val_ptr; | |
1008 | return 0; | |
1009 | } | |
1010 | } while (iter); | |
1011 | ||
1012 | return -EIO; | |
1013 | } | |
cf283ade | 1014 | EXPORT_SYMBOL(extract_attr_from_ios); |
06886a5a | 1015 | |
8ff660ab | 1016 | static int _truncate_mirrors(struct ore_io_state *ios, unsigned cur_comp, |
5d952b83 BH |
1017 | struct osd_attr *attr) |
1018 | { | |
1019 | int last_comp = cur_comp + ios->layout->mirrors_p1; | |
1020 | ||
1021 | for (; cur_comp < last_comp; ++cur_comp) { | |
8ff660ab | 1022 | struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp]; |
5d952b83 BH |
1023 | struct osd_request *or; |
1024 | ||
9e9db456 | 1025 | or = osd_start_request(_ios_od(ios, cur_comp), GFP_KERNEL); |
5d952b83 | 1026 | if (unlikely(!or)) { |
8ff660ab | 1027 | ORE_ERR("%s: osd_start_request failed\n", __func__); |
5d952b83 BH |
1028 | return -ENOMEM; |
1029 | } | |
1030 | per_dev->or = or; | |
1031 | ||
9e9db456 | 1032 | osd_req_set_attributes(or, _ios_obj(ios, cur_comp)); |
5d952b83 BH |
1033 | osd_req_add_set_attr_list(or, attr, 1); |
1034 | } | |
1035 | ||
1036 | return 0; | |
1037 | } | |
1038 | ||
16f75bb3 | 1039 | struct _trunc_info { |
eb507bc1 | 1040 | struct ore_striping_info si; |
16f75bb3 BH |
1041 | u64 prev_group_obj_off; |
1042 | u64 next_group_obj_off; | |
1043 | ||
1044 | unsigned first_group_dev; | |
1045 | unsigned nex_group_dev; | |
16f75bb3 BH |
1046 | }; |
1047 | ||
1958c7c2 HS |
1048 | static void _calc_trunk_info(struct ore_layout *layout, u64 file_offset, |
1049 | struct _trunc_info *ti) | |
16f75bb3 BH |
1050 | { |
1051 | unsigned stripe_unit = layout->stripe_unit; | |
1052 | ||
a1fec1db | 1053 | ore_calc_stripe_info(layout, file_offset, 0, &ti->si); |
16f75bb3 BH |
1054 | |
1055 | ti->prev_group_obj_off = ti->si.M * stripe_unit; | |
1056 | ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0; | |
1057 | ||
1058 | ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width); | |
1059 | ti->nex_group_dev = ti->first_group_dev + layout->group_width; | |
16f75bb3 BH |
1060 | } |
1061 | ||
5bf696da | 1062 | int ore_truncate(struct ore_layout *layout, struct ore_components *oc, |
9e9db456 | 1063 | u64 size) |
06886a5a | 1064 | { |
8ff660ab | 1065 | struct ore_io_state *ios; |
5d952b83 BH |
1066 | struct exofs_trunc_attr { |
1067 | struct osd_attr attr; | |
1068 | __be64 newsize; | |
1069 | } *size_attrs; | |
16f75bb3 | 1070 | struct _trunc_info ti; |
06886a5a BH |
1071 | int i, ret; |
1072 | ||
5bf696da | 1073 | ret = ore_get_io_state(layout, oc, &ios); |
5d952b83 BH |
1074 | if (unlikely(ret)) |
1075 | return ret; | |
1076 | ||
16f75bb3 BH |
1077 | _calc_trunk_info(ios->layout, size, &ti); |
1078 | ||
b916c5cd | 1079 | size_attrs = kcalloc(ios->oc->numdevs, sizeof(*size_attrs), |
5d952b83 BH |
1080 | GFP_KERNEL); |
1081 | if (unlikely(!size_attrs)) { | |
1082 | ret = -ENOMEM; | |
1083 | goto out; | |
1084 | } | |
06886a5a | 1085 | |
5bf696da | 1086 | ios->numdevs = ios->oc->numdevs; |
06886a5a | 1087 | |
b916c5cd | 1088 | for (i = 0; i < ios->numdevs; ++i) { |
5d952b83 BH |
1089 | struct exofs_trunc_attr *size_attr = &size_attrs[i]; |
1090 | u64 obj_size; | |
06886a5a | 1091 | |
16f75bb3 BH |
1092 | if (i < ti.first_group_dev) |
1093 | obj_size = ti.prev_group_obj_off; | |
1094 | else if (i >= ti.nex_group_dev) | |
1095 | obj_size = ti.next_group_obj_off; | |
1096 | else if (i < ti.si.dev) /* dev within this group */ | |
1097 | obj_size = ti.si.obj_offset + | |
1098 | ios->layout->stripe_unit - ti.si.unit_off; | |
1099 | else if (i == ti.si.dev) | |
1100 | obj_size = ti.si.obj_offset; | |
1101 | else /* i > ti.dev */ | |
1102 | obj_size = ti.si.obj_offset - ti.si.unit_off; | |
06886a5a | 1103 | |
5d952b83 BH |
1104 | size_attr->newsize = cpu_to_be64(obj_size); |
1105 | size_attr->attr = g_attr_logical_length; | |
1106 | size_attr->attr.val_ptr = &size_attr->newsize; | |
1107 | ||
8ff660ab | 1108 | ORE_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n", |
5bf696da | 1109 | _LLU(oc->comps->obj.id), _LLU(obj_size), i); |
5d952b83 BH |
1110 | ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1, |
1111 | &size_attr->attr); | |
1112 | if (unlikely(ret)) | |
1113 | goto out; | |
06886a5a | 1114 | } |
8ff660ab | 1115 | ret = ore_io_execute(ios); |
06886a5a BH |
1116 | |
1117 | out: | |
5d952b83 | 1118 | kfree(size_attrs); |
8ff660ab | 1119 | ore_put_io_state(ios); |
06886a5a BH |
1120 | return ret; |
1121 | } | |
cf283ade | 1122 | EXPORT_SYMBOL(ore_truncate); |
85e44df4 BH |
1123 | |
1124 | const struct osd_attr g_attr_logical_length = ATTR_DEF( | |
1125 | OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8); | |
cf283ade | 1126 | EXPORT_SYMBOL(g_attr_logical_length); |