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 <scsi/scsi_device.h>
26 #include <asm/div64.h>
30 #define EXOFS_DBGMSG2(M...) do {} while (0)
31 /* #define EXOFS_DBGMSG2 EXOFS_DBGMSG */
33 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
35 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
38 int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
39 u64 offset, void *p, unsigned length)
41 struct osd_request *or = osd_start_request(od, GFP_KERNEL);
42 /* struct osd_sense_info osi = {.key = 0};*/
46 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
49 ret = osd_req_read_kern(or, obj, offset, p, length);
51 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
55 ret = osd_finalize_request(or, 0, cred, NULL);
57 EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n", ret);
61 ret = osd_execute_request(or);
63 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
64 /* osd_req_decode_sense(or, ret); */
71 int exofs_get_io_state(struct exofs_layout *layout,
72 struct exofs_io_state **pios)
74 struct exofs_io_state *ios;
76 /*TODO: Maybe use kmem_cach per sbi of size
77 * exofs_io_state_size(layout->s_numdevs)
79 ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL);
81 EXOFS_DBGMSG("Faild kzalloc bytes=%d\n",
82 exofs_io_state_size(layout->s_numdevs));
88 ios->obj.partition = layout->s_pid;
93 void exofs_put_io_state(struct exofs_io_state *ios)
98 for (i = 0; i < ios->numdevs; i++) {
99 struct exofs_per_dev_state *per_dev = &ios->per_dev[i];
102 osd_end_request(per_dev->or);
104 bio_put(per_dev->bio);
111 unsigned exofs_layout_od_id(struct exofs_layout *layout,
112 osd_id obj_no, unsigned layout_index)
114 /* switch (layout->lay_func) {
115 case LAYOUT_MOVING_WINDOW:
117 unsigned dev_mod = obj_no;
119 return (layout_index + dev_mod * layout->mirrors_p1) %
122 case LAYOUT_FUNC_IMPLICT:
123 return layout->devs[layout_index];
127 static inline struct osd_dev *exofs_ios_od(struct exofs_io_state *ios,
128 unsigned layout_index)
130 return ios->layout->s_ods[
131 exofs_layout_od_id(ios->layout, ios->obj.id, layout_index)];
134 static void _sync_done(struct exofs_io_state *ios, void *p)
136 struct completion *waiting = p;
141 static void _last_io(struct kref *kref)
143 struct exofs_io_state *ios = container_of(
144 kref, struct exofs_io_state, kref);
146 ios->done(ios, ios->private);
149 static void _done_io(struct osd_request *or, void *p)
151 struct exofs_io_state *ios = p;
153 kref_put(&ios->kref, _last_io);
156 static int exofs_io_execute(struct exofs_io_state *ios)
158 DECLARE_COMPLETION_ONSTACK(wait);
159 bool sync = (ios->done == NULL);
163 ios->done = _sync_done;
164 ios->private = &wait;
167 for (i = 0; i < ios->numdevs; i++) {
168 struct osd_request *or = ios->per_dev[i].or;
172 ret = osd_finalize_request(or, 0, ios->cred, NULL);
174 EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n",
180 kref_init(&ios->kref);
182 for (i = 0; i < ios->numdevs; i++) {
183 struct osd_request *or = ios->per_dev[i].or;
187 kref_get(&ios->kref);
188 osd_execute_request_async(or, _done_io, ios);
191 kref_put(&ios->kref, _last_io);
195 wait_for_completion(&wait);
196 ret = exofs_check_io(ios, NULL);
201 static void _clear_bio(struct bio *bio)
206 __bio_for_each_segment(bv, bio, i, 0) {
207 unsigned this_count = bv->bv_len;
209 if (likely(PAGE_SIZE == this_count))
210 clear_highpage(bv->bv_page);
212 zero_user(bv->bv_page, bv->bv_offset, this_count);
216 int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
218 enum osd_err_priority acumulated_osd_err = 0;
219 int acumulated_lin_err = 0;
222 for (i = 0; i < ios->numdevs; i++) {
223 struct osd_sense_info osi;
224 struct osd_request *or = ios->per_dev[i].or;
230 ret = osd_req_decode_sense(or, &osi);
234 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
235 /* start read offset passed endof file */
236 _clear_bio(ios->per_dev[i].bio);
237 EXOFS_DBGMSG("start read offset passed end of file "
238 "offset=0x%llx, length=0x%llx\n",
239 _LLU(ios->per_dev[i].offset),
240 _LLU(ios->per_dev[i].length));
242 continue; /* we recovered */
245 if (osi.osd_err_pri >= acumulated_osd_err) {
246 acumulated_osd_err = osi.osd_err_pri;
247 acumulated_lin_err = ret;
251 /* TODO: raid specific residual calculations */
253 if (likely(!acumulated_lin_err))
256 *resid = ios->length;
259 return acumulated_lin_err;
262 /* REMOVEME: After review
263 Some quoteing from the standard
265 L = logical offset into the file
266 W = number of data components in a stripe
267 S = W * stripe_unit (S is Stripe length)
268 N = L / S (N is the stripe Number)
269 C = (L-(N*S)) / stripe_unit (C is the component)
270 O = (N*stripe_unit)+(L%stripe_unit) (O is the object's offset)
273 static void _offset_dev_unit_off(struct exofs_io_state *ios, u64 file_offset,
274 u64 *obj_offset, unsigned *dev, unsigned *unit_off)
276 unsigned stripe_unit = ios->layout->stripe_unit;
277 unsigned stripe_length = stripe_unit * ios->layout->group_width;
278 u64 stripe_no = file_offset;
279 unsigned stripe_mod = do_div(stripe_no, stripe_length);
281 *unit_off = stripe_mod % stripe_unit;
282 *obj_offset = stripe_no * stripe_unit + *unit_off;
283 *dev = stripe_mod / stripe_unit * ios->layout->mirrors_p1;
286 static int _add_stripe_unit(struct exofs_io_state *ios, unsigned *cur_pg,
287 unsigned pgbase, struct exofs_per_dev_state *per_dev,
290 unsigned pg = *cur_pg;
291 struct request_queue *q =
292 osd_request_queue(exofs_ios_od(ios, per_dev->dev));
294 per_dev->length += cur_len;
296 if (per_dev->bio == NULL) {
297 unsigned pages_in_stripe = ios->layout->group_width *
298 (ios->layout->stripe_unit / PAGE_SIZE);
299 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
300 ios->layout->group_width;
302 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
303 if (unlikely(!per_dev->bio)) {
304 EXOFS_DBGMSG("Faild to allocate BIO size=%u\n",
310 while (cur_len > 0) {
311 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
314 BUG_ON(ios->nr_pages <= pg);
317 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
319 if (unlikely(pglen != added_len))
330 static int _prepare_for_striping(struct exofs_io_state *ios)
332 u64 length = ios->length;
333 u64 offset = ios->offset;
334 unsigned stripe_unit = ios->layout->stripe_unit;
336 unsigned stripes = 0;
341 if (ios->kern_buff) {
342 struct exofs_per_dev_state *per_dev = &ios->per_dev[0];
345 _offset_dev_unit_off(ios, offset, &per_dev->offset,
346 &per_dev->dev, &unit_off);
347 /* no cross device without page array */
348 BUG_ON((ios->layout->group_width > 1) &&
349 (unit_off + length > stripe_unit));
351 ios->numdevs = ios->layout->mirrors_p1;
356 struct exofs_per_dev_state *per_dev = &ios->per_dev[comp];
357 unsigned cur_len, page_off;
359 if (!per_dev->length) {
362 _offset_dev_unit_off(ios, offset, &per_dev->offset,
363 &per_dev->dev, &unit_off);
365 cur_len = min_t(u64, stripe_unit - unit_off, length);
367 page_off = unit_off & ~PAGE_MASK;
368 BUG_ON(page_off != ios->pgbase);
370 cur_len = min_t(u64, stripe_unit, length);
374 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
379 comp += ios->layout->mirrors_p1;
380 comp %= ios->layout->s_numdevs;
385 ios->numdevs = stripes * ios->layout->mirrors_p1;
389 int exofs_sbi_create(struct exofs_io_state *ios)
393 for (i = 0; i < ios->layout->s_numdevs; i++) {
394 struct osd_request *or;
396 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
398 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
402 ios->per_dev[i].or = or;
405 osd_req_create_object(or, &ios->obj);
407 ret = exofs_io_execute(ios);
413 int exofs_sbi_remove(struct exofs_io_state *ios)
417 for (i = 0; i < ios->layout->s_numdevs; i++) {
418 struct osd_request *or;
420 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
422 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
426 ios->per_dev[i].or = or;
429 osd_req_remove_object(or, &ios->obj);
431 ret = exofs_io_execute(ios);
437 static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp)
439 struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp];
440 unsigned dev = ios->per_dev[cur_comp].dev;
441 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
444 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
445 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
446 struct osd_request *or;
448 or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL);
450 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
455 per_dev->offset = master_dev->offset;
460 if (per_dev != master_dev) {
461 bio = bio_kmalloc(GFP_KERNEL,
462 master_dev->bio->bi_max_vecs);
463 if (unlikely(!bio)) {
465 "Faild to allocate BIO size=%u\n",
466 master_dev->bio->bi_max_vecs);
471 __bio_clone(bio, master_dev->bio);
474 per_dev->length = master_dev->length;
478 bio = master_dev->bio;
479 /* FIXME: bio_set_dir() */
480 bio->bi_rw |= (1 << BIO_RW);
483 osd_req_write(or, &ios->obj, per_dev->offset, bio,
485 EXOFS_DBGMSG("write(0x%llx) offset=0x%llx "
486 "length=0x%llx dev=%d\n",
487 _LLU(ios->obj.id), _LLU(per_dev->offset),
488 _LLU(per_dev->length), dev);
489 } else if (ios->kern_buff) {
490 ret = osd_req_write_kern(or, &ios->obj, per_dev->offset,
491 ios->kern_buff, ios->length);
494 EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
495 "length=0x%llx dev=%d\n",
496 _LLU(ios->obj.id), _LLU(per_dev->offset),
497 _LLU(ios->length), dev);
499 osd_req_set_attributes(or, &ios->obj);
500 EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
501 _LLU(ios->obj.id), ios->out_attr_len, dev);
505 osd_req_add_set_attr_list(or, ios->out_attr,
509 osd_req_add_get_attr_list(or, ios->in_attr,
517 int exofs_sbi_write(struct exofs_io_state *ios)
522 ret = _prepare_for_striping(ios);
526 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
527 ret = _sbi_write_mirror(ios, i);
532 ret = exofs_io_execute(ios);
536 static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp)
538 struct osd_request *or;
539 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
540 unsigned first_dev = (unsigned)ios->obj.id;
542 first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
543 or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL);
545 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
551 osd_req_read(or, &ios->obj, per_dev->offset,
552 per_dev->bio, per_dev->length);
553 EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
554 " dev=%d\n", _LLU(ios->obj.id),
555 _LLU(per_dev->offset), _LLU(per_dev->length),
557 } else if (ios->kern_buff) {
558 int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset,
559 ios->kern_buff, ios->length);
560 EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
561 "length=0x%llx dev=%d ret=>%d\n",
562 _LLU(ios->obj.id), _LLU(per_dev->offset),
563 _LLU(ios->length), first_dev, ret);
567 osd_req_get_attributes(or, &ios->obj);
568 EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
569 _LLU(ios->obj.id), ios->in_attr_len, first_dev);
572 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
575 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
580 int exofs_sbi_read(struct exofs_io_state *ios)
585 ret = _prepare_for_striping(ios);
589 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
590 ret = _sbi_read_mirror(ios, i);
595 ret = exofs_io_execute(ios);
599 int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr)
601 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
607 osd_req_decode_get_attr_list(ios->per_dev[0].or,
608 &cur_attr, &nelem, &iter);
609 if ((cur_attr.attr_page == attr->attr_page) &&
610 (cur_attr.attr_id == attr->attr_id)) {
611 attr->len = cur_attr.len;
612 attr->val_ptr = cur_attr.val_ptr;
620 static int _truncate_mirrors(struct exofs_io_state *ios, unsigned cur_comp,
621 struct osd_attr *attr)
623 int last_comp = cur_comp + ios->layout->mirrors_p1;
625 for (; cur_comp < last_comp; ++cur_comp) {
626 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
627 struct osd_request *or;
629 or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL);
631 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
636 osd_req_set_attributes(or, &ios->obj);
637 osd_req_add_set_attr_list(or, attr, 1);
643 int exofs_oi_truncate(struct exofs_i_info *oi, u64 size)
645 struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info;
646 struct exofs_io_state *ios;
647 struct exofs_trunc_attr {
648 struct osd_attr attr;
656 ret = exofs_get_io_state(&sbi->layout, &ios);
660 size_attrs = kcalloc(ios->layout->group_width, sizeof(*size_attrs),
662 if (unlikely(!size_attrs)) {
667 ios->obj.id = exofs_oi_objno(oi);
668 ios->cred = oi->i_cred;
670 ios->numdevs = ios->layout->s_numdevs;
671 _offset_dev_unit_off(ios, size, &this_obj_size, &dev, &unit_off);
673 for (i = 0; i < ios->layout->group_width; ++i) {
674 struct exofs_trunc_attr *size_attr = &size_attrs[i];
678 obj_size = this_obj_size +
679 ios->layout->stripe_unit - unit_off;
681 obj_size = this_obj_size;
683 obj_size = this_obj_size - unit_off;
685 size_attr->newsize = cpu_to_be64(obj_size);
686 size_attr->attr = g_attr_logical_length;
687 size_attr->attr.val_ptr = &size_attr->newsize;
689 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
694 ret = exofs_io_execute(ios);
698 exofs_put_io_state(ios);