lustre/llite/vvp_dev.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) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * cl_device and cl_device_type implementation for VVP layer.
  32  *
  33  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  34  *   Author: Jinshan Xiong <jinshan.xiong@intel.com>
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <obd.h>
  40 #include "llite_internal.h"
  41 #include "vvp_internal.h"
  42 #include <linux/kallsyms.h>
  43
  44 /*****************************************************************************
  45  *
  46  * Vvp device and device type functions.
  47  *
  48  */
  49
  50 /*
  51  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  52  * "llite_" (var. "ll_") prefix.
  53  */
  54
  55 static struct kmem_cache *ll_thread_kmem;
  56 struct kmem_cache *vvp_object_kmem;
  57 static struct kmem_cache *vvp_session_kmem;
  58 static struct kmem_cache *vvp_thread_kmem;
  59
  60 static struct lu_kmem_descr vvp_caches[] = {
  61         {
  62                 .ckd_cache = &ll_thread_kmem,
  63                 .ckd_name  = "ll_thread_kmem",
  64                 .ckd_size  = sizeof(struct ll_thread_info),
  65         },
  66         {
  67                 .ckd_cache = &vvp_object_kmem,
  68                 .ckd_name  = "vvp_object_kmem",
  69                 .ckd_size  = sizeof(struct vvp_object),
  70         },
  71         {
  72                 .ckd_cache = &vvp_session_kmem,
  73                 .ckd_name  = "vvp_session_kmem",
  74                 .ckd_size  = sizeof (struct vvp_session)
  75         },
  76         {
  77                 .ckd_cache = &vvp_thread_kmem,
  78                 .ckd_name  = "vvp_thread_kmem",
  79                 .ckd_size  = sizeof(struct vvp_thread_info),
  80         },
  81         {
  82                 .ckd_cache = NULL
  83         }
  84 };
  85
  86 static void *ll_thread_key_init(const struct lu_context *ctx,
  87                                 struct lu_context_key *key)
  88 {
  89         struct ll_thread_info *lti;
  90
  91         OBD_SLAB_ALLOC_PTR_GFP(lti, ll_thread_kmem, GFP_NOFS);
  92         if (lti == NULL)
  93                 lti = ERR_PTR(-ENOMEM);
  94
  95         return lti;
  96 }
  97
  98 static void ll_thread_key_fini(const struct lu_context *ctx,
  99                                struct lu_context_key *key, void *data)
 100 {
 101         struct ll_thread_info *lti = data;
 102
 103         OBD_SLAB_FREE_PTR(lti, ll_thread_kmem);
 104 }
 105
 106 struct lu_context_key ll_thread_key = {
 107         .lct_tags = LCT_CL_THREAD,
 108         .lct_init = ll_thread_key_init,
 109         .lct_fini = ll_thread_key_fini,
 110 };
 111
 112 static void *vvp_session_key_init(const struct lu_context *ctx,
 113                                   struct lu_context_key *key)
 114 {
 115         struct vvp_session *session;
 116
 117         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
 118         if (session == NULL)
 119                 session = ERR_PTR(-ENOMEM);
 120         return session;
 121 }
 122
 123 static void vvp_session_key_fini(const struct lu_context *ctx,
 124                                  struct lu_context_key *key, void *data)
 125 {
 126         struct vvp_session *session = data;
 127         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 128 }
 129
 130 struct lu_context_key vvp_session_key = {
 131         .lct_tags = LCT_SESSION,
 132         .lct_init = vvp_session_key_init,
 133         .lct_fini = vvp_session_key_fini
 134 };
 135
 136 static void *vvp_thread_key_init(const struct lu_context *ctx,
 137                                  struct lu_context_key *key)
 138 {
 139         struct vvp_thread_info *vti;
 140
 141         OBD_SLAB_ALLOC_PTR_GFP(vti, vvp_thread_kmem, GFP_NOFS);
 142         if (vti == NULL)
 143                 vti = ERR_PTR(-ENOMEM);
 144         return vti;
 145 }
 146
 147 static void vvp_thread_key_fini(const struct lu_context *ctx,
 148                                 struct lu_context_key *key, void *data)
 149 {
 150         struct vvp_thread_info *vti = data;
 151         OBD_SLAB_FREE_PTR(vti, vvp_thread_kmem);
 152 }
 153
 154 struct lu_context_key vvp_thread_key = {
 155         .lct_tags = LCT_CL_THREAD,
 156         .lct_init = vvp_thread_key_init,
 157         .lct_fini = vvp_thread_key_fini,
 158 };
 159
 160 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 161 LU_TYPE_INIT_FINI(vvp, &ll_thread_key, &vvp_session_key, &vvp_thread_key);
 162
 163 static const struct lu_device_operations vvp_lu_ops = {
 164         .ldo_object_alloc      = vvp_object_alloc
 165 };
 166
 167 static struct lu_device *vvp_device_free(const struct lu_env *env,
 168                                          struct lu_device *d)
 169 {
 170         struct vvp_device *vdv  = lu2vvp_dev(d);
 171         struct cl_site    *site = lu2cl_site(d->ld_site);
 172         struct lu_device  *next = cl2lu_dev(vdv->vdv_next);
 173
 174         if (d->ld_site != NULL) {
 175                 cl_site_fini(site);
 176                 OBD_FREE_PTR(site);
 177         }
 178
 179         cl_device_fini(lu2cl_dev(d));
 180         OBD_FREE_PTR(vdv);
 181         return next;
 182 }
 183
 184 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 185                                           struct lu_device_type *t,
 186                                           struct lustre_cfg *cfg)
 187 {
 188         struct vvp_device *vdv;
 189         struct lu_device *lud;
 190         struct cl_site *site;
 191         int rc;
 192         ENTRY;
 193
 194         OBD_ALLOC_PTR(vdv);
 195         if (vdv == NULL)
 196                 RETURN(ERR_PTR(-ENOMEM));
 197
 198         lud = &vdv->vdv_cl.cd_lu_dev;
 199         cl_device_init(&vdv->vdv_cl, t);
 200         vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
 201
 202         OBD_ALLOC_PTR(site);
 203         if (site != NULL) {
 204                 rc = cl_site_init(site, &vdv->vdv_cl);
 205                 if (rc == 0)
 206                         rc = lu_site_init_finish(&site->cs_lu);
 207                 else {
 208                         LASSERT(lud->ld_site == NULL);
 209                         CERROR("Cannot init lu_site, rc %d.\n", rc);
 210                         OBD_FREE_PTR(site);
 211                 }
 212         } else
 213                 rc = -ENOMEM;
 214         if (rc != 0) {
 215                 vvp_device_free(env, lud);
 216                 lud = ERR_PTR(rc);
 217         }
 218         RETURN(lud);
 219 }
 220
 221 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
 222                            const char *name, struct lu_device *next)
 223 {
 224         struct vvp_device  *vdv;
 225         int rc;
 226         ENTRY;
 227
 228         vdv = lu2vvp_dev(d);
 229         vdv->vdv_next = lu2cl_dev(next);
 230
 231         LASSERT(d->ld_site != NULL && next->ld_type != NULL);
 232         next->ld_site = d->ld_site;
 233         rc = next->ld_type->ldt_ops->ldto_device_init(
 234                 env, next, next->ld_type->ldt_name, NULL);
 235         if (rc == 0) {
 236                 lu_device_get(next);
 237                 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
 238         }
 239         RETURN(rc);
 240 }
 241
 242 static struct lu_device *vvp_device_fini(const struct lu_env *env,
 243                                          struct lu_device *d)
 244 {
 245         return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
 246 }
 247
 248 static const struct lu_device_type_operations vvp_device_type_ops = {
 249         .ldto_init = vvp_type_init,
 250         .ldto_fini = vvp_type_fini,
 251
 252         .ldto_start = vvp_type_start,
 253         .ldto_stop  = vvp_type_stop,
 254
 255         .ldto_device_alloc      = vvp_device_alloc,
 256         .ldto_device_free       = vvp_device_free,
 257         .ldto_device_init       = vvp_device_init,
 258         .ldto_device_fini       = vvp_device_fini,
 259 };
 260
 261 struct lu_device_type vvp_device_type = {
 262         .ldt_tags     = LU_DEVICE_CL,
 263         .ldt_name     = LUSTRE_VVP_NAME,
 264         .ldt_ops      = &vvp_device_type_ops,
 265         .ldt_ctx_tags = LCT_CL_THREAD
 266 };
 267
 268 #ifndef HAVE_ACCOUNT_PAGE_DIRTIED_EXPORT
 269 unsigned int (*vvp_account_page_dirtied)(struct page *page,
 270                                          struct address_space *mapping);
 271
 272 unsigned int ll_account_page_dirtied(struct page *page,
 273                                      struct address_space *mapping)
 274 {
 275         /* must use __set_page_dirty, which means unlocking and
 276          * relocking, which hurts performance.
 277          */
 278         ll_xa_unlock(&mapping->i_pages);
 279         __set_page_dirty(page, mapping, 0);
 280         ll_xa_lock(&mapping->i_pages);
 281         return 0;
 282 }
 283 #endif
 284
 285 /**
 286  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 287  * pressure, when environments cannot be allocated.
 288  */
 289 int vvp_global_init(void)
 290 {
 291         int rc;
 292
 293         rc = lu_kmem_init(vvp_caches);
 294         if (rc != 0)
 295                 return rc;
 296
 297         rc = lu_device_type_init(&vvp_device_type);
 298         if (rc != 0)
 299                 goto out_kmem;
 300
 301 #ifndef HAVE_ACCOUNT_PAGE_DIRTIED_EXPORT
 302         /*
 303          * Kernel v5.2-5678-gac1c3e4 no longer exports account_page_dirtied
 304          */
 305         vvp_account_page_dirtied = (void *)
 306                 cfs_kallsyms_lookup_name("account_page_dirtied");
 307         if (!vvp_account_page_dirtied)
 308                 vvp_account_page_dirtied = ll_account_page_dirtied;
 309 #endif
 310
 311         return 0;
 312
 313 out_kmem:
 314         lu_kmem_fini(vvp_caches);
 315
 316         return rc;
 317 }
 318
 319 void vvp_global_fini(void)
 320 {
 321         lu_device_type_fini(&vvp_device_type);
 322         lu_kmem_fini(vvp_caches);
 323 }
 324
 325 /*****************************************************************************
 326  *
 327  * mirror obd-devices into cl devices.
 328  *
 329  */
 330
 331 int cl_sb_init(struct super_block *sb)
 332 {
 333         struct ll_sb_info *sbi;
 334         struct cl_device  *cl;
 335         struct lu_env     *env;
 336         int rc = 0;
 337         __u16 refcheck;
 338
 339         sbi  = ll_s2sbi(sb);
 340         env = cl_env_get(&refcheck);
 341         if (!IS_ERR(env)) {
 342                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 343                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 344                 if (!IS_ERR(cl)) {
 345                         sbi->ll_cl = cl;
 346                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 347                 }
 348                 cl_env_put(env, &refcheck);
 349         } else
 350                 rc = PTR_ERR(env);
 351         RETURN(rc);
 352 }
 353
 354 int cl_sb_fini(struct super_block *sb)
 355 {
 356         struct ll_sb_info *sbi;
 357         struct lu_env     *env;
 358         struct cl_device  *cld;
 359         __u16              refcheck;
 360         int                result;
 361
 362         ENTRY;
 363         sbi = ll_s2sbi(sb);
 364         env = cl_env_get(&refcheck);
 365         if (!IS_ERR(env)) {
 366                 cld = sbi->ll_cl;
 367
 368                 if (cld != NULL) {
 369                         cl_stack_fini(env, cld);
 370                         sbi->ll_cl = NULL;
 371                         sbi->ll_site = NULL;
 372                 }
 373                 cl_env_put(env, &refcheck);
 374                 result = 0;
 375         } else {
 376                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 377                 result = PTR_ERR(env);
 378         }
 379
 380         RETURN(result);
 381 }
 382
 383 /****************************************************************************
 384  *
 385  * debugfs/lustre/llite/$MNT/dump_page_cache
 386  *
 387  ****************************************************************************/
 388
 389 struct vvp_seq_private {
 390         struct ll_sb_info       *vsp_sbi;
 391         struct lu_env           *vsp_env;
 392         u16                     vsp_refcheck;
 393         struct cl_object        *vsp_clob;
 394         struct rhashtable_iter  vsp_iter;
 395         u32                     vsp_page_index;
 396         /*
 397          * prev_pos is the 'pos' of the last object returned
 398          * by ->start of ->next.
 399          */
 400         loff_t                  vvp_prev_pos;
 401 };
 402
 403 static struct page *vvp_pgcache_current(struct vvp_seq_private *priv)
 404 {
 405         struct lu_device *dev = &priv->vsp_sbi->ll_cl->cd_lu_dev;
 406         struct lu_object_header *h;
 407         struct page *vmpage = NULL;
 408
 409         rhashtable_walk_start(&priv->vsp_iter);
 410         while ((h = rhashtable_walk_next(&priv->vsp_iter)) != NULL) {
 411                 struct inode *inode;
 412                 int nr;
 413
 414                 if (IS_ERR(h)) {
 415                         if (PTR_ERR(h) == -EAGAIN)
 416                                 continue;
 417                         break;
 418                 }
 419
 420                 if (!priv->vsp_clob) {
 421                         struct lu_object *lu_obj;
 422
 423                         lu_obj = lu_object_get_first(h, dev);
 424                         if (!lu_obj)
 425                                 continue;
 426
 427                         priv->vsp_clob = lu2cl(lu_obj);
 428                         lu_object_ref_add_atomic(lu_obj, "dump", current);
 429                         priv->vsp_page_index = 0;
 430                 }
 431
 432                 inode = vvp_object_inode(priv->vsp_clob);
 433                 nr = find_get_pages_contig(inode->i_mapping,
 434                                            priv->vsp_page_index, 1, &vmpage);
 435                 if (nr > 0) {
 436                         priv->vsp_page_index = vmpage->index;
 437                         break;
 438                 }
 439                 lu_object_ref_del(&priv->vsp_clob->co_lu, "dump", current);
 440                 cl_object_put(priv->vsp_env, priv->vsp_clob);
 441                 priv->vsp_clob = NULL;
 442                 priv->vsp_page_index = 0;
 443         }
 444         rhashtable_walk_stop(&priv->vsp_iter);
 445         return vmpage;
 446 }
 447
 448 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 449         if (test_bit(PG_##flag, &(page)->flags)) {                  \
 450                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 451                 has_flags = 1;                                          \
 452         }                                                               \
 453 } while(0)
 454
 455 static void vvp_pgcache_page_show(const struct lu_env *env,
 456                                   struct seq_file *seq, struct cl_page *page)
 457 {
 458         struct vvp_page *vpg;
 459         struct page      *vmpage;
 460         int              has_flags;
 461
 462         vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
 463         vmpage = vpg->vpg_page;
 464         seq_printf(seq, " %5i | %p %p %s %s %s | %p "DFID"(%p) %lu %u [",
 465                    0 /* gen */,
 466                    vpg, page,
 467                    "none",
 468                    vpg->vpg_defer_uptodate ? "du" : "- ",
 469                    PageWriteback(vmpage) ? "wb" : "-",
 470                    vmpage,
 471                    PFID(ll_inode2fid(vmpage->mapping->host)),
 472                    vmpage->mapping->host, vmpage->index,
 473                    page_count(vmpage));
 474         has_flags = 0;
 475         seq_page_flag(seq, vmpage, locked, has_flags);
 476         seq_page_flag(seq, vmpage, error, has_flags);
 477         seq_page_flag(seq, vmpage, referenced, has_flags);
 478         seq_page_flag(seq, vmpage, uptodate, has_flags);
 479         seq_page_flag(seq, vmpage, dirty, has_flags);
 480         seq_page_flag(seq, vmpage, writeback, has_flags);
 481         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 482 }
 483
 484 static int vvp_pgcache_show(struct seq_file *f, void *v)
 485 {
 486         struct vvp_seq_private *priv = f->private;
 487         struct page *vmpage = v;
 488         struct cl_page *page;
 489
 490         seq_printf(f, "%8lx@" DFID ": ", vmpage->index,
 491                    PFID(lu_object_fid(&priv->vsp_clob->co_lu)));
 492         lock_page(vmpage);
 493         page = cl_vmpage_page(vmpage, priv->vsp_clob);
 494         unlock_page(vmpage);
 495         put_page(vmpage);
 496
 497         if (page) {
 498                 vvp_pgcache_page_show(priv->vsp_env, f, page);
 499                 cl_page_put(priv->vsp_env, page);
 500         } else {
 501                 seq_puts(f, "missing\n");
 502         }
 503
 504         return 0;
 505 }
 506
 507 static void vvp_pgcache_rewind(struct vvp_seq_private *priv)
 508 {
 509         if (priv->vvp_prev_pos) {
 510                 struct lu_site *s = priv->vsp_sbi->ll_cl->cd_lu_dev.ld_site;
 511
 512                 rhashtable_walk_exit(&priv->vsp_iter);
 513                 rhashtable_walk_enter(&s->ls_obj_hash, &priv->vsp_iter);
 514                 priv->vvp_prev_pos = 0;
 515                 if (priv->vsp_clob) {
 516                         lu_object_ref_del(&priv->vsp_clob->co_lu, "dump",
 517                                           current);
 518                         cl_object_put(priv->vsp_env, priv->vsp_clob);
 519                 }
 520                 priv->vsp_clob = NULL;
 521         }
 522 }
 523
 524 static struct page *vvp_pgcache_next_page(struct vvp_seq_private *priv)
 525 {
 526         priv->vsp_page_index += 1;
 527         return vvp_pgcache_current(priv);
 528 }
 529
 530 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 531 {
 532         struct vvp_seq_private *priv = f->private;
 533
 534         if (*pos == 0) {
 535                 vvp_pgcache_rewind(priv);
 536         } else if (*pos == priv->vvp_prev_pos) {
 537                 /* Return the current item */;
 538         } else {
 539                 WARN_ON(*pos != priv->vvp_prev_pos + 1);
 540                 priv->vsp_page_index += 1;
 541         }
 542
 543         priv->vvp_prev_pos = *pos;
 544         return vvp_pgcache_current(priv);
 545 }
 546
 547 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 548 {
 549         struct vvp_seq_private *priv = f->private;
 550
 551         WARN_ON(*pos != priv->vvp_prev_pos);
 552         *pos += 1;
 553         priv->vvp_prev_pos = *pos;
 554         return vvp_pgcache_next_page(priv);
 555 }
 556
 557 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 558 {
 559         /* Nothing to do */
 560 }
 561
 562 static const struct seq_operations vvp_pgcache_ops = {
 563         .start = vvp_pgcache_start,
 564         .next  = vvp_pgcache_next,
 565         .stop  = vvp_pgcache_stop,
 566         .show  = vvp_pgcache_show
 567 };
 568
 569 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 570 {
 571         struct vvp_seq_private *priv;
 572         struct lu_site *s;
 573
 574         priv = __seq_open_private(filp, &vvp_pgcache_ops, sizeof(*priv));
 575         if (!priv)
 576                 return -ENOMEM;
 577
 578         priv->vsp_sbi = inode->i_private;
 579         priv->vsp_env = cl_env_get(&priv->vsp_refcheck);
 580         priv->vsp_clob = NULL;
 581         if (IS_ERR(priv->vsp_env)) {
 582                 int err = PTR_ERR(priv->vsp_env);
 583
 584                 seq_release_private(inode, filp);
 585                 return err;
 586         }
 587
 588         s = priv->vsp_sbi->ll_cl->cd_lu_dev.ld_site;
 589         rhashtable_walk_enter(&s->ls_obj_hash, &priv->vsp_iter);
 590
 591         return 0;
 592 }
 593
 594 static int vvp_dump_pgcache_seq_release(struct inode *inode, struct file *file)
 595 {
 596         struct seq_file *seq = file->private_data;
 597         struct vvp_seq_private *priv = seq->private;
 598
 599         if (priv->vsp_clob) {
 600                 lu_object_ref_del(&priv->vsp_clob->co_lu, "dump", current);
 601                 cl_object_put(priv->vsp_env, priv->vsp_clob);
 602         }
 603         cl_env_put(priv->vsp_env, &priv->vsp_refcheck);
 604         rhashtable_walk_exit(&priv->vsp_iter);
 605         return seq_release_private(inode, file);
 606 }
 607
 608 const struct file_operations vvp_dump_pgcache_file_ops = {
 609         .owner   = THIS_MODULE,
 610         .open    = vvp_dump_pgcache_seq_open,
 611         .read    = seq_read,
 612         .llseek  = seq_lseek,
 613         .release = vvp_dump_pgcache_seq_release,
 614 };