2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
7 * This file is part of exofs.
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.
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.
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
25 #include <linux/slab.h>
26 #include <asm/div64.h>
28 #include <scsi/osd_ore.h>
30 #define ORE_ERR(fmt, a...) printk(KERN_ERR "ore: " fmt, ##a)
32 #ifdef CONFIG_EXOFS_DEBUG
33 #define ORE_DBGMSG(fmt, a...) \
34 printk(KERN_NOTICE "ore @%s:%d: " fmt, __func__, __LINE__, ##a)
36 #define ORE_DBGMSG(fmt, a...) \
37 do { if (0) printk(fmt, ##a); } while (0)
40 /* u64 has problems with printk this will cast it to unsigned long long */
41 #define _LLU(x) (unsigned long long)(x)
43 #define ORE_DBGMSG2(M...) do {} while (0)
44 /* #define ORE_DBGMSG2 ORE_DBGMSG */
46 MODULE_AUTHOR("Boaz Harrosh <bharrosh@panasas.com>");
47 MODULE_DESCRIPTION("Objects Raid Engine ore.ko");
48 MODULE_LICENSE("GPL");
50 static u8 *_ios_cred(struct ore_io_state *ios, unsigned index)
52 return ios->oc->comps[index & ios->oc->single_comp].cred;
55 static struct osd_obj_id *_ios_obj(struct ore_io_state *ios, unsigned index)
57 return &ios->oc->comps[index & ios->oc->single_comp].obj;
60 static struct osd_dev *_ios_od(struct ore_io_state *ios, unsigned index)
62 return ios->oc->ods[index];
65 int ore_get_rw_state(struct ore_layout *layout, struct ore_components *oc,
66 bool is_reading, u64 offset, u64 length,
67 struct ore_io_state **pios)
69 struct ore_io_state *ios;
71 /*TODO: Maybe use kmem_cach per sbi of size
72 * exofs_io_state_size(layout->s_numdevs)
74 ios = kzalloc(ore_io_state_size(oc->numdevs), GFP_KERNEL);
76 ORE_DBGMSG("Failed kzalloc bytes=%d\n",
77 ore_io_state_size(oc->numdevs));
86 ios->reading = is_reading;
91 EXPORT_SYMBOL(ore_get_rw_state);
93 int ore_get_io_state(struct ore_layout *layout, struct ore_components *oc,
94 struct ore_io_state **ios)
96 return ore_get_rw_state(layout, oc, true, 0, 0, ios);
98 EXPORT_SYMBOL(ore_get_io_state);
100 void ore_put_io_state(struct ore_io_state *ios)
105 for (i = 0; i < ios->numdevs; i++) {
106 struct ore_per_dev_state *per_dev = &ios->per_dev[i];
109 osd_end_request(per_dev->or);
111 bio_put(per_dev->bio);
117 EXPORT_SYMBOL(ore_put_io_state);
119 static void _sync_done(struct ore_io_state *ios, void *p)
121 struct completion *waiting = p;
126 static void _last_io(struct kref *kref)
128 struct ore_io_state *ios = container_of(
129 kref, struct ore_io_state, kref);
131 ios->done(ios, ios->private);
134 static void _done_io(struct osd_request *or, void *p)
136 struct ore_io_state *ios = p;
138 kref_put(&ios->kref, _last_io);
141 static int ore_io_execute(struct ore_io_state *ios)
143 DECLARE_COMPLETION_ONSTACK(wait);
144 bool sync = (ios->done == NULL);
148 ios->done = _sync_done;
149 ios->private = &wait;
152 for (i = 0; i < ios->numdevs; i++) {
153 struct osd_request *or = ios->per_dev[i].or;
157 ret = osd_finalize_request(or, 0, _ios_cred(ios, i), NULL);
159 ORE_DBGMSG("Failed to osd_finalize_request() => %d\n",
165 kref_init(&ios->kref);
167 for (i = 0; i < ios->numdevs; i++) {
168 struct osd_request *or = ios->per_dev[i].or;
172 kref_get(&ios->kref);
173 osd_execute_request_async(or, _done_io, ios);
176 kref_put(&ios->kref, _last_io);
180 wait_for_completion(&wait);
181 ret = ore_check_io(ios, NULL);
186 static void _clear_bio(struct bio *bio)
191 __bio_for_each_segment(bv, bio, i, 0) {
192 unsigned this_count = bv->bv_len;
194 if (likely(PAGE_SIZE == this_count))
195 clear_highpage(bv->bv_page);
197 zero_user(bv->bv_page, bv->bv_offset, this_count);
201 int ore_check_io(struct ore_io_state *ios, u64 *resid)
203 enum osd_err_priority acumulated_osd_err = 0;
204 int acumulated_lin_err = 0;
207 for (i = 0; i < ios->numdevs; i++) {
208 struct osd_sense_info osi;
209 struct osd_request *or = ios->per_dev[i].or;
215 ret = osd_req_decode_sense(or, &osi);
219 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
220 /* start read offset passed endof file */
221 _clear_bio(ios->per_dev[i].bio);
222 ORE_DBGMSG("start read offset passed end of file "
223 "offset=0x%llx, length=0x%llx\n",
224 _LLU(ios->per_dev[i].offset),
225 _LLU(ios->per_dev[i].length));
227 continue; /* we recovered */
230 if (osi.osd_err_pri >= acumulated_osd_err) {
231 acumulated_osd_err = osi.osd_err_pri;
232 acumulated_lin_err = ret;
236 /* TODO: raid specific residual calculations */
238 if (likely(!acumulated_lin_err))
241 *resid = ios->length;
244 return acumulated_lin_err;
246 EXPORT_SYMBOL(ore_check_io);
249 * L - logical offset into the file
251 * U - The number of bytes in a stripe within a group
253 * U = stripe_unit * group_width
255 * T - The number of bytes striped within a group of component objects
256 * (before advancing to the next group)
258 * T = stripe_unit * group_width * group_depth
260 * S - The number of bytes striped across all component objects
261 * before the pattern repeats
263 * S = stripe_unit * group_width * group_depth * group_count
265 * M - The "major" (i.e., across all components) stripe number
269 * G - Counts the groups from the beginning of the major stripe
271 * G = (L - (M * S)) / T [or (L % S) / T]
273 * H - The byte offset within the group
275 * H = (L - (M * S)) % T [or (L % S) % T]
277 * N - The "minor" (i.e., across the group) stripe number
281 * C - The component index coresponding to L
283 * C = (H - (N * U)) / stripe_unit + G * group_width
284 * [or (L % U) / stripe_unit + G * group_width]
286 * O - The component offset coresponding to L
288 * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
290 static void ore_calc_stripe_info(struct ore_layout *layout, u64 file_offset,
291 struct ore_striping_info *si)
293 u32 stripe_unit = layout->stripe_unit;
294 u32 group_width = layout->group_width;
295 u64 group_depth = layout->group_depth;
297 u32 U = stripe_unit * group_width;
298 u64 T = U * group_depth;
299 u64 S = T * layout->group_count;
300 u64 M = div64_u64(file_offset, S);
303 G = (L - (M * S)) / T
304 H = (L - (M * S)) % T
306 u64 LmodS = file_offset - M * S;
307 u32 G = div64_u64(LmodS, T);
308 u64 H = LmodS - G * T;
310 u32 N = div_u64(H, U);
312 /* "H - (N * U)" is just "H % U" so it's bound to u32 */
313 si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
314 si->dev *= layout->mirrors_p1;
316 div_u64_rem(file_offset, stripe_unit, &si->unit_off);
318 si->obj_offset = si->unit_off + (N * stripe_unit) +
319 (M * group_depth * stripe_unit);
321 si->group_length = T - H;
325 static int _add_stripe_unit(struct ore_io_state *ios, unsigned *cur_pg,
326 unsigned pgbase, struct ore_per_dev_state *per_dev,
329 unsigned pg = *cur_pg;
330 struct request_queue *q =
331 osd_request_queue(_ios_od(ios, per_dev->dev));
333 per_dev->length += cur_len;
335 if (per_dev->bio == NULL) {
336 unsigned pages_in_stripe = ios->layout->group_width *
337 (ios->layout->stripe_unit / PAGE_SIZE);
338 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
339 ios->layout->group_width;
341 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
342 if (unlikely(!per_dev->bio)) {
343 ORE_DBGMSG("Failed to allocate BIO size=%u\n",
349 while (cur_len > 0) {
350 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
353 BUG_ON(ios->nr_pages <= pg);
356 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
358 if (unlikely(pglen != added_len))
369 static int _prepare_one_group(struct ore_io_state *ios, u64 length,
370 struct ore_striping_info *si)
372 unsigned stripe_unit = ios->layout->stripe_unit;
373 unsigned mirrors_p1 = ios->layout->mirrors_p1;
374 unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
375 unsigned dev = si->dev;
376 unsigned first_dev = dev - (dev % devs_in_group);
377 unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
378 unsigned cur_pg = ios->pages_consumed;
382 struct ore_per_dev_state *per_dev = &ios->per_dev[dev];
383 unsigned cur_len, page_off = 0;
385 if (!per_dev->length) {
388 per_dev->offset = si->obj_offset + stripe_unit -
390 cur_len = stripe_unit;
391 } else if (dev == si->dev) {
392 per_dev->offset = si->obj_offset;
393 cur_len = stripe_unit - si->unit_off;
394 page_off = si->unit_off & ~PAGE_MASK;
395 BUG_ON(page_off && (page_off != ios->pgbase));
396 } else { /* dev > si->dev */
397 per_dev->offset = si->obj_offset - si->unit_off;
398 cur_len = stripe_unit;
404 cur_len = stripe_unit;
406 if (cur_len >= length)
409 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
415 dev = (dev % devs_in_group) + first_dev;
420 ios->numdevs = max_comp + mirrors_p1;
421 ios->pages_consumed = cur_pg;
425 static int _prepare_for_striping(struct ore_io_state *ios)
427 u64 length = ios->length;
428 u64 offset = ios->offset;
429 struct ore_striping_info si;
433 if (ios->kern_buff) {
434 struct ore_per_dev_state *per_dev = &ios->per_dev[0];
436 ore_calc_stripe_info(ios->layout, ios->offset, &si);
437 per_dev->offset = si.obj_offset;
438 per_dev->dev = si.dev;
440 /* no cross device without page array */
441 BUG_ON((ios->layout->group_width > 1) &&
442 (si.unit_off + ios->length >
443 ios->layout->stripe_unit));
445 ios->numdevs = ios->layout->mirrors_p1;
450 ore_calc_stripe_info(ios->layout, offset, &si);
452 if (length < si.group_length)
453 si.group_length = length;
455 ret = _prepare_one_group(ios, si.group_length, &si);
459 offset += si.group_length;
460 length -= si.group_length;
467 int ore_create(struct ore_io_state *ios)
471 for (i = 0; i < ios->oc->numdevs; i++) {
472 struct osd_request *or;
474 or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
476 ORE_ERR("%s: osd_start_request failed\n", __func__);
480 ios->per_dev[i].or = or;
483 osd_req_create_object(or, _ios_obj(ios, i));
485 ret = ore_io_execute(ios);
490 EXPORT_SYMBOL(ore_create);
492 int ore_remove(struct ore_io_state *ios)
496 for (i = 0; i < ios->oc->numdevs; i++) {
497 struct osd_request *or;
499 or = osd_start_request(_ios_od(ios, i), GFP_KERNEL);
501 ORE_ERR("%s: osd_start_request failed\n", __func__);
505 ios->per_dev[i].or = or;
508 osd_req_remove_object(or, _ios_obj(ios, i));
510 ret = ore_io_execute(ios);
515 EXPORT_SYMBOL(ore_remove);
517 static int _write_mirror(struct ore_io_state *ios, int cur_comp)
519 struct ore_per_dev_state *master_dev = &ios->per_dev[cur_comp];
520 unsigned dev = ios->per_dev[cur_comp].dev;
521 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
524 if (ios->pages && !master_dev->length)
525 return 0; /* Just an empty slot */
527 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
528 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
529 struct osd_request *or;
531 or = osd_start_request(_ios_od(ios, dev), GFP_KERNEL);
533 ORE_ERR("%s: osd_start_request failed\n", __func__);
538 per_dev->offset = master_dev->offset;
543 if (per_dev != master_dev) {
544 bio = bio_kmalloc(GFP_KERNEL,
545 master_dev->bio->bi_max_vecs);
546 if (unlikely(!bio)) {
548 "Failed to allocate BIO size=%u\n",
549 master_dev->bio->bi_max_vecs);
554 __bio_clone(bio, master_dev->bio);
557 per_dev->length = master_dev->length;
561 bio = master_dev->bio;
562 /* FIXME: bio_set_dir() */
563 bio->bi_rw |= REQ_WRITE;
566 osd_req_write(or, _ios_obj(ios, dev), per_dev->offset,
567 bio, per_dev->length);
568 ORE_DBGMSG("write(0x%llx) offset=0x%llx "
569 "length=0x%llx dev=%d\n",
570 _LLU(_ios_obj(ios, dev)->id),
571 _LLU(per_dev->offset),
572 _LLU(per_dev->length), dev);
573 } else if (ios->kern_buff) {
574 ret = osd_req_write_kern(or, _ios_obj(ios, dev),
576 ios->kern_buff, ios->length);
579 ORE_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
580 "length=0x%llx dev=%d\n",
581 _LLU(_ios_obj(ios, dev)->id),
582 _LLU(per_dev->offset),
583 _LLU(ios->length), dev);
585 osd_req_set_attributes(or, _ios_obj(ios, dev));
586 ORE_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
587 _LLU(_ios_obj(ios, dev)->id),
588 ios->out_attr_len, dev);
592 osd_req_add_set_attr_list(or, ios->out_attr,
596 osd_req_add_get_attr_list(or, ios->in_attr,
604 int ore_write(struct ore_io_state *ios)
609 ret = _prepare_for_striping(ios);
613 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
614 ret = _write_mirror(ios, i);
619 ret = ore_io_execute(ios);
622 EXPORT_SYMBOL(ore_write);
624 static int _read_mirror(struct ore_io_state *ios, unsigned cur_comp)
626 struct osd_request *or;
627 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
628 struct osd_obj_id *obj = _ios_obj(ios, cur_comp);
629 unsigned first_dev = (unsigned)obj->id;
631 if (ios->pages && !per_dev->length)
632 return 0; /* Just an empty slot */
634 first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
635 or = osd_start_request(_ios_od(ios, first_dev), GFP_KERNEL);
637 ORE_ERR("%s: osd_start_request failed\n", __func__);
643 osd_req_read(or, obj, per_dev->offset,
644 per_dev->bio, per_dev->length);
645 ORE_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
646 " dev=%d\n", _LLU(obj->id),
647 _LLU(per_dev->offset), _LLU(per_dev->length),
649 } else if (ios->kern_buff) {
650 int ret = osd_req_read_kern(or, obj, per_dev->offset,
651 ios->kern_buff, ios->length);
652 ORE_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
653 "length=0x%llx dev=%d ret=>%d\n",
654 _LLU(obj->id), _LLU(per_dev->offset),
655 _LLU(ios->length), first_dev, ret);
659 osd_req_get_attributes(or, obj);
660 ORE_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
662 ios->in_attr_len, first_dev);
665 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
668 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
673 int ore_read(struct ore_io_state *ios)
678 ret = _prepare_for_striping(ios);
682 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
683 ret = _read_mirror(ios, i);
688 ret = ore_io_execute(ios);
691 EXPORT_SYMBOL(ore_read);
693 int extract_attr_from_ios(struct ore_io_state *ios, struct osd_attr *attr)
695 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
701 osd_req_decode_get_attr_list(ios->per_dev[0].or,
702 &cur_attr, &nelem, &iter);
703 if ((cur_attr.attr_page == attr->attr_page) &&
704 (cur_attr.attr_id == attr->attr_id)) {
705 attr->len = cur_attr.len;
706 attr->val_ptr = cur_attr.val_ptr;
713 EXPORT_SYMBOL(extract_attr_from_ios);
715 static int _truncate_mirrors(struct ore_io_state *ios, unsigned cur_comp,
716 struct osd_attr *attr)
718 int last_comp = cur_comp + ios->layout->mirrors_p1;
720 for (; cur_comp < last_comp; ++cur_comp) {
721 struct ore_per_dev_state *per_dev = &ios->per_dev[cur_comp];
722 struct osd_request *or;
724 or = osd_start_request(_ios_od(ios, cur_comp), GFP_KERNEL);
726 ORE_ERR("%s: osd_start_request failed\n", __func__);
731 osd_req_set_attributes(or, _ios_obj(ios, cur_comp));
732 osd_req_add_set_attr_list(or, attr, 1);
739 struct ore_striping_info si;
740 u64 prev_group_obj_off;
741 u64 next_group_obj_off;
743 unsigned first_group_dev;
744 unsigned nex_group_dev;
748 static void _calc_trunk_info(struct ore_layout *layout, u64 file_offset,
749 struct _trunc_info *ti)
751 unsigned stripe_unit = layout->stripe_unit;
753 ore_calc_stripe_info(layout, file_offset, &ti->si);
755 ti->prev_group_obj_off = ti->si.M * stripe_unit;
756 ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
758 ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
759 ti->nex_group_dev = ti->first_group_dev + layout->group_width;
760 ti->max_devs = layout->group_width * layout->group_count;
763 int ore_truncate(struct ore_layout *layout, struct ore_components *oc,
766 struct ore_io_state *ios;
767 struct exofs_trunc_attr {
768 struct osd_attr attr;
771 struct _trunc_info ti;
774 ret = ore_get_io_state(layout, oc, &ios);
778 _calc_trunk_info(ios->layout, size, &ti);
780 size_attrs = kcalloc(ti.max_devs, sizeof(*size_attrs),
782 if (unlikely(!size_attrs)) {
787 ios->numdevs = ios->oc->numdevs;
789 for (i = 0; i < ti.max_devs; ++i) {
790 struct exofs_trunc_attr *size_attr = &size_attrs[i];
793 if (i < ti.first_group_dev)
794 obj_size = ti.prev_group_obj_off;
795 else if (i >= ti.nex_group_dev)
796 obj_size = ti.next_group_obj_off;
797 else if (i < ti.si.dev) /* dev within this group */
798 obj_size = ti.si.obj_offset +
799 ios->layout->stripe_unit - ti.si.unit_off;
800 else if (i == ti.si.dev)
801 obj_size = ti.si.obj_offset;
802 else /* i > ti.dev */
803 obj_size = ti.si.obj_offset - ti.si.unit_off;
805 size_attr->newsize = cpu_to_be64(obj_size);
806 size_attr->attr = g_attr_logical_length;
807 size_attr->attr.val_ptr = &size_attr->newsize;
809 ORE_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
810 _LLU(oc->comps->obj.id), _LLU(obj_size), i);
811 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
816 ret = ore_io_execute(ios);
820 ore_put_io_state(ios);
823 EXPORT_SYMBOL(ore_truncate);
825 const struct osd_attr g_attr_logical_length = ATTR_DEF(
826 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
827 EXPORT_SYMBOL(g_attr_logical_length);