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
  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 /**
 269  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 270  * pressure, when environments cannot be allocated.
 271  */
 272 int vvp_global_init(void)
 273 {
 274         int rc;
 275
 276         rc = lu_kmem_init(vvp_caches);
 277         if (rc != 0)
 278                 return rc;
 279
 280         rc = lu_device_type_init(&vvp_device_type);
 281         if (rc != 0)
 282                 goto out_kmem;
 283
 284         return 0;
 285
 286 out_kmem:
 287         lu_kmem_fini(vvp_caches);
 288
 289         return rc;
 290 }
 291
 292 void vvp_global_fini(void)
 293 {
 294         lu_device_type_fini(&vvp_device_type);
 295         lu_kmem_fini(vvp_caches);
 296 }
 297
 298 /*****************************************************************************
 299  *
 300  * mirror obd-devices into cl devices.
 301  *
 302  */
 303
 304 int cl_sb_init(struct super_block *sb)
 305 {
 306         struct ll_sb_info *sbi;
 307         struct cl_device  *cl;
 308         struct lu_env     *env;
 309         int rc = 0;
 310         __u16 refcheck;
 311
 312         sbi  = ll_s2sbi(sb);
 313         env = cl_env_get(&refcheck);
 314         if (!IS_ERR(env)) {
 315                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 316                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 317                 if (!IS_ERR(cl)) {
 318                         sbi->ll_cl = cl;
 319                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 320                 }
 321                 cl_env_put(env, &refcheck);
 322         } else
 323                 rc = PTR_ERR(env);
 324         RETURN(rc);
 325 }
 326
 327 int cl_sb_fini(struct super_block *sb)
 328 {
 329         struct ll_sb_info *sbi;
 330         struct lu_env     *env;
 331         struct cl_device  *cld;
 332         __u16              refcheck;
 333         int                result;
 334
 335         ENTRY;
 336         sbi = ll_s2sbi(sb);
 337         env = cl_env_get(&refcheck);
 338         if (!IS_ERR(env)) {
 339                 cld = sbi->ll_cl;
 340
 341                 if (cld != NULL) {
 342                         cl_stack_fini(env, cld);
 343                         sbi->ll_cl = NULL;
 344                         sbi->ll_site = NULL;
 345                 }
 346                 cl_env_put(env, &refcheck);
 347                 result = 0;
 348         } else {
 349                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 350                 result = PTR_ERR(env);
 351         }
 352
 353         RETURN(result);
 354 }
 355
 356 /****************************************************************************
 357  *
 358  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 359  *
 360  ****************************************************************************/
 361
 362 /*
 363  * To represent contents of a page cache as a byte stream, following
 364  * information if encoded in 64bit offset:
 365  *
 366  *       - file hash bucket in lu_site::ls_hash[]       28bits
 367  *
 368  *       - how far file is from bucket head              4bits
 369  *
 370  *       - page index                                   32bits
 371  *
 372  * First two data identify a file in the cache uniquely.
 373  */
 374
 375 #define PGC_OBJ_SHIFT (32 + 4)
 376 #define PGC_DEPTH_SHIFT (32)
 377
 378 struct vvp_pgcache_id {
 379         unsigned                 vpi_bucket;
 380         unsigned                 vpi_depth;
 381         uint32_t                 vpi_index;
 382
 383         unsigned                 vpi_curdep;
 384         struct lu_object_header *vpi_obj;
 385 };
 386
 387 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 388 {
 389         CLASSERT(sizeof(pos) == sizeof(__u64));
 390
 391         id->vpi_index  = pos & 0xffffffff;
 392         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 393         id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
 394 }
 395
 396 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 397 {
 398         return
 399                 ((__u64)id->vpi_index) |
 400                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 401                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 402 }
 403
 404 static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 405                                struct hlist_node *hnode, void *data)
 406 {
 407         struct vvp_pgcache_id   *id  = data;
 408         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 409
 410         if (id->vpi_curdep-- > 0)
 411                 return 0; /* continue */
 412
 413         if (lu_object_is_dying(hdr))
 414                 return 1;
 415
 416         cfs_hash_get(hs, hnode);
 417         id->vpi_obj = hdr;
 418         return 1;
 419 }
 420
 421 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 422                                          struct lu_device *dev,
 423                                          struct vvp_pgcache_id *id)
 424 {
 425         LASSERT(lu_device_is_cl(dev));
 426
 427         id->vpi_depth &= 0xf;
 428         id->vpi_obj    = NULL;
 429         id->vpi_curdep = id->vpi_depth;
 430
 431         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 432                                 vvp_pgcache_obj_get, id);
 433         if (id->vpi_obj != NULL) {
 434                 struct lu_object *lu_obj;
 435
 436                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 437                 if (lu_obj != NULL) {
 438                         lu_object_ref_add(lu_obj, "dump", current);
 439                         return lu2cl(lu_obj);
 440                 }
 441                 lu_object_put(env, lu_object_top(id->vpi_obj));
 442
 443         } else if (id->vpi_curdep > 0) {
 444                 id->vpi_depth = 0xf;
 445         }
 446         return NULL;
 447 }
 448
 449 static loff_t vvp_pgcache_find(const struct lu_env *env,
 450                                struct lu_device *dev, loff_t pos)
 451 {
 452         struct cl_object     *clob;
 453         struct lu_site       *site;
 454         struct vvp_pgcache_id id;
 455
 456         site = dev->ld_site;
 457         vvp_pgcache_id_unpack(pos, &id);
 458
 459         while (1) {
 460                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 461                         return ~0ULL;
 462                 clob = vvp_pgcache_obj(env, dev, &id);
 463                 if (clob != NULL) {
 464                         struct inode *inode = vvp_object_inode(clob);
 465                         struct page *vmpage;
 466                         int nr;
 467
 468                         nr = find_get_pages_contig(inode->i_mapping,
 469                                                    id.vpi_index, 1, &vmpage);
 470                         if (nr > 0) {
 471                                 id.vpi_index = vmpage->index;
 472                                 /* Cant support over 16T file */
 473                                 nr = !(vmpage->index > 0xffffffff);
 474                                 put_page(vmpage);
 475                         }
 476
 477                         lu_object_ref_del(&clob->co_lu, "dump", current);
 478                         cl_object_put(env, clob);
 479                         if (nr > 0)
 480                                 return vvp_pgcache_id_pack(&id);
 481                 }
 482                 /* to the next object. */
 483                 ++id.vpi_depth;
 484                 id.vpi_depth &= 0xf;
 485                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 486                         return ~0ULL;
 487                 id.vpi_index = 0;
 488         }
 489 }
 490
 491 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 492         if (test_bit(PG_##flag, &(page)->flags)) {                  \
 493                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 494                 has_flags = 1;                                          \
 495         }                                                               \
 496 } while(0)
 497
 498 static void vvp_pgcache_page_show(const struct lu_env *env,
 499                                   struct seq_file *seq, struct cl_page *page)
 500 {
 501         struct vvp_page *vpg;
 502         struct page      *vmpage;
 503         int              has_flags;
 504
 505         vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
 506         vmpage = vpg->vpg_page;
 507         seq_printf(seq, " %5i | %p %p %s %s %s | %p "DFID"(%p) %lu %u [",
 508                    0 /* gen */,
 509                    vpg, page,
 510                    "none",
 511                    vpg->vpg_defer_uptodate ? "du" : "- ",
 512                    PageWriteback(vmpage) ? "wb" : "-",
 513                    vmpage,
 514                    PFID(ll_inode2fid(vmpage->mapping->host)),
 515                    vmpage->mapping->host, vmpage->index,
 516                    page_count(vmpage));
 517         has_flags = 0;
 518         seq_page_flag(seq, vmpage, locked, has_flags);
 519         seq_page_flag(seq, vmpage, error, has_flags);
 520         seq_page_flag(seq, vmpage, referenced, has_flags);
 521         seq_page_flag(seq, vmpage, uptodate, has_flags);
 522         seq_page_flag(seq, vmpage, dirty, has_flags);
 523         seq_page_flag(seq, vmpage, writeback, has_flags);
 524         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 525 }
 526
 527 static int vvp_pgcache_show(struct seq_file *f, void *v)
 528 {
 529         loff_t                   pos;
 530         struct ll_sb_info       *sbi;
 531         struct cl_object        *clob;
 532         struct lu_env           *env;
 533         struct vvp_pgcache_id    id;
 534         __u16                    refcheck;
 535         int                      result;
 536
 537         env = cl_env_get(&refcheck);
 538         if (!IS_ERR(env)) {
 539                 pos = *(loff_t *) v;
 540                 vvp_pgcache_id_unpack(pos, &id);
 541                 sbi = f->private;
 542                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 543                 if (clob != NULL) {
 544                         struct inode *inode = vvp_object_inode(clob);
 545                         struct cl_page *page = NULL;
 546                         struct page *vmpage;
 547
 548                         result = find_get_pages_contig(inode->i_mapping,
 549                                                       id.vpi_index, 1, &vmpage);
 550                         if (result > 0) {
 551                                 lock_page(vmpage);
 552                                 page = cl_vmpage_page(vmpage, clob);
 553                                 unlock_page(vmpage);
 554
 555                                 put_page(vmpage);
 556                         }
 557
 558                         seq_printf(f, "%8x@"DFID": ", id.vpi_index,
 559                                    PFID(lu_object_fid(&clob->co_lu)));
 560                         if (page != NULL) {
 561                                 vvp_pgcache_page_show(env, f, page);
 562                                 cl_page_put(env, page);
 563                         } else
 564                                 seq_puts(f, "missing\n");
 565                         lu_object_ref_del(&clob->co_lu, "dump", current);
 566                         cl_object_put(env, clob);
 567                 } else
 568                         seq_printf(f, "%llx missing\n", pos);
 569                 cl_env_put(env, &refcheck);
 570                 result = 0;
 571         } else
 572                 result = PTR_ERR(env);
 573         return result;
 574 }
 575
 576 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 577 {
 578         struct ll_sb_info *sbi;
 579         struct lu_env     *env;
 580         __u16              refcheck;
 581
 582         sbi = f->private;
 583
 584         env = cl_env_get(&refcheck);
 585         if (!IS_ERR(env)) {
 586                 sbi = f->private;
 587                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 588                         pos = ERR_PTR(-EFBIG);
 589                 else {
 590                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 591                                                 *pos);
 592                         if (*pos == ~0ULL)
 593                                 pos = NULL;
 594                 }
 595                 cl_env_put(env, &refcheck);
 596         }
 597         return pos;
 598 }
 599
 600 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 601 {
 602         struct ll_sb_info *sbi;
 603         struct lu_env     *env;
 604         __u16              refcheck;
 605
 606         env = cl_env_get(&refcheck);
 607         if (!IS_ERR(env)) {
 608                 sbi = f->private;
 609                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 610                 if (*pos == ~0ULL)
 611                         pos = NULL;
 612                 cl_env_put(env, &refcheck);
 613         }
 614         return pos;
 615 }
 616
 617 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 618 {
 619         /* Nothing to do */
 620 }
 621
 622 static struct seq_operations vvp_pgcache_ops = {
 623         .start = vvp_pgcache_start,
 624         .next  = vvp_pgcache_next,
 625         .stop  = vvp_pgcache_stop,
 626         .show  = vvp_pgcache_show
 627 };
 628
 629 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 630 {
 631         struct ll_sb_info       *sbi = PDE_DATA(inode);
 632         struct seq_file         *seq;
 633         int                     result;
 634
 635         result = seq_open(filp, &vvp_pgcache_ops);
 636         if (result == 0) {
 637                 seq = filp->private_data;
 638                 seq->private = sbi;
 639         }
 640         return result;
 641 }
 642
 643 const struct file_operations vvp_dump_pgcache_file_ops = {
 644         .owner   = THIS_MODULE,
 645         .open    = vvp_dump_pgcache_seq_open,
 646         .read    = seq_read,
 647         .llseek  = seq_lseek,
 648         .release = seq_release,
 649 };