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 <scsi/scsi_device.h>
27 #include <asm/div64.h>
31 #define EXOFS_DBGMSG2(M...) do {} while (0)
32 /* #define EXOFS_DBGMSG2 EXOFS_DBGMSG */
34 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
36 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
39 int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
40 u64 offset, void *p, unsigned length)
42 struct osd_request *or = osd_start_request(od, GFP_KERNEL);
43 /* struct osd_sense_info osi = {.key = 0};*/
47 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
50 ret = osd_req_read_kern(or, obj, offset, p, length);
52 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
56 ret = osd_finalize_request(or, 0, cred, NULL);
58 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
62 ret = osd_execute_request(or);
64 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
65 /* osd_req_decode_sense(or, ret); */
72 int exofs_get_io_state(struct exofs_layout *layout,
73 struct exofs_io_state **pios)
75 struct exofs_io_state *ios;
77 /*TODO: Maybe use kmem_cach per sbi of size
78 * exofs_io_state_size(layout->s_numdevs)
80 ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL);
82 EXOFS_DBGMSG("Failed kzalloc bytes=%d\n",
83 exofs_io_state_size(layout->s_numdevs));
89 ios->obj.partition = layout->s_pid;
94 void exofs_put_io_state(struct exofs_io_state *ios)
99 for (i = 0; i < ios->numdevs; i++) {
100 struct exofs_per_dev_state *per_dev = &ios->per_dev[i];
103 osd_end_request(per_dev->or);
105 bio_put(per_dev->bio);
112 unsigned exofs_layout_od_id(struct exofs_layout *layout,
113 osd_id obj_no, unsigned layout_index)
115 /* switch (layout->lay_func) {
116 case LAYOUT_MOVING_WINDOW:
118 unsigned dev_mod = obj_no;
120 return (layout_index + dev_mod * layout->mirrors_p1) %
123 case LAYOUT_FUNC_IMPLICT:
124 return layout->devs[layout_index];
128 static inline struct osd_dev *exofs_ios_od(struct exofs_io_state *ios,
129 unsigned layout_index)
131 return ios->layout->s_ods[
132 exofs_layout_od_id(ios->layout, ios->obj.id, layout_index)];
135 static void _sync_done(struct exofs_io_state *ios, void *p)
137 struct completion *waiting = p;
142 static void _last_io(struct kref *kref)
144 struct exofs_io_state *ios = container_of(
145 kref, struct exofs_io_state, kref);
147 ios->done(ios, ios->private);
150 static void _done_io(struct osd_request *or, void *p)
152 struct exofs_io_state *ios = p;
154 kref_put(&ios->kref, _last_io);
157 static int exofs_io_execute(struct exofs_io_state *ios)
159 DECLARE_COMPLETION_ONSTACK(wait);
160 bool sync = (ios->done == NULL);
164 ios->done = _sync_done;
165 ios->private = &wait;
168 for (i = 0; i < ios->numdevs; i++) {
169 struct osd_request *or = ios->per_dev[i].or;
173 ret = osd_finalize_request(or, 0, ios->cred, NULL);
175 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n",
181 kref_init(&ios->kref);
183 for (i = 0; i < ios->numdevs; i++) {
184 struct osd_request *or = ios->per_dev[i].or;
188 kref_get(&ios->kref);
189 osd_execute_request_async(or, _done_io, ios);
192 kref_put(&ios->kref, _last_io);
196 wait_for_completion(&wait);
197 ret = exofs_check_io(ios, NULL);
202 static void _clear_bio(struct bio *bio)
207 __bio_for_each_segment(bv, bio, i, 0) {
208 unsigned this_count = bv->bv_len;
210 if (likely(PAGE_SIZE == this_count))
211 clear_highpage(bv->bv_page);
213 zero_user(bv->bv_page, bv->bv_offset, this_count);
217 int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
219 enum osd_err_priority acumulated_osd_err = 0;
220 int acumulated_lin_err = 0;
223 for (i = 0; i < ios->numdevs; i++) {
224 struct osd_sense_info osi;
225 struct osd_request *or = ios->per_dev[i].or;
231 ret = osd_req_decode_sense(or, &osi);
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));
243 continue; /* we recovered */
246 if (osi.osd_err_pri >= acumulated_osd_err) {
247 acumulated_osd_err = osi.osd_err_pri;
248 acumulated_lin_err = ret;
252 /* TODO: raid specific residual calculations */
254 if (likely(!acumulated_lin_err))
257 *resid = ios->length;
260 return acumulated_lin_err;
264 * L - logical offset into the file
266 * U - The number of bytes in a stripe within a group
268 * U = stripe_unit * group_width
270 * T - The number of bytes striped within a group of component objects
271 * (before advancing to the next group)
273 * T = stripe_unit * group_width * group_depth
275 * S - The number of bytes striped across all component objects
276 * before the pattern repeats
278 * S = stripe_unit * group_width * group_depth * group_count
280 * M - The "major" (i.e., across all components) stripe number
284 * G - Counts the groups from the beginning of the major stripe
286 * G = (L - (M * S)) / T [or (L % S) / T]
288 * H - The byte offset within the group
290 * H = (L - (M * S)) % T [or (L % S) % T]
292 * N - The "minor" (i.e., across the group) stripe number
296 * C - The component index coresponding to L
298 * C = (H - (N * U)) / stripe_unit + G * group_width
299 * [or (L % U) / stripe_unit + G * group_width]
301 * O - The component offset coresponding to L
303 * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
305 struct _striping_info {
308 u64 M; /* for truncate */
313 static void _calc_stripe_info(struct exofs_layout *layout, u64 file_offset,
314 struct _striping_info *si)
316 u32 stripe_unit = layout->stripe_unit;
317 u32 group_width = layout->group_width;
318 u64 group_depth = layout->group_depth;
320 u32 U = stripe_unit * group_width;
321 u64 T = U * group_depth;
322 u64 S = T * layout->group_count;
323 u64 M = div64_u64(file_offset, S);
326 G = (L - (M * S)) / T
327 H = (L - (M * S)) % T
329 u64 LmodS = file_offset - M * S;
330 u32 G = div64_u64(LmodS, T);
331 u64 H = LmodS - G * T;
333 u32 N = div_u64(H, U);
335 /* "H - (N * U)" is just "H % U" so it's bound to u32 */
336 si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
337 si->dev *= layout->mirrors_p1;
339 div_u64_rem(file_offset, stripe_unit, &si->unit_off);
341 si->obj_offset = si->unit_off + (N * stripe_unit) +
342 (M * group_depth * stripe_unit);
344 si->group_length = T - H;
348 static int _add_stripe_unit(struct exofs_io_state *ios, unsigned *cur_pg,
349 unsigned pgbase, struct exofs_per_dev_state *per_dev,
352 unsigned pg = *cur_pg;
353 struct request_queue *q =
354 osd_request_queue(exofs_ios_od(ios, per_dev->dev));
356 per_dev->length += cur_len;
358 if (per_dev->bio == NULL) {
359 unsigned pages_in_stripe = ios->layout->group_width *
360 (ios->layout->stripe_unit / PAGE_SIZE);
361 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
362 ios->layout->group_width;
364 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
365 if (unlikely(!per_dev->bio)) {
366 EXOFS_DBGMSG("Failed to allocate BIO size=%u\n",
372 while (cur_len > 0) {
373 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
376 BUG_ON(ios->nr_pages <= pg);
379 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
381 if (unlikely(pglen != added_len))
392 static int _prepare_one_group(struct exofs_io_state *ios, u64 length,
393 struct _striping_info *si)
395 unsigned stripe_unit = ios->layout->stripe_unit;
396 unsigned mirrors_p1 = ios->layout->mirrors_p1;
397 unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
398 unsigned dev = si->dev;
399 unsigned first_dev = dev - (dev % devs_in_group);
400 unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
401 unsigned cur_pg = ios->pages_consumed;
405 struct exofs_per_dev_state *per_dev = &ios->per_dev[dev];
406 unsigned cur_len, page_off = 0;
408 if (!per_dev->length) {
411 per_dev->offset = si->obj_offset + stripe_unit -
413 cur_len = stripe_unit;
414 } else if (dev == si->dev) {
415 per_dev->offset = si->obj_offset;
416 cur_len = stripe_unit - si->unit_off;
417 page_off = si->unit_off & ~PAGE_MASK;
418 BUG_ON(page_off && (page_off != ios->pgbase));
419 } else { /* dev > si->dev */
420 per_dev->offset = si->obj_offset - si->unit_off;
421 cur_len = stripe_unit;
427 cur_len = stripe_unit;
429 if (cur_len >= length)
432 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
438 dev = (dev % devs_in_group) + first_dev;
443 ios->numdevs = max_comp + mirrors_p1;
444 ios->pages_consumed = cur_pg;
448 static int _prepare_for_striping(struct exofs_io_state *ios)
450 u64 length = ios->length;
451 u64 offset = ios->offset;
452 struct _striping_info si;
456 if (ios->kern_buff) {
457 struct exofs_per_dev_state *per_dev = &ios->per_dev[0];
459 _calc_stripe_info(ios->layout, ios->offset, &si);
460 per_dev->offset = si.obj_offset;
461 per_dev->dev = si.dev;
463 /* no cross device without page array */
464 BUG_ON((ios->layout->group_width > 1) &&
465 (si.unit_off + ios->length >
466 ios->layout->stripe_unit));
468 ios->numdevs = ios->layout->mirrors_p1;
473 _calc_stripe_info(ios->layout, offset, &si);
475 if (length < si.group_length)
476 si.group_length = length;
478 ret = _prepare_one_group(ios, si.group_length, &si);
482 offset += si.group_length;
483 length -= si.group_length;
490 int exofs_sbi_create(struct exofs_io_state *ios)
494 for (i = 0; i < ios->layout->s_numdevs; i++) {
495 struct osd_request *or;
497 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
499 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
503 ios->per_dev[i].or = or;
506 osd_req_create_object(or, &ios->obj);
508 ret = exofs_io_execute(ios);
514 int exofs_sbi_remove(struct exofs_io_state *ios)
518 for (i = 0; i < ios->layout->s_numdevs; i++) {
519 struct osd_request *or;
521 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
523 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
527 ios->per_dev[i].or = or;
530 osd_req_remove_object(or, &ios->obj);
532 ret = exofs_io_execute(ios);
538 static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp)
540 struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp];
541 unsigned dev = ios->per_dev[cur_comp].dev;
542 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
545 if (ios->pages && !master_dev->length)
546 return 0; /* Just an empty slot */
548 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
549 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
550 struct osd_request *or;
552 or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL);
554 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
559 per_dev->offset = master_dev->offset;
564 if (per_dev != master_dev) {
565 bio = bio_kmalloc(GFP_KERNEL,
566 master_dev->bio->bi_max_vecs);
567 if (unlikely(!bio)) {
569 "Failed to allocate BIO size=%u\n",
570 master_dev->bio->bi_max_vecs);
575 __bio_clone(bio, master_dev->bio);
578 per_dev->length = master_dev->length;
582 bio = master_dev->bio;
583 /* FIXME: bio_set_dir() */
584 bio->bi_rw |= REQ_WRITE;
587 osd_req_write(or, &ios->obj, per_dev->offset, bio,
589 EXOFS_DBGMSG("write(0x%llx) offset=0x%llx "
590 "length=0x%llx dev=%d\n",
591 _LLU(ios->obj.id), _LLU(per_dev->offset),
592 _LLU(per_dev->length), dev);
593 } else if (ios->kern_buff) {
594 ret = osd_req_write_kern(or, &ios->obj, per_dev->offset,
595 ios->kern_buff, ios->length);
598 EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
599 "length=0x%llx dev=%d\n",
600 _LLU(ios->obj.id), _LLU(per_dev->offset),
601 _LLU(ios->length), dev);
603 osd_req_set_attributes(or, &ios->obj);
604 EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
605 _LLU(ios->obj.id), ios->out_attr_len, dev);
609 osd_req_add_set_attr_list(or, ios->out_attr,
613 osd_req_add_get_attr_list(or, ios->in_attr,
621 int exofs_sbi_write(struct exofs_io_state *ios)
626 ret = _prepare_for_striping(ios);
630 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
631 ret = _sbi_write_mirror(ios, i);
636 ret = exofs_io_execute(ios);
640 static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp)
642 struct osd_request *or;
643 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
644 unsigned first_dev = (unsigned)ios->obj.id;
646 if (ios->pages && !per_dev->length)
647 return 0; /* Just an empty slot */
649 first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
650 or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL);
652 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
658 osd_req_read(or, &ios->obj, per_dev->offset,
659 per_dev->bio, per_dev->length);
660 EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
661 " dev=%d\n", _LLU(ios->obj.id),
662 _LLU(per_dev->offset), _LLU(per_dev->length),
664 } else if (ios->kern_buff) {
665 int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset,
666 ios->kern_buff, ios->length);
667 EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
668 "length=0x%llx dev=%d ret=>%d\n",
669 _LLU(ios->obj.id), _LLU(per_dev->offset),
670 _LLU(ios->length), first_dev, ret);
674 osd_req_get_attributes(or, &ios->obj);
675 EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
676 _LLU(ios->obj.id), ios->in_attr_len, first_dev);
679 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
682 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
687 int exofs_sbi_read(struct exofs_io_state *ios)
692 ret = _prepare_for_striping(ios);
696 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
697 ret = _sbi_read_mirror(ios, i);
702 ret = exofs_io_execute(ios);
706 int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr)
708 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
714 osd_req_decode_get_attr_list(ios->per_dev[0].or,
715 &cur_attr, &nelem, &iter);
716 if ((cur_attr.attr_page == attr->attr_page) &&
717 (cur_attr.attr_id == attr->attr_id)) {
718 attr->len = cur_attr.len;
719 attr->val_ptr = cur_attr.val_ptr;
727 static int _truncate_mirrors(struct exofs_io_state *ios, unsigned cur_comp,
728 struct osd_attr *attr)
730 int last_comp = cur_comp + ios->layout->mirrors_p1;
732 for (; cur_comp < last_comp; ++cur_comp) {
733 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
734 struct osd_request *or;
736 or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL);
738 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
743 osd_req_set_attributes(or, &ios->obj);
744 osd_req_add_set_attr_list(or, attr, 1);
751 struct _striping_info si;
752 u64 prev_group_obj_off;
753 u64 next_group_obj_off;
755 unsigned first_group_dev;
756 unsigned nex_group_dev;
760 void _calc_trunk_info(struct exofs_layout *layout, u64 file_offset,
761 struct _trunc_info *ti)
763 unsigned stripe_unit = layout->stripe_unit;
765 _calc_stripe_info(layout, file_offset, &ti->si);
767 ti->prev_group_obj_off = ti->si.M * stripe_unit;
768 ti->next_group_obj_off = ti->si.M ? (ti->si.M - 1) * stripe_unit : 0;
770 ti->first_group_dev = ti->si.dev - (ti->si.dev % layout->group_width);
771 ti->nex_group_dev = ti->first_group_dev + layout->group_width;
772 ti->max_devs = layout->group_width * layout->group_count;
775 int exofs_oi_truncate(struct exofs_i_info *oi, u64 size)
777 struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info;
778 struct exofs_io_state *ios;
779 struct exofs_trunc_attr {
780 struct osd_attr attr;
783 struct _trunc_info ti;
786 ret = exofs_get_io_state(&sbi->layout, &ios);
790 _calc_trunk_info(ios->layout, size, &ti);
792 size_attrs = kcalloc(ti.max_devs, sizeof(*size_attrs),
794 if (unlikely(!size_attrs)) {
799 ios->obj.id = exofs_oi_objno(oi);
800 ios->cred = oi->i_cred;
801 ios->numdevs = ios->layout->s_numdevs;
803 for (i = 0; i < ti.max_devs; ++i) {
804 struct exofs_trunc_attr *size_attr = &size_attrs[i];
807 if (i < ti.first_group_dev)
808 obj_size = ti.prev_group_obj_off;
809 else if (i >= ti.nex_group_dev)
810 obj_size = ti.next_group_obj_off;
811 else if (i < ti.si.dev) /* dev within this group */
812 obj_size = ti.si.obj_offset +
813 ios->layout->stripe_unit - ti.si.unit_off;
814 else if (i == ti.si.dev)
815 obj_size = ti.si.obj_offset;
816 else /* i > ti.dev */
817 obj_size = ti.si.obj_offset - ti.si.unit_off;
819 size_attr->newsize = cpu_to_be64(obj_size);
820 size_attr->attr = g_attr_logical_length;
821 size_attr->attr.val_ptr = &size_attr->newsize;
823 EXOFS_DBGMSG("trunc(0x%llx) obj_offset=0x%llx dev=%d\n",
824 _LLU(ios->obj.id), _LLU(obj_size), i);
825 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
830 ret = exofs_io_execute(ios);
834 exofs_put_io_state(ios);