lustre/ofd/lproc_ofd.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/ofd/lproc_ofd.c
  32  *
  33  * This file provides functions of procfs interface for OBD Filter Device (OFD).
  34  *
  35  * Author: Andreas Dilger <andreas.dilger@intel.com>
  36  * Author: Mikhail Pershin <mike.pershin@intel.com>
  37  * Author: Johann Lombardi <johann.lombardi@intel.com>
  38  * Author: Fan Yong <fan.yong@intel.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_CLASS
  42
  43 #include <obd.h>
  44 #include <lprocfs_status.h>
  45 #include <linux/seq_file.h>
  46 #include <lustre_lfsck.h>
  47 #include <uapi/linux/lustre/lustre_access_log.h>
  48
  49 #include "ofd_internal.h"
  50
  51 #ifdef CONFIG_PROC_FS
  52
  53 /**
  54  * Show number of FID allocation sequences.
  55  *
  56  * \param[in] m         seq_file handle
  57  * \param[in] data      unused for single entry
  58  *
  59  * \retval              0 on success
  60  * \retval              negative value on error
  61  */
  62 static ssize_t seqs_allocated_show(struct kobject *kobj, struct attribute *attr,
  63                                    char *buf)
  64 {
  65         struct obd_device *obd = container_of(kobj, struct obd_device,
  66                                               obd_kset.kobj);
  67         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
  68
  69         return sprintf(buf, "%u\n", ofd->ofd_seq_count);
  70 }
  71 LUSTRE_RO_ATTR(seqs_allocated);
  72
  73 /**
  74  * Show total number of grants for precreate.
  75  *
  76  * \param[in] m         seq_file handle
  77  * \param[in] data      unused for single entry
  78  *
  79  * \retval              0 on success
  80  * \retval              negative value on error
  81  */
  82 static ssize_t grant_precreate_show(struct kobject *kobj,
  83                                     struct attribute *attr,
  84                                     char *buf)
  85 {
  86         struct obd_device *obd = container_of(kobj, struct obd_device,
  87                                               obd_kset.kobj);
  88
  89         return sprintf(buf, "%ld\n",
  90                        obd->obd_self_export->exp_target_data.ted_grant);
  91 }
  92 LUSTRE_RO_ATTR(grant_precreate);
  93
  94 /**
  95  * Show number of precreates allowed in a single transaction.
  96  *
  97  * \param[in] m         seq_file handle
  98  * \param[in] data      unused for single entry
  99  *
 100  * \retval              0 on success
 101  * \retval              negative value on error
 102  */
 103 static ssize_t precreate_batch_show(struct kobject *kobj,
 104                                     struct attribute *attr,
 105                                     char *buf)
 106 {
 107         struct obd_device *obd = container_of(kobj, struct obd_device,
 108                                               obd_kset.kobj);
 109         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 110
 111         return sprintf(buf, "%d\n", ofd->ofd_precreate_batch);
 112 }
 113
 114 /**
 115  * Change number of precreates allowed in a single transaction.
 116  *
 117  * \param[in] file      proc file
 118  * \param[in] buffer    string which represents maximum number
 119  * \param[in] count     \a buffer length
 120  * \param[in] off       unused for single entry
 121  *
 122  * \retval              \a count on success
 123  * \retval              negative number on error
 124  */
 125 static ssize_t precreate_batch_store(struct kobject *kobj,
 126                                      struct attribute *attr,
 127                                      const char *buffer, size_t count)
 128 {
 129         struct obd_device *obd = container_of(kobj, struct obd_device,
 130                                               obd_kset.kobj);
 131         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 132         unsigned int val;
 133         int rc;
 134
 135         rc = kstrtouint(buffer, 0, &val);
 136         if (rc)
 137                 return rc;
 138
 139         if (val < 1 || val > 65536)
 140                 return -EINVAL;
 141
 142         spin_lock(&ofd->ofd_batch_lock);
 143         ofd->ofd_precreate_batch = val;
 144         spin_unlock(&ofd->ofd_batch_lock);
 145         return count;
 146 }
 147 LUSTRE_RW_ATTR(precreate_batch);
 148
 149 /**
 150  * Show number of seconds to delay atime
 151  *
 152  * \param[in] m         seq_file handle
 153  * \param[in] data      unused for single entry
 154  *
 155  * \retval              0 on success
 156  * \retval              negative value on error
 157  */
 158 static ssize_t atime_diff_show(struct kobject *kobj, struct attribute *attr,
 159                                char *buf)
 160 {
 161         struct obd_device *obd = container_of(kobj, struct obd_device,
 162                                               obd_kset.kobj);
 163         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 164
 165         return scnprintf(buf, PAGE_SIZE, "%lld\n", ofd->ofd_atime_diff);
 166 }
 167
 168 /**
 169  * Change number of seconds to delay atime
 170  *
 171  * \param[in] file      proc file
 172  * \param[in] buffer    string which represents maximum number
 173  * \param[in] count     \a buffer length
 174  * \param[in] off       unused for single entry
 175  *
 176  * \retval              \a count on success
 177  * \retval              negative number on error
 178  */
 179 static ssize_t atime_diff_store(struct kobject *kobj, struct attribute *attr,
 180                                 const char *buffer, size_t count)
 181 {
 182         struct obd_device *obd = container_of(kobj, struct obd_device,
 183                                               obd_kset.kobj);
 184         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 185         unsigned int val;
 186         int rc;
 187
 188         rc = kstrtouint(buffer, 0, &val);
 189         if (rc)
 190                 return rc;
 191
 192         if (val > 86400)
 193                 return -EINVAL;
 194
 195         ofd->ofd_atime_diff = val;
 196         return count;
 197 }
 198 LUSTRE_RW_ATTR(atime_diff);
 199
 200 /**
 201  * Show the last used ID for each FID sequence used by OFD.
 202  *
 203  * \param[in] m         seq_file handle
 204  * \param[in] data      unused for single entry
 205  *
 206  * \retval              0 on success
 207  * \retval              negative value on error
 208  */
 209 static int ofd_last_id_seq_show(struct seq_file *m, void *data)
 210 {
 211         struct obd_device       *obd = m->private;
 212         struct ofd_device       *ofd;
 213         struct ofd_seq          *oseq = NULL;
 214
 215         if (obd == NULL)
 216                 return 0;
 217
 218         ofd = ofd_dev(obd->obd_lu_dev);
 219
 220         read_lock(&ofd->ofd_seq_list_lock);
 221         list_for_each_entry(oseq, &ofd->ofd_seq_list, os_list) {
 222                 __u64 seq;
 223
 224                 seq = ostid_seq(&oseq->os_oi) == 0 ?
 225                       fid_idif_seq(ostid_id(&oseq->os_oi),
 226                                 ofd->ofd_lut.lut_lsd.lsd_osd_index) & ~0xFFFF :
 227                                 ostid_seq(&oseq->os_oi);
 228                 seq_printf(m, DOSTID"\n", seq, ostid_id(&oseq->os_oi));
 229         }
 230         read_unlock(&ofd->ofd_seq_list_lock);
 231         return 0;
 232 }
 233
 234 LPROC_SEQ_FOPS_RO(ofd_last_id);
 235
 236 /**
 237  * Show if the OFD is in degraded mode.
 238  *
 239  * Degraded means OFD has a failed drive or is undergoing RAID rebuild.
 240  * The MDS will try to avoid using this OST for new object allocations
 241  * to reduce the impact to global IO performance when clients writing to
 242  * this OST are slowed down.  It also reduces the contention on the OST
 243  * RAID device, allowing it to rebuild more quickly.
 244  *
 245  * \retval              count of bytes written
 246  */
 247 static ssize_t degraded_show(struct kobject *kobj, struct attribute *attr,
 248                              char *buf)
 249 {
 250         struct obd_device *obd = container_of(kobj, struct obd_device,
 251                                               obd_kset.kobj);
 252         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 253
 254         return sprintf(buf, "%u\n", ofd->ofd_raid_degraded);
 255 }
 256
 257 /**
 258  * Set OFD to degraded mode.
 259  *
 260  * This is used to interface to userspace administrative tools for
 261  * the underlying RAID storage, so that they can mark an OST
 262  * as having degraded performance.
 263  *
 264  * \param[in] count     \a buffer length
 265  * \param[in] off       unused for single entry
 266  *
 267  * \retval              \a count on success
 268  * \retval              negative number on error
 269  */
 270 static ssize_t degraded_store(struct kobject *kobj, struct attribute *attr,
 271                               const char *buffer, size_t count)
 272 {
 273         struct obd_device *obd = container_of(kobj, struct obd_device,
 274                                               obd_kset.kobj);
 275         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 276         bool val;
 277         int rc;
 278
 279         rc = kstrtobool(buffer, &val);
 280         if (rc)
 281                 return rc;
 282
 283         spin_lock(&ofd->ofd_flags_lock);
 284         ofd->ofd_raid_degraded = val;
 285         spin_unlock(&ofd->ofd_flags_lock);
 286         return count;
 287 }
 288 LUSTRE_RW_ATTR(degraded);
 289
 290 /**
 291  * Show if the OFD is in no precreate mode.
 292  *
 293  * This means OFD has been adminstratively disabled at the OST to prevent
 294  * the MDS from creating any new files on the OST, though existing files
 295  * can still be read, written, and unlinked.
 296  *
 297  * \retval              number of bytes written
 298  */
 299 static ssize_t no_precreate_show(struct kobject *kobj, struct attribute *attr,
 300                                  char *buf)
 301 {
 302         struct obd_device *obd = container_of(kobj, struct obd_device,
 303                                               obd_kset.kobj);
 304         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 305
 306         return scnprintf(buf, PAGE_SIZE, "%u\n", ofd->ofd_no_precreate);
 307 }
 308
 309 /**
 310  * Set OFD to no precreate mode.
 311  *
 312  * This is used to interface to userspace administrative tools to
 313  * disable new object creation on the OST.
 314  *
 315  * \param[in] count     \a buffer length
 316  *
 317  * \retval              \a count on success
 318  * \retval              negative number on error
 319  */
 320 static ssize_t no_precreate_store(struct kobject *kobj, struct attribute *attr,
 321                                   const char *buffer, size_t count)
 322 {
 323         struct obd_device *obd = container_of(kobj, struct obd_device,
 324                                               obd_kset.kobj);
 325         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 326         bool val;
 327         int rc;
 328
 329         rc = kstrtobool(buffer, &val);
 330         if (rc)
 331                 return rc;
 332
 333         spin_lock(&ofd->ofd_flags_lock);
 334         ofd->ofd_no_precreate = val;
 335         spin_unlock(&ofd->ofd_flags_lock);
 336
 337         return count;
 338 }
 339 LUSTRE_RW_ATTR(no_precreate);
 340
 341 /**
 342  * Show OFD filesystem type.
 343  *
 344  * \param[in] m         seq_file handle
 345  * \param[in] data      unused for single entry
 346  *
 347  * \retval              0 on success
 348  * \retval              negative value on error
 349  */
 350 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
 351                            char *buf)
 352 {
 353         struct obd_device *obd = container_of(kobj, struct obd_device,
 354                                               obd_kset.kobj);
 355         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 356         struct lu_device  *d;
 357
 358         LASSERT(ofd->ofd_osd);
 359         d = &ofd->ofd_osd->dd_lu_dev;
 360         LASSERT(d->ld_type);
 361         return sprintf(buf, "%s\n", d->ld_type->ldt_name);
 362 }
 363 LUSTRE_RO_ATTR(fstype);
 364
 365 /**
 366  * Show journal handling mode: synchronous or asynchronous.
 367  *
 368  * When running in asynchronous mode the journal transactions are not
 369  * committed to disk before the RPC is replied back to the client.
 370  * This will typically improve client performance when only a small number
 371  * of clients are writing, since the client(s) can have more write RPCs
 372  * in flight. However, it also means that the client has to handle recovery
 373  * on bulk RPCs, and will have to keep more dirty pages in cache before they
 374  * are committed on the OST.
 375  *
 376  * \param[in] m         seq_file handle
 377  * \param[in] data      unused for single entry
 378  *
 379  * \retval              0 on success
 380  * \retval              negative value on error
 381  */
 382 static ssize_t sync_journal_show(struct kobject *kobj, struct attribute *attr,
 383                                 char *buf)
 384 {
 385         struct obd_device *obd = container_of(kobj, struct obd_device,
 386                                               obd_kset.kobj);
 387         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 388
 389         return sprintf(buf, "%u\n", ofd->ofd_sync_journal);
 390 }
 391
 392 /**
 393  * Set journal mode to synchronous or asynchronous.
 394  *
 395  * \param[in] file      proc file
 396  * \param[in] buffer    string which represents mode
 397  *                      1: synchronous mode
 398  *                      0: asynchronous mode
 399  * \param[in] count     \a buffer length
 400  * \param[in] off       unused for single entry
 401  *
 402  * \retval              \a count on success
 403  * \retval              negative number on error
 404  */
 405 static ssize_t sync_journal_store(struct kobject *kobj, struct attribute *attr,
 406                                  const char *buffer, size_t count)
 407 {
 408         struct obd_device *obd = container_of(kobj, struct obd_device,
 409                                               obd_kset.kobj);
 410         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 411         bool val;
 412         int rc;
 413
 414         rc = kstrtobool(buffer, &val);
 415         if (rc)
 416                 return rc;
 417
 418         spin_lock(&ofd->ofd_flags_lock);
 419         ofd->ofd_sync_journal = val;
 420         ofd_slc_set(ofd);
 421         spin_unlock(&ofd->ofd_flags_lock);
 422
 423         return count;
 424 }
 425 LUSTRE_RW_ATTR(sync_journal);
 426
 427 static int ofd_brw_size_seq_show(struct seq_file *m, void *data)
 428 {
 429         struct obd_device       *obd = m->private;
 430         struct ofd_device       *ofd = ofd_dev(obd->obd_lu_dev);
 431
 432         seq_printf(m, "%u\n", ofd->ofd_brw_size / ONE_MB_BRW_SIZE);
 433         return 0;
 434 }
 435
 436 static ssize_t
 437 ofd_brw_size_seq_write(struct file *file, const char __user *buffer,
 438                        size_t count, loff_t *off)
 439 {
 440         struct seq_file *m = file->private_data;
 441         struct obd_device *obd = m->private;
 442         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 443         char kernbuf[22] = "";
 444         u64 val;
 445         int rc;
 446
 447         if (count >= sizeof(kernbuf))
 448                 return -EINVAL;
 449
 450         if (copy_from_user(kernbuf, buffer, count))
 451                 return -EFAULT;
 452         kernbuf[count] = 0;
 453
 454         rc = sysfs_memparse(kernbuf, count, &val, "MiB");
 455         if (rc < 0)
 456                 return rc;
 457
 458         if (val == 0)
 459                 return -EINVAL;
 460
 461         if (val > DT_MAX_BRW_SIZE ||
 462             val < (1 << ofd->ofd_lut.lut_tgd.tgd_blockbits))
 463                 return -ERANGE;
 464
 465         spin_lock(&ofd->ofd_flags_lock);
 466         ofd->ofd_brw_size = val;
 467         spin_unlock(&ofd->ofd_flags_lock);
 468
 469         return count;
 470 }
 471 LPROC_SEQ_FOPS(ofd_brw_size);
 472
 473 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 16, 53, 0)
 474 static bool sync_on_lock_cancel_warned;
 475 static ssize_t sync_on_lock_cancel_show(struct kobject *kobj,
 476                                         struct attribute *attr, char *buf)
 477 {
 478         if (!sync_on_lock_cancel_warned) {
 479                 sync_on_lock_cancel_warned = true;
 480                 pr_info("ofd: 'obdfilter.*.sync_on_lock_cancel' is deprecated, use 'obdfilter.*.sync_lock_cancel' instead\n");
 481         }
 482         return sync_lock_cancel_show(kobj, attr, buf);
 483 }
 484
 485 static ssize_t sync_on_lock_cancel_store(struct kobject *kobj,
 486                                          struct attribute *attr,
 487                                          const char *buffer, size_t count)
 488 {
 489         if (!sync_on_lock_cancel_warned) {
 490                 sync_on_lock_cancel_warned = true;
 491                 pr_info("ofd: 'obdfilter.*.sync_on_lock_cancel' is deprecated, use 'obdfilter.*.sync_lock_cancel' instead\n");
 492         }
 493         return sync_lock_cancel_store(kobj, attr, buffer, count);
 494 }
 495 LUSTRE_RW_ATTR(sync_on_lock_cancel);
 496 #endif
 497
 498 /**
 499  * Show the limit of soft sync RPCs.
 500  *
 501  * This value defines how many IO RPCs with OBD_BRW_SOFT_SYNC flag
 502  * are allowed before sync update will be triggered.
 503  *
 504  * \param[in] m         seq_file handle
 505  * \param[in] data      unused for single entry
 506  *
 507  * \retval              0 on success
 508  * \retval              negative value on error
 509  */
 510 static ssize_t soft_sync_limit_show(struct kobject *kobj,
 511                                     struct attribute *attr, char *buf)
 512 {
 513         struct obd_device *obd = container_of(kobj, struct obd_device,
 514                                               obd_kset.kobj);
 515         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 516
 517         return sprintf(buf, "%u\n", ofd->ofd_soft_sync_limit);
 518 }
 519
 520 /**
 521  * Change the limit of soft sync RPCs.
 522  *
 523  * Define how many IO RPCs with OBD_BRW_SOFT_SYNC flag
 524  * allowed before sync update will be done.
 525  *
 526  * This limit is global across all exports.
 527  *
 528  * \param[in] file      proc file
 529  * \param[in] buffer    string which represents limit
 530  * \param[in] count     \a buffer length
 531  * \param[in] off       unused for single entry
 532  *
 533  * \retval              \a count on success
 534  * \retval              negative number on error
 535  */
 536 static ssize_t soft_sync_limit_store(struct kobject *kobj,
 537                                      struct attribute *attr,
 538                                      const char *buffer, size_t count)
 539 {
 540         struct obd_device *obd = container_of(kobj, struct obd_device,
 541                                               obd_kset.kobj);
 542         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 543         unsigned int val;
 544         int rc;
 545
 546         rc = kstrtouint(buffer, 0, &val);
 547         if (rc < 0)
 548                 return rc;
 549
 550         ofd->ofd_soft_sync_limit = val;
 551         return 0;
 552 }
 553 LUSTRE_RW_ATTR(soft_sync_limit);
 554
 555 /**
 556  * Show the LFSCK speed limit.
 557  *
 558  * The maximum number of items scanned per second.
 559  *
 560  * \param[in] m         seq_file handle
 561  * \param[in] data      unused for single entry
 562  *
 563  * \retval              0 on success
 564  * \retval              negative value on error
 565  */
 566 static ssize_t lfsck_speed_limit_show(struct kobject *kobj,
 567                                       struct attribute *attr, char *buf)
 568 {
 569         struct obd_device *obd = container_of(kobj, struct obd_device,
 570                                               obd_kset.kobj);
 571         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 572
 573         return lfsck_get_speed(buf, ofd->ofd_osd);
 574 }
 575
 576 /**
 577  * Change the LFSCK speed limit.
 578  *
 579  * Limit number of items that may be scanned per second.
 580  *
 581  * \param[in] file      proc file
 582  * \param[in] buffer    string which represents limit
 583  * \param[in] count     \a buffer length
 584  * \param[in] off       unused for single entry
 585  *
 586  * \retval              \a count on success
 587  * \retval              negative number on error
 588  */
 589 static ssize_t lfsck_speed_limit_store(struct kobject *kobj,
 590                                        struct attribute *attr,
 591                                        const char *buffer, size_t count)
 592 {
 593         struct obd_device *obd = container_of(kobj, struct obd_device,
 594                                               obd_kset.kobj);
 595         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 596         unsigned int val;
 597         int rc;
 598
 599         rc = kstrtouint(buffer, 0, &val);
 600         if (rc != 0)
 601                 return rc;
 602
 603         rc = lfsck_set_speed(ofd->ofd_osd, val);
 604
 605         return rc != 0 ? rc : count;
 606 }
 607 LUSTRE_RW_ATTR(lfsck_speed_limit);
 608
 609 /**
 610  * Show LFSCK layout verification stats from the most recent LFSCK run.
 611  *
 612  * \param[in] m         seq_file handle
 613  * \param[in] data      unused for single entry
 614  *
 615  * \retval              0 on success
 616  * \retval              negative value on error
 617  */
 618 static int ofd_lfsck_layout_seq_show(struct seq_file *m, void *data)
 619 {
 620         struct obd_device *obd = m->private;
 621         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 622
 623         return lfsck_dump(m, ofd->ofd_osd, LFSCK_TYPE_LAYOUT);
 624 }
 625
 626 LPROC_SEQ_FOPS_RO(ofd_lfsck_layout);
 627
 628 /**
 629  * Show if LFSCK performed parent FID verification.
 630  *
 631  * \param[in] m         seq_file handle
 632  * \param[in] data      unused for single entry
 633  *
 634  * \retval              0 on success
 635  * \retval              negative value on error
 636  */
 637 static int ofd_lfsck_verify_pfid_seq_show(struct seq_file *m, void *data)
 638 {
 639         struct obd_device *obd = m->private;
 640         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 641
 642         seq_printf(m, "switch: %s\ndetected: %llu\nrepaired: %llu\n",
 643                    ofd->ofd_lfsck_verify_pfid ? "on" : "off",
 644                    ofd->ofd_inconsistency_self_detected,
 645                    ofd->ofd_inconsistency_self_repaired);
 646         return 0;
 647 }
 648
 649 /**
 650  * Set the LFSCK behavior to verify parent FID correctness.
 651  *
 652  * If flag ofd_lfsck_verify_pfid is set then LFSCK does parent FID
 653  * verification during read/write operations.
 654  *
 655  * \param[in] file      proc file
 656  * \param[in] buffer    string which represents behavior
 657  *                      1: verify parent FID
 658  *                      0: don't verify parent FID
 659  * \param[in] count     \a buffer length
 660  * \param[in] off       unused for single entry
 661  *
 662  * \retval              \a count on success
 663  * \retval              negative number on error
 664  */
 665 static ssize_t
 666 ofd_lfsck_verify_pfid_seq_write(struct file *file, const char __user *buffer,
 667                                 size_t count, loff_t *off)
 668 {
 669         struct seq_file *m = file->private_data;
 670         struct obd_device *obd = m->private;
 671         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 672         bool val;
 673         int rc;
 674
 675         rc = kstrtobool_from_user(buffer, count, &val);
 676         if (rc)
 677                 return rc;
 678
 679         ofd->ofd_lfsck_verify_pfid = val;
 680         if (!ofd->ofd_lfsck_verify_pfid) {
 681                 ofd->ofd_inconsistency_self_detected = 0;
 682                 ofd->ofd_inconsistency_self_repaired = 0;
 683         }
 684
 685         return count;
 686 }
 687
 688 LPROC_SEQ_FOPS(ofd_lfsck_verify_pfid);
 689
 690 static ssize_t access_log_mask_show(struct kobject *kobj,
 691                         struct attribute *attr, char *buf)
 692 {
 693         struct obd_device *obd = container_of(kobj, struct obd_device,
 694                                               obd_kset.kobj);
 695         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 696
 697         return scnprintf(buf, PAGE_SIZE, "%s%s%s\n",
 698                 (ofd->ofd_access_log_mask == 0) ? "0" : "",
 699                 (ofd->ofd_access_log_mask & OFD_ACCESS_READ) ? "r" : "",
 700                 (ofd->ofd_access_log_mask & OFD_ACCESS_WRITE) ? "w" : "");
 701 }
 702
 703 static ssize_t access_log_mask_store(struct kobject *kobj,
 704                                      struct attribute *attr,
 705                                      const char *buffer, size_t count)
 706 {
 707         struct obd_device *obd = container_of(kobj, struct obd_device,
 708                                               obd_kset.kobj);
 709         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 710         unsigned int mask = 0;
 711         size_t i;
 712
 713         for (i = 0; i < count; i++) {
 714                 switch (tolower(buffer[i])) {
 715                 case '0':
 716                         break;
 717                 case 'r':
 718                         mask |= OFD_ACCESS_READ;
 719                         break;
 720                 case 'w':
 721                         mask |= OFD_ACCESS_WRITE;
 722                         break;
 723                 default:
 724                         return -EINVAL;
 725                 }
 726         }
 727
 728         ofd->ofd_access_log_mask = mask;
 729
 730         return count;
 731 }
 732 LUSTRE_RW_ATTR(access_log_mask);
 733
 734 static ssize_t access_log_size_show(struct kobject *kobj,
 735                                     struct attribute *attr, char *buf)
 736 {
 737         struct obd_device *obd = container_of(kobj, struct obd_device,
 738                                               obd_kset.kobj);
 739         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 740
 741         return scnprintf(buf, PAGE_SIZE, "%u\n", ofd->ofd_access_log_size);
 742 }
 743
 744 static ssize_t access_log_size_store(struct kobject *kobj,
 745                                      struct attribute *attr,
 746                                      const char *buffer, size_t count)
 747 {
 748         struct obd_device *obd = container_of(kobj, struct obd_device,
 749                                               obd_kset.kobj);
 750         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 751         struct ofd_access_log *oal;
 752         unsigned int size;
 753         ssize_t rc;
 754
 755         rc = kstrtouint(buffer, 0, &size);
 756         if (rc < 0)
 757                 return rc;
 758
 759         if (!ofd_access_log_size_is_valid(size))
 760                 return -EINVAL;
 761
 762         /* The size of the ofd_access_log cannot be changed after it
 763          * has been created.
 764          */
 765         if (ofd->ofd_access_log_size == size)
 766                 return count;
 767
 768         oal = ofd_access_log_create(obd->obd_name, size);
 769         if (IS_ERR(oal))
 770                 return PTR_ERR(oal);
 771
 772         spin_lock(&ofd->ofd_flags_lock);
 773         if (ofd->ofd_access_log != NULL) {
 774                 rc = -EBUSY;
 775         } else {
 776                 ofd->ofd_access_log = oal;
 777                 ofd->ofd_access_log_size = size;
 778                 oal = NULL;
 779                 rc = count;
 780         }
 781         spin_unlock(&ofd->ofd_flags_lock);
 782
 783         ofd_access_log_delete(oal);
 784
 785         return rc;
 786 }
 787 LUSTRE_RW_ATTR(access_log_size);
 788
 789 static int ofd_site_stats_seq_show(struct seq_file *m, void *data)
 790 {
 791         struct obd_device *obd = m->private;
 792
 793         return lu_site_stats_seq_print(obd->obd_lu_dev->ld_site, m);
 794 }
 795
 796 LPROC_SEQ_FOPS_RO(ofd_site_stats);
 797
 798 /**
 799  * Show if the OFD enforces T10PI checksum.
 800  *
 801  * \param[in] m         seq_file handle
 802  * \param[in] data      unused for single entry
 803  *
 804  * \retval              0 on success
 805  * \retval              negative value on error
 806  */
 807 static ssize_t checksum_t10pi_enforce_show(struct kobject *kobj,
 808                                            struct attribute *attr,
 809                                            char *buf)
 810 {
 811         struct obd_device *obd = container_of(kobj, struct obd_device,
 812                                               obd_kset.kobj);
 813         struct lu_target *lut = obd->u.obt.obt_lut;
 814
 815         return scnprintf(buf, PAGE_SIZE, "%u\n", lut->lut_cksum_t10pi_enforce);
 816 }
 817
 818 /**
 819  * Force specific T10PI checksum modes to be enabled
 820  *
 821  * If T10PI *is* supported in hardware, allow only the supported T10PI type
 822  * to be used. If T10PI is *not* supported by the OSD, setting the enforce
 823  * parameter forces all T10PI types to be enabled (even if slower) for
 824  * testing.
 825  *
 826  * The final determination of which algorithm to be used depends whether
 827  * the client supports T10PI or not, and is handled at client connect time.
 828  *
 829  * \param[in] file      proc file
 830  * \param[in] buffer    string which represents mode
 831  *                      1: set T10PI checksums enforced
 832  *                      0: unset T10PI checksums enforced
 833  * \param[in] count     \a buffer length
 834  * \param[in] off       unused for single entry
 835  *
 836  * \retval              \a count on success
 837  * \retval              negative number on error
 838  */
 839 static ssize_t checksum_t10pi_enforce_store(struct kobject *kobj,
 840                                             struct attribute *attr,
 841                                             const char *buffer, size_t count)
 842 {
 843         struct obd_device *obd = container_of(kobj, struct obd_device,
 844                                               obd_kset.kobj);
 845         struct lu_target *lut = obd->u.obt.obt_lut;
 846         bool enforce;
 847         int rc;
 848
 849         rc = kstrtobool(buffer, &enforce);
 850         if (rc)
 851                 return rc;
 852
 853         spin_lock(&lut->lut_flags_lock);
 854         lut->lut_cksum_t10pi_enforce = enforce;
 855         spin_unlock(&lut->lut_flags_lock);
 856         return count;
 857 }
 858 LUSTRE_RW_ATTR(checksum_t10pi_enforce);
 859
 860 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
 861 static bool max_file_warned;
 862 static bool rd_cache_warned;
 863 static bool wr_cache_warned;
 864
 865 static ssize_t read_cache_enable_show(struct kobject *kobj,
 866                                       struct attribute *attr,
 867                                       char *buf)
 868 {
 869         struct obd_device *obd = container_of(kobj, struct obd_device,
 870                                               obd_kset.kobj);
 871         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 872
 873         if (!rd_cache_warned) {
 874                 rd_cache_warned = true;
 875                 pr_info("ofd: 'obdfilter.*.read_cache_enabled' is deprecated, use 'osd-*.read_cache_enabled' instead\n");
 876         }
 877
 878         if (!ofd->ofd_read_cache_enable)
 879                 return -EOPNOTSUPP;
 880
 881         return lustre_attr_show(&ofd->ofd_osd->dd_kobj,
 882                                 ofd->ofd_read_cache_enable, buf);
 883 }
 884
 885 static ssize_t read_cache_enable_store(struct kobject *kobj,
 886                                        struct attribute *attr,
 887                                        const char *buffer, size_t count)
 888 {
 889         struct obd_device *obd = container_of(kobj, struct obd_device,
 890                                               obd_kset.kobj);
 891         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 892
 893         if (!rd_cache_warned) {
 894                 rd_cache_warned = true;
 895                 pr_info("ofd: 'obdfilter.*.read_cache_enabled' is deprecated, use 'osd-*.read_cache_enabled' instead\n");
 896         }
 897
 898         if (!ofd->ofd_read_cache_enable)
 899                 return -EOPNOTSUPP;
 900
 901         return lustre_attr_store(&ofd->ofd_osd->dd_kobj,
 902                                  ofd->ofd_read_cache_enable, buffer, count);
 903 }
 904 LUSTRE_RW_ATTR(read_cache_enable);
 905
 906 static ssize_t readcache_max_filesize_show(struct kobject *kobj,
 907                                            struct attribute *attr,
 908                                            char *buf)
 909 {
 910         struct obd_device *obd = container_of(kobj, struct obd_device,
 911                                               obd_kset.kobj);
 912         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 913
 914         if (!max_file_warned) {
 915                 max_file_warned = true;
 916                 pr_info("ofd: 'obdfilter.*.readcache_max_filesize' is deprecated, use 'osd-*.readcache_max_filesize' instead\n");
 917         }
 918
 919         if (!ofd->ofd_read_cache_max_filesize)
 920                 return -EOPNOTSUPP;
 921
 922         return lustre_attr_show(&ofd->ofd_osd->dd_kobj,
 923                                 ofd->ofd_read_cache_max_filesize, buf);
 924 }
 925
 926 static ssize_t readcache_max_filesize_store(struct kobject *kobj,
 927                                             struct attribute *attr,
 928                                             const char *buffer, size_t count)
 929 {
 930         struct obd_device *obd = container_of(kobj, struct obd_device,
 931                                               obd_kset.kobj);
 932         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 933
 934         if (!max_file_warned) {
 935                 max_file_warned = true;
 936                 pr_info("ofd: 'obdfilter.*.readcache_max_filesize' is deprecated, use 'osd-*.readcache_max_filesize' instead\n");
 937         }
 938
 939         if (!ofd->ofd_read_cache_max_filesize)
 940                 return -EOPNOTSUPP;
 941
 942         return lustre_attr_store(&ofd->ofd_osd->dd_kobj,
 943                                  ofd->ofd_read_cache_max_filesize,
 944                                  buffer, count);
 945 }
 946 LUSTRE_RW_ATTR(readcache_max_filesize);
 947
 948 static ssize_t writethrough_cache_enable_show(struct kobject *kobj,
 949                                               struct attribute *attr,
 950                                               char *buf)
 951 {
 952         struct obd_device *obd = container_of(kobj, struct obd_device,
 953                                               obd_kset.kobj);
 954         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 955
 956         if (!wr_cache_warned) {
 957                 wr_cache_warned = true;
 958                 pr_info("ofd: 'obdfilter.*.writethrough_cache_enabled' is deprecated, use 'osd-*.writethrough_cache_enabled' instead\n");
 959         }
 960
 961         if (!ofd->ofd_write_cache_enable)
 962                 return -EOPNOTSUPP;
 963
 964         return lustre_attr_show(&ofd->ofd_osd->dd_kobj,
 965                                 ofd->ofd_write_cache_enable, buf);
 966 }
 967
 968 static ssize_t writethrough_cache_enable_store(struct kobject *kobj,
 969                                                struct attribute *attr,
 970                                                const char *buffer, size_t count)
 971 {
 972         struct obd_device *obd = container_of(kobj, struct obd_device,
 973                                               obd_kset.kobj);
 974         struct ofd_device *ofd = ofd_dev(obd->obd_lu_dev);
 975
 976         if (!ofd->ofd_write_cache_enable)
 977                 return -EOPNOTSUPP;
 978
 979         return lustre_attr_store(&ofd->ofd_osd->dd_kobj,
 980                                  ofd->ofd_write_cache_enable,
 981                                  buffer, count);
 982 }
 983 LUSTRE_RW_ATTR(writethrough_cache_enable);
 984 #endif
 985
 986 LPROC_SEQ_FOPS_RO_TYPE(ofd, recovery_status);
 987 LUSTRE_RW_ATTR(recovery_time_hard);
 988 LUSTRE_RW_ATTR(recovery_time_soft);
 989 LUSTRE_RW_ATTR(ir_factor);
 990
 991 LPROC_SEQ_FOPS_WR_ONLY(ofd, evict_client);
 992 LPROC_SEQ_FOPS_RW_TYPE(ofd, checksum_dump);
 993 LUSTRE_RW_ATTR(job_cleanup_interval);
 994
 995 LUSTRE_RO_ATTR(tot_dirty);
 996 LUSTRE_RO_ATTR(tot_granted);
 997 LUSTRE_RO_ATTR(tot_pending);
 998 LUSTRE_RW_ATTR(grant_compat_disable);
 999 LUSTRE_RO_ATTR(instance);
1000
1001 LUSTRE_RO_ATTR(num_exports);
1002
1003 struct lprocfs_vars lprocfs_ofd_obd_vars[] = {
1004         { .name =       "last_id",
1005           .fops =       &ofd_last_id_fops               },
1006         { .name =       "recovery_status",
1007           .fops =       &ofd_recovery_status_fops       },
1008         { .name =       "evict_client",
1009           .fops =       &ofd_evict_client_fops          },
1010         { .name =       "brw_size",
1011           .fops =       &ofd_brw_size_fops              },
1012         { .name =       "checksum_dump",
1013           .fops =       &ofd_checksum_dump_fops         },
1014         { .name =       "lfsck_layout",
1015           .fops =       &ofd_lfsck_layout_fops          },
1016         { .name =       "lfsck_verify_pfid",
1017           .fops =       &ofd_lfsck_verify_pfid_fops     },
1018         { .name =       "site_stats",
1019           .fops =       &ofd_site_stats_fops            },
1020         { NULL }
1021 };
1022
1023 /**
1024  * Initialize OFD statistics counters
1025  *
1026  * param[in] stats      statistics counters
1027  */
1028 void ofd_stats_counter_init(struct lprocfs_stats *stats, unsigned int offset)
1029 {
1030         LASSERT(stats && stats->ls_num >= LPROC_OFD_STATS_LAST);
1031
1032         lprocfs_counter_init(stats, LPROC_OFD_STATS_READ_BYTES,
1033                              LPROCFS_TYPE_BYTES_FULL, "read_bytes", "bytes");
1034         lprocfs_counter_init(stats, LPROC_OFD_STATS_WRITE_BYTES,
1035                              LPROCFS_TYPE_BYTES_FULL, "write_bytes", "bytes");
1036         lprocfs_counter_init(stats, LPROC_OFD_STATS_READ,
1037                              LPROCFS_TYPE_LATENCY, "read", "usecs");
1038         lprocfs_counter_init(stats, LPROC_OFD_STATS_WRITE,
1039                              LPROCFS_TYPE_LATENCY, "write", "usecs");
1040         lprocfs_counter_init(stats, LPROC_OFD_STATS_GETATTR,
1041                              LPROCFS_TYPE_LATENCY, "getattr", "usecs");
1042         lprocfs_counter_init(stats, LPROC_OFD_STATS_SETATTR,
1043                              LPROCFS_TYPE_LATENCY, "setattr", "usecs");
1044         lprocfs_counter_init(stats, LPROC_OFD_STATS_PUNCH,
1045                              LPROCFS_TYPE_LATENCY, "punch", "usecs");
1046         lprocfs_counter_init(stats, LPROC_OFD_STATS_SYNC,
1047                              LPROCFS_TYPE_LATENCY, "sync", "usecs");
1048         lprocfs_counter_init(stats, LPROC_OFD_STATS_DESTROY,
1049                              LPROCFS_TYPE_LATENCY, "destroy", "usecs");
1050         lprocfs_counter_init(stats, LPROC_OFD_STATS_CREATE,
1051                              LPROCFS_TYPE_LATENCY, "create", "usecs");
1052         lprocfs_counter_init(stats, LPROC_OFD_STATS_STATFS,
1053                              LPROCFS_TYPE_LATENCY, "statfs", "usecs");
1054         lprocfs_counter_init(stats, LPROC_OFD_STATS_GET_INFO,
1055                              LPROCFS_TYPE_LATENCY, "get_info", "usecs");
1056         lprocfs_counter_init(stats, LPROC_OFD_STATS_SET_INFO,
1057                              LPROCFS_TYPE_LATENCY, "set_info", "usecs");
1058         lprocfs_counter_init(stats, LPROC_OFD_STATS_QUOTACTL,
1059                              LPROCFS_TYPE_LATENCY, "quotactl", "usecs");
1060         lprocfs_counter_init(stats, LPROC_OFD_STATS_PREALLOC,
1061                              LPROCFS_TYPE_LATENCY, "prealloc", "usecs");
1062 }
1063
1064 LPROC_SEQ_FOPS(lprocfs_nid_stats_clear);
1065
1066 static struct attribute *ofd_attrs[] = {
1067         &lustre_attr_tot_dirty.attr,
1068         &lustre_attr_tot_granted.attr,
1069         &lustre_attr_tot_pending.attr,
1070         &lustre_attr_grant_compat_disable.attr,
1071         &lustre_attr_instance.attr,
1072         &lustre_attr_recovery_time_hard.attr,
1073         &lustre_attr_recovery_time_soft.attr,
1074         &lustre_attr_ir_factor.attr,
1075         &lustre_attr_num_exports.attr,
1076         &lustre_attr_seqs_allocated.attr,
1077         &lustre_attr_grant_precreate.attr,
1078         &lustre_attr_precreate_batch.attr,
1079         &lustre_attr_atime_diff.attr,
1080         &lustre_attr_degraded.attr,
1081         &lustre_attr_fstype.attr,
1082         &lustre_attr_no_precreate.attr,
1083         &lustre_attr_sync_journal.attr,
1084 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 16, 53, 0)
1085         &lustre_attr_sync_on_lock_cancel.attr,
1086 #endif
1087         &lustre_attr_soft_sync_limit.attr,
1088         &lustre_attr_lfsck_speed_limit.attr,
1089         &lustre_attr_access_log_mask.attr,
1090         &lustre_attr_access_log_size.attr,
1091         &lustre_attr_job_cleanup_interval.attr,
1092         &lustre_attr_checksum_t10pi_enforce.attr,
1093 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 53, 0)
1094         &lustre_attr_read_cache_enable.attr,
1095         &lustre_attr_readcache_max_filesize.attr,
1096         &lustre_attr_writethrough_cache_enable.attr,
1097 #endif
1098         NULL,
1099 };
1100
1101 /**
1102  * Initialize all needed procfs entries for OFD device.
1103  *
1104  * \param[in] ofd       OFD device
1105  *
1106  * \retval              0 if successful
1107  * \retval              negative value on error
1108  */
1109 int ofd_tunables_init(struct ofd_device *ofd)
1110 {
1111         struct obd_device *obd = ofd_obd(ofd);
1112         struct proc_dir_entry *entry;
1113         int rc = 0;
1114
1115         ENTRY;
1116         /* lprocfs must be setup before the ofd so state can be safely added
1117          * to /proc incrementally as the ofd is setup
1118          */
1119         obd->obd_ktype.default_attrs = ofd_attrs;
1120         obd->obd_vars = lprocfs_ofd_obd_vars;
1121         rc = lprocfs_obd_setup(obd, false);
1122         if (rc) {
1123                 CERROR("%s: lprocfs_obd_setup failed: %d.\n",
1124                        obd->obd_name, rc);
1125                 RETURN(rc);
1126         }
1127
1128         rc = tgt_tunables_init(&ofd->ofd_lut);
1129         if (rc) {
1130                 CERROR("%s: tgt_tunables_init failed: rc = %d\n",
1131                        obd->obd_name, rc);
1132                 GOTO(obd_cleanup, rc);
1133         }
1134
1135         rc = lprocfs_alloc_obd_stats(obd, LPROC_OFD_STATS_LAST);
1136         if (rc) {
1137                 CERROR("%s: lprocfs_alloc_obd_stats failed: %d.\n",
1138                        obd->obd_name, rc);
1139                 GOTO(tgt_cleanup, rc);
1140         }
1141
1142         entry = lprocfs_register("exports", obd->obd_proc_entry, NULL, NULL);
1143         if (IS_ERR(entry)) {
1144                 rc = PTR_ERR(entry);
1145                 CERROR("%s: error %d setting up lprocfs for %s\n",
1146                        obd->obd_name, rc, "exports");
1147                 GOTO(obd_free_stats, rc);
1148         }
1149         obd->obd_proc_exports_entry = entry;
1150
1151         entry = lprocfs_add_simple(obd->obd_proc_exports_entry, "clear",
1152                                    obd, &lprocfs_nid_stats_clear_fops);
1153         if (IS_ERR(entry)) {
1154                 rc = PTR_ERR(entry);
1155                 CERROR("%s: add proc entry 'clear' failed: %d.\n",
1156                        obd->obd_name, rc);
1157                 GOTO(obd_free_stats, rc);
1158         }
1159
1160         ofd_stats_counter_init(obd->obd_stats, 0);
1161
1162         rc = lprocfs_job_stats_init(obd, LPROC_OFD_STATS_LAST,
1163                                     ofd_stats_counter_init);
1164         if (rc)
1165                 GOTO(obd_free_stats, rc);
1166
1167         RETURN(0);
1168
1169 obd_free_stats:
1170         lprocfs_free_obd_stats(obd);
1171 tgt_cleanup:
1172         tgt_tunables_fini(&ofd->ofd_lut);
1173 obd_cleanup:
1174         lprocfs_obd_cleanup(obd);
1175
1176         return rc;
1177 }
1178 #endif /* CONFIG_PROC_FS */