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