linux/fs/fscache/operation.c
<<
>>
Prefs
   1/* FS-Cache worker operation management routines
   2 *
   3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
   4 * Written by David Howells (dhowells@redhat.com)
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License
   8 * as published by the Free Software Foundation; either version
   9 * 2 of the License, or (at your option) any later version.
  10 *
  11 * See Documentation/filesystems/caching/operations.txt
  12 */
  13
  14#define FSCACHE_DEBUG_LEVEL OPERATION
  15#include <linux/module.h>
  16#include <linux/seq_file.h>
  17#include "internal.h"
  18
  19atomic_t fscache_op_debug_id;
  20EXPORT_SYMBOL(fscache_op_debug_id);
  21
  22/**
  23 * fscache_enqueue_operation - Enqueue an operation for processing
  24 * @op: The operation to enqueue
  25 *
  26 * Enqueue an operation for processing by the FS-Cache thread pool.
  27 *
  28 * This will get its own ref on the object.
  29 */
  30void fscache_enqueue_operation(struct fscache_operation *op)
  31{
  32        _enter("{OBJ%x OP%x,%u}",
  33               op->object->debug_id, op->debug_id, atomic_read(&op->usage));
  34
  35        fscache_set_op_state(op, "EnQ");
  36
  37        ASSERT(list_empty(&op->pend_link));
  38        ASSERT(op->processor != NULL);
  39        ASSERTCMP(op->object->state, >=, FSCACHE_OBJECT_AVAILABLE);
  40        ASSERTCMP(atomic_read(&op->usage), >, 0);
  41
  42        fscache_stat(&fscache_n_op_enqueue);
  43        switch (op->flags & FSCACHE_OP_TYPE) {
  44        case FSCACHE_OP_FAST:
  45                _debug("queue fast");
  46                atomic_inc(&op->usage);
  47                if (!schedule_work(&op->fast_work))
  48                        fscache_put_operation(op);
  49                break;
  50        case FSCACHE_OP_SLOW:
  51                _debug("queue slow");
  52                slow_work_enqueue(&op->slow_work);
  53                break;
  54        case FSCACHE_OP_MYTHREAD:
  55                _debug("queue for caller's attention");
  56                break;
  57        default:
  58                printk(KERN_ERR "FS-Cache: Unexpected op type %lx",
  59                       op->flags);
  60                BUG();
  61                break;
  62        }
  63}
  64EXPORT_SYMBOL(fscache_enqueue_operation);
  65
  66/*
  67 * start an op running
  68 */
  69static void fscache_run_op(struct fscache_object *object,
  70                           struct fscache_operation *op)
  71{
  72        fscache_set_op_state(op, "Run");
  73
  74        object->n_in_progress++;
  75        if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
  76                wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
  77        if (op->processor)
  78                fscache_enqueue_operation(op);
  79        fscache_stat(&fscache_n_op_run);
  80}
  81
  82/*
  83 * submit an exclusive operation for an object
  84 * - other ops are excluded from running simultaneously with this one
  85 * - this gets any extra refs it needs on an op
  86 */
  87int fscache_submit_exclusive_op(struct fscache_object *object,
  88                                struct fscache_operation *op)
  89{
  90        int ret;
  91
  92        _enter("{OBJ%x OP%x},", object->debug_id, op->debug_id);
  93
  94        fscache_set_op_state(op, "SubmitX");
  95
  96        spin_lock(&object->lock);
  97        ASSERTCMP(object->n_ops, >=, object->n_in_progress);
  98        ASSERTCMP(object->n_ops, >=, object->n_exclusive);
  99        ASSERT(list_empty(&op->pend_link));
 100
 101        ret = -ENOBUFS;
 102        if (fscache_object_is_active(object)) {
 103                op->object = object;
 104                object->n_ops++;
 105                object->n_exclusive++;  /* reads and writes must wait */
 106
 107                if (object->n_ops > 0) {
 108                        atomic_inc(&op->usage);
 109                        list_add_tail(&op->pend_link, &object->pending_ops);
 110                        fscache_stat(&fscache_n_op_pend);
 111                } else if (!list_empty(&object->pending_ops)) {
 112                        atomic_inc(&op->usage);
 113                        list_add_tail(&op->pend_link, &object->pending_ops);
 114                        fscache_stat(&fscache_n_op_pend);
 115                        fscache_start_operations(object);
 116                } else {
 117                        ASSERTCMP(object->n_in_progress, ==, 0);
 118                        fscache_run_op(object, op);
 119                }
 120
 121                /* need to issue a new write op after this */
 122                clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
 123                ret = 0;
 124        } else if (object->state == FSCACHE_OBJECT_CREATING) {
 125                op->object = object;
 126                object->n_ops++;
 127                object->n_exclusive++;  /* reads and writes must wait */
 128                atomic_inc(&op->usage);
 129                list_add_tail(&op->pend_link, &object->pending_ops);
 130                fscache_stat(&fscache_n_op_pend);
 131                ret = 0;
 132        } else {
 133                /* not allowed to submit ops in any other state */
 134                BUG();
 135        }
 136
 137        spin_unlock(&object->lock);
 138        return ret;
 139}
 140
 141/*
 142 * report an unexpected submission
 143 */
 144static void fscache_report_unexpected_submission(struct fscache_object *object,
 145                                                 struct fscache_operation *op,
 146                                                 unsigned long ostate)
 147{
 148        static bool once_only;
 149        struct fscache_operation *p;
 150        unsigned n;
 151
 152        if (once_only)
 153                return;
 154        once_only = true;
 155
 156        kdebug("unexpected submission OP%x [OBJ%x %s]",
 157               op->debug_id, object->debug_id,
 158               fscache_object_states[object->state]);
 159        kdebug("objstate=%s [%s]",
 160               fscache_object_states[object->state],
 161               fscache_object_states[ostate]);
 162        kdebug("objflags=%lx", object->flags);
 163        kdebug("objevent=%lx [%lx]", object->events, object->event_mask);
 164        kdebug("ops=%u inp=%u exc=%u",
 165               object->n_ops, object->n_in_progress, object->n_exclusive);
 166
 167        if (!list_empty(&object->pending_ops)) {
 168                n = 0;
 169                list_for_each_entry(p, &object->pending_ops, pend_link) {
 170                        ASSERTCMP(p->object, ==, object);
 171                        kdebug("%p %p", op->processor, op->release);
 172                        n++;
 173                }
 174
 175                kdebug("n=%u", n);
 176        }
 177
 178        dump_stack();
 179}
 180
 181/*
 182 * submit an operation for an object
 183 * - objects may be submitted only in the following states:
 184 *   - during object creation (write ops may be submitted)
 185 *   - whilst the object is active
 186 *   - after an I/O error incurred in one of the two above states (op rejected)
 187 * - this gets any extra refs it needs on an op
 188 */
 189int fscache_submit_op(struct fscache_object *object,
 190                      struct fscache_operation *op)
 191{
 192        unsigned long ostate;
 193        int ret;
 194
 195        _enter("{OBJ%x OP%x},{%u}",
 196               object->debug_id, op->debug_id, atomic_read(&op->usage));
 197
 198        ASSERTCMP(atomic_read(&op->usage), >, 0);
 199
 200        fscache_set_op_state(op, "Submit");
 201
 202        spin_lock(&object->lock);
 203        ASSERTCMP(object->n_ops, >=, object->n_in_progress);
 204        ASSERTCMP(object->n_ops, >=, object->n_exclusive);
 205        ASSERT(list_empty(&op->pend_link));
 206
 207        ostate = object->state;
 208        smp_rmb();
 209
 210        if (fscache_object_is_active(object)) {
 211                op->object = object;
 212                object->n_ops++;
 213
 214                if (object->n_exclusive > 0) {
 215                        atomic_inc(&op->usage);
 216                        list_add_tail(&op->pend_link, &object->pending_ops);
 217                        fscache_stat(&fscache_n_op_pend);
 218                } else if (!list_empty(&object->pending_ops)) {
 219                        atomic_inc(&op->usage);
 220                        list_add_tail(&op->pend_link, &object->pending_ops);
 221                        fscache_stat(&fscache_n_op_pend);
 222                        fscache_start_operations(object);
 223                } else {
 224                        ASSERTCMP(object->n_exclusive, ==, 0);
 225                        fscache_run_op(object, op);
 226                }
 227                ret = 0;
 228        } else if (object->state == FSCACHE_OBJECT_CREATING) {
 229                op->object = object;
 230                object->n_ops++;
 231                atomic_inc(&op->usage);
 232                list_add_tail(&op->pend_link, &object->pending_ops);
 233                fscache_stat(&fscache_n_op_pend);
 234                ret = 0;
 235        } else if (object->state == FSCACHE_OBJECT_DYING ||
 236                   object->state == FSCACHE_OBJECT_LC_DYING ||
 237                   object->state == FSCACHE_OBJECT_WITHDRAWING) {
 238                fscache_stat(&fscache_n_op_rejected);
 239                ret = -ENOBUFS;
 240        } else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
 241                fscache_report_unexpected_submission(object, op, ostate);
 242                ASSERT(!fscache_object_is_active(object));
 243                ret = -ENOBUFS;
 244        } else {
 245                ret = -ENOBUFS;
 246        }
 247
 248        spin_unlock(&object->lock);
 249        return ret;
 250}
 251
 252/*
 253 * queue an object for withdrawal on error, aborting all following asynchronous
 254 * operations
 255 */
 256void fscache_abort_object(struct fscache_object *object)
 257{
 258        _enter("{OBJ%x}", object->debug_id);
 259
 260        fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
 261}
 262
 263/*
 264 * jump start the operation processing on an object
 265 * - caller must hold object->lock
 266 */
 267void fscache_start_operations(struct fscache_object *object)
 268{
 269        struct fscache_operation *op;
 270        bool stop = false;
 271
 272        while (!list_empty(&object->pending_ops) && !stop) {
 273                op = list_entry(object->pending_ops.next,
 274                                struct fscache_operation, pend_link);
 275
 276                if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
 277                        if (object->n_in_progress > 0)
 278                                break;
 279                        stop = true;
 280                }
 281                list_del_init(&op->pend_link);
 282                fscache_run_op(object, op);
 283
 284                /* the pending queue was holding a ref on the object */
 285                fscache_put_operation(op);
 286        }
 287
 288        ASSERTCMP(object->n_in_progress, <=, object->n_ops);
 289
 290        _debug("woke %d ops on OBJ%x",
 291               object->n_in_progress, object->debug_id);
 292}
 293
 294/*
 295 * cancel an operation that's pending on an object
 296 */
 297int fscache_cancel_op(struct fscache_operation *op)
 298{
 299        struct fscache_object *object = op->object;
 300        int ret;
 301
 302        _enter("OBJ%x OP%x}", op->object->debug_id, op->debug_id);
 303
 304        spin_lock(&object->lock);
 305
 306        ret = -EBUSY;
 307        if (!list_empty(&op->pend_link)) {
 308                fscache_stat(&fscache_n_op_cancelled);
 309                list_del_init(&op->pend_link);
 310                object->n_ops--;
 311                if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
 312                        object->n_exclusive--;
 313                if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
 314                        wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
 315                fscache_put_operation(op);
 316                ret = 0;
 317        }
 318
 319        spin_unlock(&object->lock);
 320        _leave(" = %d", ret);
 321        return ret;
 322}
 323
 324/*
 325 * release an operation
 326 * - queues pending ops if this is the last in-progress op
 327 */
 328void fscache_put_operation(struct fscache_operation *op)
 329{
 330        struct fscache_object *object;
 331        struct fscache_cache *cache;
 332
 333        _enter("{OBJ%x OP%x,%d}",
 334               op->object->debug_id, op->debug_id, atomic_read(&op->usage));
 335
 336        ASSERTCMP(atomic_read(&op->usage), >, 0);
 337
 338        if (!atomic_dec_and_test(&op->usage))
 339                return;
 340
 341        fscache_set_op_state(op, "Put");
 342
 343        _debug("PUT OP");
 344        if (test_and_set_bit(FSCACHE_OP_DEAD, &op->flags))
 345                BUG();
 346
 347        fscache_stat(&fscache_n_op_release);
 348
 349        if (op->release) {
 350                op->release(op);
 351                op->release = NULL;
 352        }
 353
 354        object = op->object;
 355
 356        if (test_bit(FSCACHE_OP_DEC_READ_CNT, &op->flags))
 357                atomic_dec(&object->n_reads);
 358
 359        /* now... we may get called with the object spinlock held, so we
 360         * complete the cleanup here only if we can immediately acquire the
 361         * lock, and defer it otherwise */
 362        if (!spin_trylock(&object->lock)) {
 363                _debug("defer put");
 364                fscache_stat(&fscache_n_op_deferred_release);
 365
 366                cache = object->cache;
 367                spin_lock(&cache->op_gc_list_lock);
 368                list_add_tail(&op->pend_link, &cache->op_gc_list);
 369                spin_unlock(&cache->op_gc_list_lock);
 370                schedule_work(&cache->op_gc);
 371                _leave(" [defer]");
 372                return;
 373        }
 374
 375        if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
 376                ASSERTCMP(object->n_exclusive, >, 0);
 377                object->n_exclusive--;
 378        }
 379
 380        ASSERTCMP(object->n_in_progress, >, 0);
 381        object->n_in_progress--;
 382        if (object->n_in_progress == 0)
 383                fscache_start_operations(object);
 384
 385        ASSERTCMP(object->n_ops, >, 0);
 386        object->n_ops--;
 387        if (object->n_ops == 0)
 388                fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
 389
 390        spin_unlock(&object->lock);
 391
 392        kfree(op);
 393        _leave(" [done]");
 394}
 395EXPORT_SYMBOL(fscache_put_operation);
 396
 397/*
 398 * garbage collect operations that have had their release deferred
 399 */
 400void fscache_operation_gc(struct work_struct *work)
 401{
 402        struct fscache_operation *op;
 403        struct fscache_object *object;
 404        struct fscache_cache *cache =
 405                container_of(work, struct fscache_cache, op_gc);
 406        int count = 0;
 407
 408        _enter("");
 409
 410        do {
 411                spin_lock(&cache->op_gc_list_lock);
 412                if (list_empty(&cache->op_gc_list)) {
 413                        spin_unlock(&cache->op_gc_list_lock);
 414                        break;
 415                }
 416
 417                op = list_entry(cache->op_gc_list.next,
 418                                struct fscache_operation, pend_link);
 419                list_del(&op->pend_link);
 420                spin_unlock(&cache->op_gc_list_lock);
 421
 422                object = op->object;
 423
 424                _debug("GC DEFERRED REL OBJ%x OP%x",
 425                       object->debug_id, op->debug_id);
 426                fscache_stat(&fscache_n_op_gc);
 427
 428                ASSERTCMP(atomic_read(&op->usage), ==, 0);
 429
 430                spin_lock(&object->lock);
 431                if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
 432                        ASSERTCMP(object->n_exclusive, >, 0);
 433                        object->n_exclusive--;
 434                }
 435
 436                ASSERTCMP(object->n_in_progress, >, 0);
 437                object->n_in_progress--;
 438                if (object->n_in_progress == 0)
 439                        fscache_start_operations(object);
 440
 441                ASSERTCMP(object->n_ops, >, 0);
 442                object->n_ops--;
 443                if (object->n_ops == 0)
 444                        fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
 445
 446                spin_unlock(&object->lock);
 447
 448        } while (count++ < 20);
 449
 450        if (!list_empty(&cache->op_gc_list))
 451                schedule_work(&cache->op_gc);
 452
 453        _leave("");
 454}
 455
 456/*
 457 * allow the slow work item processor to get a ref on an operation
 458 */
 459static int fscache_op_get_ref(struct slow_work *work)
 460{
 461        struct fscache_operation *op =
 462                container_of(work, struct fscache_operation, slow_work);
 463
 464        atomic_inc(&op->usage);
 465        return 0;
 466}
 467
 468/*
 469 * allow the slow work item processor to discard a ref on an operation
 470 */
 471static void fscache_op_put_ref(struct slow_work *work)
 472{
 473        struct fscache_operation *op =
 474                container_of(work, struct fscache_operation, slow_work);
 475
 476        fscache_put_operation(op);
 477}
 478
 479/*
 480 * execute an operation using the slow thread pool to provide processing context
 481 * - the caller holds a ref to this object, so we don't need to hold one
 482 */
 483static void fscache_op_execute(struct slow_work *work)
 484{
 485        struct fscache_operation *op =
 486                container_of(work, struct fscache_operation, slow_work);
 487        unsigned long start;
 488
 489        _enter("{OBJ%x OP%x,%d}",
 490               op->object->debug_id, op->debug_id, atomic_read(&op->usage));
 491
 492        ASSERT(op->processor != NULL);
 493        start = jiffies;
 494        op->processor(op);
 495        fscache_hist(fscache_ops_histogram, start);
 496
 497        _leave("");
 498}
 499
 500/*
 501 * describe an operation for slow-work debugging
 502 */
 503#ifdef CONFIG_SLOW_WORK_PROC
 504static void fscache_op_desc(struct slow_work *work, struct seq_file *m)
 505{
 506        struct fscache_operation *op =
 507                container_of(work, struct fscache_operation, slow_work);
 508
 509        seq_printf(m, "FSC: OBJ%x OP%x: %s/%s fl=%lx",
 510                   op->object->debug_id, op->debug_id,
 511                   op->name, op->state, op->flags);
 512}
 513#endif
 514
 515const struct slow_work_ops fscache_op_slow_work_ops = {
 516        .owner          = THIS_MODULE,
 517        .get_ref        = fscache_op_get_ref,
 518        .put_ref        = fscache_op_put_ref,
 519        .execute        = fscache_op_execute,
 520#ifdef CONFIG_SLOW_WORK_PROC
 521        .desc           = fscache_op_desc,
 522#endif
 523};
 524
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.