drivers/staging/lustre/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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * cl_device and cl_device_type implementation for VVP layer.
  37  *
  38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43
  44 #include "../include/obd.h"
  45 #include "../include/lustre_lite.h"
  46 #include "llite_internal.h"
  47 #include "vvp_internal.h"
  48
  49 /*****************************************************************************
  50  *
  51  * Vvp device and device type functions.
  52  *
  53  */
  54
  55 /*
  56  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  57  * "llite_" (var. "ll_") prefix.
  58  */
  59
  60 static struct kmem_cache *vvp_thread_kmem;
  61 static struct kmem_cache *vvp_session_kmem;
  62 static struct lu_kmem_descr vvp_caches[] = {
  63         {
  64                 .ckd_cache = &vvp_thread_kmem,
  65                 .ckd_name  = "vvp_thread_kmem",
  66                 .ckd_size  = sizeof(struct vvp_thread_info),
  67         },
  68         {
  69                 .ckd_cache = &vvp_session_kmem,
  70                 .ckd_name  = "vvp_session_kmem",
  71                 .ckd_size  = sizeof(struct vvp_session)
  72         },
  73         {
  74                 .ckd_cache = NULL
  75         }
  76 };
  77
  78 static void *vvp_key_init(const struct lu_context *ctx,
  79                           struct lu_context_key *key)
  80 {
  81         struct vvp_thread_info *info;
  82
  83         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, GFP_NOFS);
  84         if (info == NULL)
  85                 info = ERR_PTR(-ENOMEM);
  86         return info;
  87 }
  88
  89 static void vvp_key_fini(const struct lu_context *ctx,
  90                          struct lu_context_key *key, void *data)
  91 {
  92         struct vvp_thread_info *info = data;
  93
  94         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
  95 }
  96
  97 static void *vvp_session_key_init(const struct lu_context *ctx,
  98                                   struct lu_context_key *key)
  99 {
 100         struct vvp_session *session;
 101
 102         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
 103         if (session == NULL)
 104                 session = ERR_PTR(-ENOMEM);
 105         return session;
 106 }
 107
 108 static void vvp_session_key_fini(const struct lu_context *ctx,
 109                                  struct lu_context_key *key, void *data)
 110 {
 111         struct vvp_session *session = data;
 112
 113         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 114 }
 115
 116
 117 struct lu_context_key vvp_key = {
 118         .lct_tags = LCT_CL_THREAD,
 119         .lct_init = vvp_key_init,
 120         .lct_fini = vvp_key_fini
 121 };
 122
 123 struct lu_context_key vvp_session_key = {
 124         .lct_tags = LCT_SESSION,
 125         .lct_init = vvp_session_key_init,
 126         .lct_fini = vvp_session_key_fini
 127 };
 128
 129 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 130 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
 131
 132 static const struct lu_device_operations vvp_lu_ops = {
 133         .ldo_object_alloc      = vvp_object_alloc
 134 };
 135
 136 static const struct cl_device_operations vvp_cl_ops = {
 137         .cdo_req_init = ccc_req_init
 138 };
 139
 140 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 141                                           struct lu_device_type *t,
 142                                           struct lustre_cfg *cfg)
 143 {
 144         return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
 145 }
 146
 147 static const struct lu_device_type_operations vvp_device_type_ops = {
 148         .ldto_init = vvp_type_init,
 149         .ldto_fini = vvp_type_fini,
 150
 151         .ldto_start = vvp_type_start,
 152         .ldto_stop  = vvp_type_stop,
 153
 154         .ldto_device_alloc = vvp_device_alloc,
 155         .ldto_device_free  = ccc_device_free,
 156         .ldto_device_init  = ccc_device_init,
 157         .ldto_device_fini  = ccc_device_fini
 158 };
 159
 160 struct lu_device_type vvp_device_type = {
 161         .ldt_tags     = LU_DEVICE_CL,
 162         .ldt_name     = LUSTRE_VVP_NAME,
 163         .ldt_ops      = &vvp_device_type_ops,
 164         .ldt_ctx_tags = LCT_CL_THREAD
 165 };
 166
 167 /**
 168  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 169  * pressure, when environments cannot be allocated.
 170  */
 171 int vvp_global_init(void)
 172 {
 173         int result;
 174
 175         result = lu_kmem_init(vvp_caches);
 176         if (result == 0) {
 177                 result = ccc_global_init(&vvp_device_type);
 178                 if (result != 0)
 179                         lu_kmem_fini(vvp_caches);
 180         }
 181         return result;
 182 }
 183
 184 void vvp_global_fini(void)
 185 {
 186         ccc_global_fini(&vvp_device_type);
 187         lu_kmem_fini(vvp_caches);
 188 }
 189
 190
 191 /*****************************************************************************
 192  *
 193  * mirror obd-devices into cl devices.
 194  *
 195  */
 196
 197 int cl_sb_init(struct super_block *sb)
 198 {
 199         struct ll_sb_info *sbi;
 200         struct cl_device  *cl;
 201         struct lu_env     *env;
 202         int rc = 0;
 203         int refcheck;
 204
 205         sbi  = ll_s2sbi(sb);
 206         env = cl_env_get(&refcheck);
 207         if (!IS_ERR(env)) {
 208                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 209                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 210                 if (!IS_ERR(cl)) {
 211                         cl2ccc_dev(cl)->cdv_sb = sb;
 212                         sbi->ll_cl = cl;
 213                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 214                 }
 215                 cl_env_put(env, &refcheck);
 216         } else
 217                 rc = PTR_ERR(env);
 218         return rc;
 219 }
 220
 221 int cl_sb_fini(struct super_block *sb)
 222 {
 223         struct ll_sb_info *sbi;
 224         struct lu_env     *env;
 225         struct cl_device  *cld;
 226         int             refcheck;
 227         int             result;
 228
 229         sbi = ll_s2sbi(sb);
 230         env = cl_env_get(&refcheck);
 231         if (!IS_ERR(env)) {
 232                 cld = sbi->ll_cl;
 233
 234                 if (cld != NULL) {
 235                         cl_stack_fini(env, cld);
 236                         sbi->ll_cl = NULL;
 237                         sbi->ll_site = NULL;
 238                 }
 239                 cl_env_put(env, &refcheck);
 240                 result = 0;
 241         } else {
 242                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 243                 result = PTR_ERR(env);
 244         }
 245         /*
 246          * If mount failed (sbi->ll_cl == NULL), and this there are no other
 247          * mounts, stop device types manually (this usually happens
 248          * automatically when last device is destroyed).
 249          */
 250         lu_types_stop();
 251         return result;
 252 }
 253
 254 /****************************************************************************
 255  *
 256  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 257  *
 258  ****************************************************************************/
 259
 260 /*
 261  * To represent contents of a page cache as a byte stream, following
 262  * information if encoded in 64bit offset:
 263  *
 264  *       - file hash bucket in lu_site::ls_hash[]       28bits
 265  *
 266  *       - how far file is from bucket head           4bits
 267  *
 268  *       - page index                              32bits
 269  *
 270  * First two data identify a file in the cache uniquely.
 271  */
 272
 273 #define PGC_OBJ_SHIFT (32 + 4)
 274 #define PGC_DEPTH_SHIFT (32)
 275
 276 struct vvp_pgcache_id {
 277         unsigned                 vpi_bucket;
 278         unsigned                 vpi_depth;
 279         uint32_t                 vpi_index;
 280
 281         unsigned                 vpi_curdep;
 282         struct lu_object_header *vpi_obj;
 283 };
 284
 285 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 286 {
 287         CLASSERT(sizeof(pos) == sizeof(__u64));
 288
 289         id->vpi_index  = pos & 0xffffffff;
 290         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 291         id->vpi_bucket = (unsigned long long)pos >> PGC_OBJ_SHIFT;
 292 }
 293
 294 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 295 {
 296         return
 297                 ((__u64)id->vpi_index) |
 298                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 299                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 300 }
 301
 302 static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 303                                struct hlist_node *hnode, void *data)
 304 {
 305         struct vvp_pgcache_id   *id  = data;
 306         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 307
 308         if (id->vpi_curdep-- > 0)
 309                 return 0; /* continue */
 310
 311         if (lu_object_is_dying(hdr))
 312                 return 1;
 313
 314         cfs_hash_get(hs, hnode);
 315         id->vpi_obj = hdr;
 316         return 1;
 317 }
 318
 319 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 320                                          struct lu_device *dev,
 321                                          struct vvp_pgcache_id *id)
 322 {
 323         LASSERT(lu_device_is_cl(dev));
 324
 325         id->vpi_depth &= 0xf;
 326         id->vpi_obj    = NULL;
 327         id->vpi_curdep = id->vpi_depth;
 328
 329         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 330                                 vvp_pgcache_obj_get, id);
 331         if (id->vpi_obj != NULL) {
 332                 struct lu_object *lu_obj;
 333
 334                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 335                 if (lu_obj != NULL) {
 336                         lu_object_ref_add(lu_obj, "dump", current);
 337                         return lu2cl(lu_obj);
 338                 }
 339                 lu_object_put(env, lu_object_top(id->vpi_obj));
 340
 341         } else if (id->vpi_curdep > 0) {
 342                 id->vpi_depth = 0xf;
 343         }
 344         return NULL;
 345 }
 346
 347 static loff_t vvp_pgcache_find(const struct lu_env *env,
 348                                struct lu_device *dev, loff_t pos)
 349 {
 350         struct cl_object     *clob;
 351         struct lu_site       *site;
 352         struct vvp_pgcache_id id;
 353
 354         site = dev->ld_site;
 355         vvp_pgcache_id_unpack(pos, &id);
 356
 357         while (1) {
 358                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 359                         return ~0ULL;
 360                 clob = vvp_pgcache_obj(env, dev, &id);
 361                 if (clob != NULL) {
 362                         struct cl_object_header *hdr;
 363                         int                   nr;
 364                         struct cl_page    *pg;
 365
 366                         /* got an object. Find next page. */
 367                         hdr = cl_object_header(clob);
 368
 369                         spin_lock(&hdr->coh_page_guard);
 370                         nr = radix_tree_gang_lookup(&hdr->coh_tree,
 371                                                     (void **)&pg,
 372                                                     id.vpi_index, 1);
 373                         if (nr > 0) {
 374                                 id.vpi_index = pg->cp_index;
 375                                 /* Cant support over 16T file */
 376                                 nr = !(pg->cp_index > 0xffffffff);
 377                         }
 378                         spin_unlock(&hdr->coh_page_guard);
 379
 380                         lu_object_ref_del(&clob->co_lu, "dump", current);
 381                         cl_object_put(env, clob);
 382                         if (nr > 0)
 383                                 return vvp_pgcache_id_pack(&id);
 384                 }
 385                 /* to the next object. */
 386                 ++id.vpi_depth;
 387                 id.vpi_depth &= 0xf;
 388                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 389                         return ~0ULL;
 390                 id.vpi_index = 0;
 391         }
 392 }
 393
 394 #define seq_page_flag(seq, page, flag, has_flags) do {            \
 395         if (test_bit(PG_##flag, &(page)->flags)) {                \
 396                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 397                 has_flags = 1;                                    \
 398         }                                                              \
 399 } while (0)
 400
 401 static void vvp_pgcache_page_show(const struct lu_env *env,
 402                                   struct seq_file *seq, struct cl_page *page)
 403 {
 404         struct ccc_page *cpg;
 405         struct page      *vmpage;
 406         int           has_flags;
 407
 408         cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
 409         vmpage = cpg->cpg_page;
 410         seq_printf(seq, " %5i | %p %p %s %s %s %s | %p %lu/%u(%p) %lu %u [",
 411                    0 /* gen */,
 412                    cpg, page,
 413                    "none",
 414                    cpg->cpg_write_queued ? "wq" : "- ",
 415                    cpg->cpg_defer_uptodate ? "du" : "- ",
 416                    PageWriteback(vmpage) ? "wb" : "-",
 417                    vmpage, vmpage->mapping->host->i_ino,
 418                    vmpage->mapping->host->i_generation,
 419                    vmpage->mapping->host, vmpage->index,
 420                    page_count(vmpage));
 421         has_flags = 0;
 422         seq_page_flag(seq, vmpage, locked, has_flags);
 423         seq_page_flag(seq, vmpage, error, has_flags);
 424         seq_page_flag(seq, vmpage, referenced, has_flags);
 425         seq_page_flag(seq, vmpage, uptodate, has_flags);
 426         seq_page_flag(seq, vmpage, dirty, has_flags);
 427         seq_page_flag(seq, vmpage, writeback, has_flags);
 428         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 429 }
 430
 431 static int vvp_pgcache_show(struct seq_file *f, void *v)
 432 {
 433         loff_t             pos;
 434         struct ll_sb_info       *sbi;
 435         struct cl_object        *clob;
 436         struct lu_env      *env;
 437         struct cl_page    *page;
 438         struct cl_object_header *hdr;
 439         struct vvp_pgcache_id    id;
 440         int                   refcheck;
 441         int                   result;
 442
 443         env = cl_env_get(&refcheck);
 444         if (!IS_ERR(env)) {
 445                 pos = *(loff_t *) v;
 446                 vvp_pgcache_id_unpack(pos, &id);
 447                 sbi = f->private;
 448                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 449                 if (clob != NULL) {
 450                         hdr = cl_object_header(clob);
 451
 452                         spin_lock(&hdr->coh_page_guard);
 453                         page = cl_page_lookup(hdr, id.vpi_index);
 454                         spin_unlock(&hdr->coh_page_guard);
 455
 456                         seq_printf(f, "%8x@"DFID": ",
 457                                    id.vpi_index, PFID(&hdr->coh_lu.loh_fid));
 458                         if (page != NULL) {
 459                                 vvp_pgcache_page_show(env, f, page);
 460                                 cl_page_put(env, page);
 461                         } else
 462                                 seq_puts(f, "missing\n");
 463                         lu_object_ref_del(&clob->co_lu, "dump", current);
 464                         cl_object_put(env, clob);
 465                 } else
 466                         seq_printf(f, "%llx missing\n", pos);
 467                 cl_env_put(env, &refcheck);
 468                 result = 0;
 469         } else
 470                 result = PTR_ERR(env);
 471         return result;
 472 }
 473
 474 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 475 {
 476         struct ll_sb_info *sbi;
 477         struct lu_env     *env;
 478         int             refcheck;
 479
 480         sbi = f->private;
 481
 482         env = cl_env_get(&refcheck);
 483         if (!IS_ERR(env)) {
 484                 sbi = f->private;
 485                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 486                         pos = ERR_PTR(-EFBIG);
 487                 else {
 488                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 489                                                 *pos);
 490                         if (*pos == ~0ULL)
 491                                 pos = NULL;
 492                 }
 493                 cl_env_put(env, &refcheck);
 494         }
 495         return pos;
 496 }
 497
 498 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 499 {
 500         struct ll_sb_info *sbi;
 501         struct lu_env     *env;
 502         int             refcheck;
 503
 504         env = cl_env_get(&refcheck);
 505         if (!IS_ERR(env)) {
 506                 sbi = f->private;
 507                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 508                 if (*pos == ~0ULL)
 509                         pos = NULL;
 510                 cl_env_put(env, &refcheck);
 511         }
 512         return pos;
 513 }
 514
 515 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 516 {
 517         /* Nothing to do */
 518 }
 519
 520 static struct seq_operations vvp_pgcache_ops = {
 521         .start = vvp_pgcache_start,
 522         .next  = vvp_pgcache_next,
 523         .stop  = vvp_pgcache_stop,
 524         .show  = vvp_pgcache_show
 525 };
 526
 527 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 528 {
 529         struct ll_sb_info     *sbi = PDE_DATA(inode);
 530         struct seq_file       *seq;
 531         int                 result;
 532
 533         result = seq_open(filp, &vvp_pgcache_ops);
 534         if (result == 0) {
 535                 seq = filp->private_data;
 536                 seq->private = sbi;
 537         }
 538         return result;
 539 }
 540
 541 const struct file_operations vvp_dump_pgcache_file_ops = {
 542         .owner   = THIS_MODULE,
 543         .open    = vvp_dump_pgcache_seq_open,
 544         .read    = seq_read,
 545         .llseek  = seq_lseek,
 546         .release = seq_release,
 547 };