SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/exofs/ios.c

   1 /*
   2  * Copyright (C) 2005, 2006
   3  * Avishay Traeger (avishay@gmail.com)
   4  * Copyright (C) 2008, 2009
   5  * Boaz Harrosh <bharrosh@panasas.com>
   6  *
   7  * This file is part of exofs.
   8  *
   9  * exofs is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation.  Since it is based on ext2, and the only
  12  * valid version of GPL for the Linux kernel is version 2, the only valid
  13  * version of GPL for exofs is version 2.
  14  *
  15  * exofs is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with exofs; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  23  */
  24
  25 #include <scsi/scsi_device.h>
  26
  27 #include "exofs.h"
  28
  29 #define EXOFS_DBGMSG2(M...) do {} while (0)
  30 /* #define EXOFS_DBGMSG2 EXOFS_DBGMSG */
  31
  32 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
  33 {
  34         osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
  35 }
  36
  37 int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
  38                     u64 offset, void *p, unsigned length)
  39 {
  40         struct osd_request *or = osd_start_request(od, GFP_KERNEL);
  41 /*      struct osd_sense_info osi = {.key = 0};*/
  42         int ret;
  43
  44         if (unlikely(!or)) {
  45                 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
  46                 return -ENOMEM;
  47         }
  48         ret = osd_req_read_kern(or, obj, offset, p, length);
  49         if (unlikely(ret)) {
  50                 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
  51                 goto out;
  52         }
  53
  54         ret = osd_finalize_request(or, 0, cred, NULL);
  55         if (unlikely(ret)) {
  56                 EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n", ret);
  57                 goto out;
  58         }
  59
  60         ret = osd_execute_request(or);
  61         if (unlikely(ret))
  62                 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
  63         /* osd_req_decode_sense(or, ret); */
  64
  65 out:
  66         osd_end_request(or);
  67         return ret;
  68 }
  69
  70 int exofs_get_io_state(struct exofs_layout *layout,
  71                        struct exofs_io_state **pios)
  72 {
  73         struct exofs_io_state *ios;
  74
  75         /*TODO: Maybe use kmem_cach per sbi of size
  76          * exofs_io_state_size(layout->s_numdevs)
  77          */
  78         ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL);
  79         if (unlikely(!ios)) {
  80                 EXOFS_DBGMSG("Faild kzalloc bytes=%d\n",
  81                              exofs_io_state_size(layout->s_numdevs));
  82                 *pios = NULL;
  83                 return -ENOMEM;
  84         }
  85
  86         ios->layout = layout;
  87         ios->obj.partition = layout->s_pid;
  88         *pios = ios;
  89         return 0;
  90 }
  91
  92 void exofs_put_io_state(struct exofs_io_state *ios)
  93 {
  94         if (ios) {
  95                 unsigned i;
  96
  97                 for (i = 0; i < ios->numdevs; i++) {
  98                         struct exofs_per_dev_state *per_dev = &ios->per_dev[i];
  99
 100                         if (per_dev->or)
 101                                 osd_end_request(per_dev->or);
 102                         if (per_dev->bio)
 103                                 bio_put(per_dev->bio);
 104                 }
 105
 106                 kfree(ios);
 107         }
 108 }
 109
 110 unsigned exofs_layout_od_id(struct exofs_layout *layout,
 111                             osd_id obj_no, unsigned layout_index)
 112 {
 113         return layout_index;
 114 }
 115
 116 static inline struct osd_dev *exofs_ios_od(struct exofs_io_state *ios,
 117                                            unsigned layout_index)
 118 {
 119         return ios->layout->s_ods[
 120                 exofs_layout_od_id(ios->layout, ios->obj.id, layout_index)];
 121 }
 122
 123 static void _sync_done(struct exofs_io_state *ios, void *p)
 124 {
 125         struct completion *waiting = p;
 126
 127         complete(waiting);
 128 }
 129
 130 static void _last_io(struct kref *kref)
 131 {
 132         struct exofs_io_state *ios = container_of(
 133                                         kref, struct exofs_io_state, kref);
 134
 135         ios->done(ios, ios->private);
 136 }
 137
 138 static void _done_io(struct osd_request *or, void *p)
 139 {
 140         struct exofs_io_state *ios = p;
 141
 142         kref_put(&ios->kref, _last_io);
 143 }
 144
 145 static int exofs_io_execute(struct exofs_io_state *ios)
 146 {
 147         DECLARE_COMPLETION_ONSTACK(wait);
 148         bool sync = (ios->done == NULL);
 149         int i, ret;
 150
 151         if (sync) {
 152                 ios->done = _sync_done;
 153                 ios->private = &wait;
 154         }
 155
 156         for (i = 0; i < ios->numdevs; i++) {
 157                 struct osd_request *or = ios->per_dev[i].or;
 158                 if (unlikely(!or))
 159                         continue;
 160
 161                 ret = osd_finalize_request(or, 0, ios->cred, NULL);
 162                 if (unlikely(ret)) {
 163                         EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n",
 164                                      ret);
 165                         return ret;
 166                 }
 167         }
 168
 169         kref_init(&ios->kref);
 170
 171         for (i = 0; i < ios->numdevs; i++) {
 172                 struct osd_request *or = ios->per_dev[i].or;
 173                 if (unlikely(!or))
 174                         continue;
 175
 176                 kref_get(&ios->kref);
 177                 osd_execute_request_async(or, _done_io, ios);
 178         }
 179
 180         kref_put(&ios->kref, _last_io);
 181         ret = 0;
 182
 183         if (sync) {
 184                 wait_for_completion(&wait);
 185                 ret = exofs_check_io(ios, NULL);
 186         }
 187         return ret;
 188 }
 189
 190 static void _clear_bio(struct bio *bio)
 191 {
 192         struct bio_vec *bv;
 193         unsigned i;
 194
 195         __bio_for_each_segment(bv, bio, i, 0) {
 196                 unsigned this_count = bv->bv_len;
 197
 198                 if (likely(PAGE_SIZE == this_count))
 199                         clear_highpage(bv->bv_page);
 200                 else
 201                         zero_user(bv->bv_page, bv->bv_offset, this_count);
 202         }
 203 }
 204
 205 int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
 206 {
 207         enum osd_err_priority acumulated_osd_err = 0;
 208         int acumulated_lin_err = 0;
 209         int i;
 210
 211         for (i = 0; i < ios->numdevs; i++) {
 212                 struct osd_sense_info osi;
 213                 struct osd_request *or = ios->per_dev[i].or;
 214                 int ret;
 215
 216                 if (unlikely(!or))
 217                         continue;
 218
 219                 ret = osd_req_decode_sense(or, &osi);
 220                 if (likely(!ret))
 221                         continue;
 222
 223                 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
 224                         /* start read offset passed endof file */
 225                         _clear_bio(ios->per_dev[i].bio);
 226                         EXOFS_DBGMSG("start read offset passed end of file "
 227                                 "offset=0x%llx, length=0x%llx\n",
 228                                 _LLU(ios->offset),
 229                                 _LLU(ios->length));
 230
 231                         continue; /* we recovered */
 232                 }
 233
 234                 if (osi.osd_err_pri >= acumulated_osd_err) {
 235                         acumulated_osd_err = osi.osd_err_pri;
 236                         acumulated_lin_err = ret;
 237                 }
 238         }
 239
 240         /* TODO: raid specific residual calculations */
 241         if (resid) {
 242                 if (likely(!acumulated_lin_err))
 243                         *resid = 0;
 244                 else
 245                         *resid = ios->length;
 246         }
 247
 248         return acumulated_lin_err;
 249 }
 250
 251 int exofs_sbi_create(struct exofs_io_state *ios)
 252 {
 253         int i, ret;
 254
 255         for (i = 0; i < ios->layout->s_numdevs; i++) {
 256                 struct osd_request *or;
 257
 258                 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
 259                 if (unlikely(!or)) {
 260                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 261                         ret = -ENOMEM;
 262                         goto out;
 263                 }
 264                 ios->per_dev[i].or = or;
 265                 ios->numdevs++;
 266
 267                 osd_req_create_object(or, &ios->obj);
 268         }
 269         ret = exofs_io_execute(ios);
 270
 271 out:
 272         return ret;
 273 }
 274
 275 int exofs_sbi_remove(struct exofs_io_state *ios)
 276 {
 277         int i, ret;
 278
 279         for (i = 0; i < ios->layout->s_numdevs; i++) {
 280                 struct osd_request *or;
 281
 282                 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
 283                 if (unlikely(!or)) {
 284                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 285                         ret = -ENOMEM;
 286                         goto out;
 287                 }
 288                 ios->per_dev[i].or = or;
 289                 ios->numdevs++;
 290
 291                 osd_req_remove_object(or, &ios->obj);
 292         }
 293         ret = exofs_io_execute(ios);
 294
 295 out:
 296         return ret;
 297 }
 298
 299 int exofs_sbi_write(struct exofs_io_state *ios)
 300 {
 301         int i, ret;
 302
 303         for (i = 0; i < ios->layout->s_numdevs; i++) {
 304                 struct osd_request *or;
 305
 306                 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
 307                 if (unlikely(!or)) {
 308                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 309                         ret = -ENOMEM;
 310                         goto out;
 311                 }
 312                 ios->per_dev[i].or = or;
 313                 ios->numdevs++;
 314
 315                 if (ios->bio) {
 316                         struct bio *bio;
 317
 318                         if (i != 0) {
 319                                 bio = bio_kmalloc(GFP_KERNEL,
 320                                                   ios->bio->bi_max_vecs);
 321                                 if (unlikely(!bio)) {
 322                                         EXOFS_DBGMSG(
 323                                               "Faild to allocate BIO size=%u\n",
 324                                               ios->bio->bi_max_vecs);
 325                                         ret = -ENOMEM;
 326                                         goto out;
 327                                 }
 328
 329                                 __bio_clone(bio, ios->bio);
 330                                 bio->bi_bdev = NULL;
 331                                 bio->bi_next = NULL;
 332                                 ios->per_dev[i].bio =  bio;
 333                         } else {
 334                                 bio = ios->bio;
 335                         }
 336
 337                         osd_req_write(or, &ios->obj, ios->offset, bio,
 338                                       ios->length);
 339                         EXOFS_DBGMSG("write(0x%llx) offset=0x%llx "
 340                                       "length=0x%llx dev=%d\n",
 341                                      _LLU(ios->obj.id), _LLU(ios->offset),
 342                                      _LLU(ios->length), i);
 343                 } else if (ios->kern_buff) {
 344                         osd_req_write_kern(or, &ios->obj, ios->offset,
 345                                            ios->kern_buff, ios->length);
 346                         EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
 347                                       "length=0x%llx dev=%d\n",
 348                                      _LLU(ios->obj.id), _LLU(ios->offset),
 349                                      _LLU(ios->length), i);
 350                 } else {
 351                         osd_req_set_attributes(or, &ios->obj);
 352                         EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
 353                                      _LLU(ios->obj.id), ios->out_attr_len, i);
 354                 }
 355
 356                 if (ios->out_attr)
 357                         osd_req_add_set_attr_list(or, ios->out_attr,
 358                                                   ios->out_attr_len);
 359
 360                 if (ios->in_attr)
 361                         osd_req_add_get_attr_list(or, ios->in_attr,
 362                                                   ios->in_attr_len);
 363         }
 364         ret = exofs_io_execute(ios);
 365
 366 out:
 367         return ret;
 368 }
 369
 370 int exofs_sbi_read(struct exofs_io_state *ios)
 371 {
 372         struct osd_request *or;
 373         struct exofs_per_dev_state *per_dev = &ios->per_dev[0];
 374         unsigned first_dev = (unsigned)ios->obj.id;
 375
 376         first_dev %= ios->layout->s_numdevs;
 377         or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL);
 378         if (unlikely(!or)) {
 379                 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 380                 return -ENOMEM;
 381         }
 382         per_dev->or = or;
 383         ios->numdevs++;
 384
 385         if (ios->bio) {
 386                 osd_req_read(or, &ios->obj, ios->offset, ios->bio, ios->length);
 387                 EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
 388                              " dev=%d\n", _LLU(ios->obj.id),
 389                              _LLU(ios->offset), _LLU(ios->length),
 390                              first_dev);
 391         } else if (ios->kern_buff) {
 392                 int ret = osd_req_read_kern(or, &ios->obj, ios->offset,
 393                                             ios->kern_buff, ios->length);
 394
 395                 EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
 396                               "length=0x%llx dev=%d ret=>%d\n",
 397                               _LLU(ios->obj.id), _LLU(ios->offset),
 398                               _LLU(ios->length), first_dev, ret);
 399                 if (unlikely(ret))
 400                         return ret;
 401         } else {
 402                 osd_req_get_attributes(or, &ios->obj);
 403                 EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
 404                               _LLU(ios->obj.id), ios->in_attr_len, first_dev);
 405         }
 406
 407         if (ios->out_attr)
 408                 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
 409
 410         if (ios->in_attr)
 411                 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
 412
 413         return exofs_io_execute(ios);
 414 }
 415
 416 int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr)
 417 {
 418         struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
 419         void *iter = NULL;
 420         int nelem;
 421
 422         do {
 423                 nelem = 1;
 424                 osd_req_decode_get_attr_list(ios->per_dev[0].or,
 425                                              &cur_attr, &nelem, &iter);
 426                 if ((cur_attr.attr_page == attr->attr_page) &&
 427                     (cur_attr.attr_id == attr->attr_id)) {
 428                         attr->len = cur_attr.len;
 429                         attr->val_ptr = cur_attr.val_ptr;
 430                         return 0;
 431                 }
 432         } while (iter);
 433
 434         return -EIO;
 435 }
 436
 437 int exofs_oi_truncate(struct exofs_i_info *oi, u64 size)
 438 {
 439         struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info;
 440         struct exofs_io_state *ios;
 441         struct osd_attr attr;
 442         __be64 newsize;
 443         int i, ret;
 444
 445         if (exofs_get_io_state(&sbi->layout, &ios))
 446                 return -ENOMEM;
 447
 448         ios->obj.id = exofs_oi_objno(oi);
 449         ios->cred = oi->i_cred;
 450
 451         newsize = cpu_to_be64(size);
 452         attr = g_attr_logical_length;
 453         attr.val_ptr = &newsize;
 454
 455         for (i = 0; i < sbi->layout.s_numdevs; i++) {
 456                 struct osd_request *or;
 457
 458                 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
 459                 if (unlikely(!or)) {
 460                         EXOFS_ERR("%s: osd_start_request failed\n", __func__);
 461                         ret = -ENOMEM;
 462                         goto out;
 463                 }
 464                 ios->per_dev[i].or = or;
 465                 ios->numdevs++;
 466
 467                 osd_req_set_attributes(or, &ios->obj);
 468                 osd_req_add_set_attr_list(or, &attr, 1);
 469         }
 470         ret = exofs_io_execute(ios);
 471
 472 out:
 473         exofs_put_io_state(ios);
 474         return ret;
 475 }