1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/stdinc.h" 35 36 #include "spdk/blobfs.h" 37 #include "spdk/conf.h" 38 #include "blobfs_internal.h" 39 40 #include "spdk/queue.h" 41 #include "spdk/io_channel.h" 42 #include "spdk/assert.h" 43 #include "spdk/env.h" 44 #include "spdk/util.h" 45 #include "spdk_internal/log.h" 46 47 #define BLOBFS_TRACE(file, str, args...) \ 48 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s " str, file->name, ##args) 49 50 #define BLOBFS_TRACE_RW(file, str, args...) \ 51 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS_RW, "file=%s " str, file->name, ##args) 52 53 #define BLOBFS_CACHE_SIZE (4ULL * 1024 * 1024 * 1024) 54 #define SPDK_BLOBFS_OPTS_CLUSTER_SZ (1024 * 1024) 55 56 static uint64_t g_fs_cache_size = BLOBFS_CACHE_SIZE; 57 static struct spdk_mempool *g_cache_pool; 58 static TAILQ_HEAD(, spdk_file) g_caches; 59 static int g_fs_count = 0; 60 static pthread_mutex_t g_cache_init_lock = PTHREAD_MUTEX_INITIALIZER; 61 static pthread_spinlock_t g_caches_lock; 62 63 static void 64 __sem_post(void *arg, int bserrno) 65 { 66 sem_t *sem = arg; 67 68 sem_post(sem); 69 } 70 71 void 72 spdk_cache_buffer_free(struct cache_buffer *cache_buffer) 73 { 74 spdk_mempool_put(g_cache_pool, cache_buffer->buf); 75 free(cache_buffer); 76 } 77 78 #define CACHE_READAHEAD_THRESHOLD (128 * 1024) 79 80 struct spdk_file { 81 struct spdk_filesystem *fs; 82 struct spdk_blob *blob; 83 char *name; 84 uint64_t length; 85 bool is_deleted; 86 bool open_for_writing; 87 uint64_t length_flushed; 88 uint64_t append_pos; 89 uint64_t seq_byte_count; 90 uint64_t next_seq_offset; 91 uint32_t priority; 92 TAILQ_ENTRY(spdk_file) tailq; 93 spdk_blob_id blobid; 94 uint32_t ref_count; 95 pthread_spinlock_t lock; 96 struct cache_buffer *last; 97 struct cache_tree *tree; 98 TAILQ_HEAD(open_requests_head, spdk_fs_request) open_requests; 99 TAILQ_HEAD(sync_requests_head, spdk_fs_request) sync_requests; 100 TAILQ_ENTRY(spdk_file) cache_tailq; 101 }; 102 103 struct spdk_deleted_file { 104 spdk_blob_id id; 105 TAILQ_ENTRY(spdk_deleted_file) tailq; 106 }; 107 108 struct spdk_filesystem { 109 struct spdk_blob_store *bs; 110 TAILQ_HEAD(, spdk_file) files; 111 struct spdk_bs_opts bs_opts; 112 struct spdk_bs_dev *bdev; 113 fs_send_request_fn send_request; 114 115 struct { 116 uint32_t max_ops; 117 struct spdk_io_channel *sync_io_channel; 118 struct spdk_fs_channel *sync_fs_channel; 119 } sync_target; 120 121 struct { 122 uint32_t max_ops; 123 struct spdk_io_channel *md_io_channel; 124 struct spdk_fs_channel *md_fs_channel; 125 } md_target; 126 127 struct { 128 uint32_t max_ops; 129 } io_target; 130 }; 131 132 struct spdk_fs_cb_args { 133 union { 134 spdk_fs_op_with_handle_complete fs_op_with_handle; 135 spdk_fs_op_complete fs_op; 136 spdk_file_op_with_handle_complete file_op_with_handle; 137 spdk_file_op_complete file_op; 138 spdk_file_stat_op_complete stat_op; 139 } fn; 140 void *arg; 141 sem_t *sem; 142 struct spdk_filesystem *fs; 143 struct spdk_file *file; 144 int rc; 145 bool from_request; 146 union { 147 struct { 148 TAILQ_HEAD(, spdk_deleted_file) deleted_files; 149 } fs_load; 150 struct { 151 uint64_t length; 152 } truncate; 153 struct { 154 struct spdk_io_channel *channel; 155 void *user_buf; 156 void *pin_buf; 157 int is_read; 158 off_t offset; 159 size_t length; 160 uint64_t start_page; 161 uint64_t num_pages; 162 uint32_t blocklen; 163 } rw; 164 struct { 165 const char *old_name; 166 const char *new_name; 167 } rename; 168 struct { 169 struct cache_buffer *cache_buffer; 170 uint64_t length; 171 } flush; 172 struct { 173 struct cache_buffer *cache_buffer; 174 uint64_t length; 175 uint64_t offset; 176 } readahead; 177 struct { 178 uint64_t offset; 179 TAILQ_ENTRY(spdk_fs_request) tailq; 180 bool xattr_in_progress; 181 } sync; 182 struct { 183 uint32_t num_clusters; 184 } resize; 185 struct { 186 const char *name; 187 uint32_t flags; 188 TAILQ_ENTRY(spdk_fs_request) tailq; 189 } open; 190 struct { 191 const char *name; 192 struct spdk_blob *blob; 193 } create; 194 struct { 195 const char *name; 196 } delete; 197 struct { 198 const char *name; 199 } stat; 200 } op; 201 }; 202 203 static void cache_free_buffers(struct spdk_file *file); 204 205 void 206 spdk_fs_opts_init(struct spdk_blobfs_opts *opts) 207 { 208 opts->cluster_sz = SPDK_BLOBFS_OPTS_CLUSTER_SZ; 209 } 210 211 static void 212 __initialize_cache(void) 213 { 214 assert(g_cache_pool == NULL); 215 216 g_cache_pool = spdk_mempool_create("spdk_fs_cache", 217 g_fs_cache_size / CACHE_BUFFER_SIZE, 218 CACHE_BUFFER_SIZE, 219 SPDK_MEMPOOL_DEFAULT_CACHE_SIZE, 220 SPDK_ENV_SOCKET_ID_ANY); 221 TAILQ_INIT(&g_caches); 222 pthread_spin_init(&g_caches_lock, 0); 223 } 224 225 static void 226 __free_cache(void) 227 { 228 assert(g_cache_pool != NULL); 229 230 spdk_mempool_free(g_cache_pool); 231 g_cache_pool = NULL; 232 } 233 234 static uint64_t 235 __file_get_blob_size(struct spdk_file *file) 236 { 237 uint64_t cluster_sz; 238 239 cluster_sz = file->fs->bs_opts.cluster_sz; 240 return cluster_sz * spdk_blob_get_num_clusters(file->blob); 241 } 242 243 struct spdk_fs_request { 244 struct spdk_fs_cb_args args; 245 TAILQ_ENTRY(spdk_fs_request) link; 246 struct spdk_fs_channel *channel; 247 }; 248 249 struct spdk_fs_channel { 250 struct spdk_fs_request *req_mem; 251 TAILQ_HEAD(, spdk_fs_request) reqs; 252 sem_t sem; 253 struct spdk_filesystem *fs; 254 struct spdk_io_channel *bs_channel; 255 fs_send_request_fn send_request; 256 bool sync; 257 pthread_spinlock_t lock; 258 }; 259 260 static struct spdk_fs_request * 261 alloc_fs_request(struct spdk_fs_channel *channel) 262 { 263 struct spdk_fs_request *req; 264 265 if (channel->sync) { 266 pthread_spin_lock(&channel->lock); 267 } 268 269 req = TAILQ_FIRST(&channel->reqs); 270 if (req) { 271 TAILQ_REMOVE(&channel->reqs, req, link); 272 } 273 274 if (channel->sync) { 275 pthread_spin_unlock(&channel->lock); 276 } 277 278 if (req == NULL) { 279 return NULL; 280 } 281 memset(req, 0, sizeof(*req)); 282 req->channel = channel; 283 req->args.from_request = true; 284 285 return req; 286 } 287 288 static void 289 free_fs_request(struct spdk_fs_request *req) 290 { 291 struct spdk_fs_channel *channel = req->channel; 292 293 if (channel->sync) { 294 pthread_spin_lock(&channel->lock); 295 } 296 297 TAILQ_INSERT_HEAD(&req->channel->reqs, req, link); 298 299 if (channel->sync) { 300 pthread_spin_unlock(&channel->lock); 301 } 302 } 303 304 static int 305 _spdk_fs_channel_create(struct spdk_filesystem *fs, struct spdk_fs_channel *channel, 306 uint32_t max_ops) 307 { 308 uint32_t i; 309 310 channel->req_mem = calloc(max_ops, sizeof(struct spdk_fs_request)); 311 if (!channel->req_mem) { 312 return -1; 313 } 314 315 TAILQ_INIT(&channel->reqs); 316 sem_init(&channel->sem, 0, 0); 317 318 for (i = 0; i < max_ops; i++) { 319 TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link); 320 } 321 322 channel->fs = fs; 323 324 return 0; 325 } 326 327 static int 328 _spdk_fs_md_channel_create(void *io_device, void *ctx_buf) 329 { 330 struct spdk_filesystem *fs; 331 struct spdk_fs_channel *channel = ctx_buf; 332 333 fs = SPDK_CONTAINEROF(io_device, struct spdk_filesystem, md_target); 334 335 return _spdk_fs_channel_create(fs, channel, fs->md_target.max_ops); 336 } 337 338 static int 339 _spdk_fs_sync_channel_create(void *io_device, void *ctx_buf) 340 { 341 struct spdk_filesystem *fs; 342 struct spdk_fs_channel *channel = ctx_buf; 343 344 fs = SPDK_CONTAINEROF(io_device, struct spdk_filesystem, sync_target); 345 346 return _spdk_fs_channel_create(fs, channel, fs->sync_target.max_ops); 347 } 348 349 static int 350 _spdk_fs_io_channel_create(void *io_device, void *ctx_buf) 351 { 352 struct spdk_filesystem *fs; 353 struct spdk_fs_channel *channel = ctx_buf; 354 355 fs = SPDK_CONTAINEROF(io_device, struct spdk_filesystem, io_target); 356 357 return _spdk_fs_channel_create(fs, channel, fs->io_target.max_ops); 358 } 359 360 static void 361 _spdk_fs_channel_destroy(void *io_device, void *ctx_buf) 362 { 363 struct spdk_fs_channel *channel = ctx_buf; 364 365 free(channel->req_mem); 366 if (channel->bs_channel != NULL) { 367 spdk_bs_free_io_channel(channel->bs_channel); 368 } 369 } 370 371 static void 372 __send_request_direct(fs_request_fn fn, void *arg) 373 { 374 fn(arg); 375 } 376 377 static void 378 common_fs_bs_init(struct spdk_filesystem *fs, struct spdk_blob_store *bs) 379 { 380 fs->bs = bs; 381 fs->bs_opts.cluster_sz = spdk_bs_get_cluster_size(bs); 382 fs->md_target.md_fs_channel->bs_channel = spdk_bs_alloc_io_channel(fs->bs); 383 fs->md_target.md_fs_channel->send_request = __send_request_direct; 384 fs->sync_target.sync_fs_channel->bs_channel = spdk_bs_alloc_io_channel(fs->bs); 385 fs->sync_target.sync_fs_channel->send_request = __send_request_direct; 386 387 pthread_mutex_lock(&g_cache_init_lock); 388 if (g_fs_count == 0) { 389 __initialize_cache(); 390 } 391 g_fs_count++; 392 pthread_mutex_unlock(&g_cache_init_lock); 393 } 394 395 static void 396 init_cb(void *ctx, struct spdk_blob_store *bs, int bserrno) 397 { 398 struct spdk_fs_request *req = ctx; 399 struct spdk_fs_cb_args *args = &req->args; 400 struct spdk_filesystem *fs = args->fs; 401 402 if (bserrno == 0) { 403 common_fs_bs_init(fs, bs); 404 } else { 405 free(fs); 406 fs = NULL; 407 } 408 409 args->fn.fs_op_with_handle(args->arg, fs, bserrno); 410 free_fs_request(req); 411 } 412 413 static void 414 fs_conf_parse(void) 415 { 416 struct spdk_conf_section *sp; 417 418 sp = spdk_conf_find_section(NULL, "Blobfs"); 419 if (sp == NULL) { 420 g_fs_cache_buffer_shift = CACHE_BUFFER_SHIFT_DEFAULT; 421 return; 422 } 423 424 g_fs_cache_buffer_shift = spdk_conf_section_get_intval(sp, "CacheBufferShift"); 425 if (g_fs_cache_buffer_shift <= 0) { 426 g_fs_cache_buffer_shift = CACHE_BUFFER_SHIFT_DEFAULT; 427 } 428 } 429 430 static struct spdk_filesystem * 431 fs_alloc(struct spdk_bs_dev *dev, fs_send_request_fn send_request_fn) 432 { 433 struct spdk_filesystem *fs; 434 435 fs = calloc(1, sizeof(*fs)); 436 if (fs == NULL) { 437 return NULL; 438 } 439 440 fs->bdev = dev; 441 fs->send_request = send_request_fn; 442 TAILQ_INIT(&fs->files); 443 444 fs->md_target.max_ops = 512; 445 spdk_io_device_register(&fs->md_target, _spdk_fs_md_channel_create, _spdk_fs_channel_destroy, 446 sizeof(struct spdk_fs_channel)); 447 fs->md_target.md_io_channel = spdk_get_io_channel(&fs->md_target); 448 fs->md_target.md_fs_channel = spdk_io_channel_get_ctx(fs->md_target.md_io_channel); 449 450 fs->sync_target.max_ops = 512; 451 spdk_io_device_register(&fs->sync_target, _spdk_fs_sync_channel_create, _spdk_fs_channel_destroy, 452 sizeof(struct spdk_fs_channel)); 453 fs->sync_target.sync_io_channel = spdk_get_io_channel(&fs->sync_target); 454 fs->sync_target.sync_fs_channel = spdk_io_channel_get_ctx(fs->sync_target.sync_io_channel); 455 456 fs->io_target.max_ops = 512; 457 spdk_io_device_register(&fs->io_target, _spdk_fs_io_channel_create, _spdk_fs_channel_destroy, 458 sizeof(struct spdk_fs_channel)); 459 460 return fs; 461 } 462 463 void 464 spdk_fs_init(struct spdk_bs_dev *dev, struct spdk_blobfs_opts *opt, 465 fs_send_request_fn send_request_fn, 466 spdk_fs_op_with_handle_complete cb_fn, void *cb_arg) 467 { 468 struct spdk_filesystem *fs; 469 struct spdk_fs_request *req; 470 struct spdk_fs_cb_args *args; 471 struct spdk_bs_opts opts = {}; 472 473 fs = fs_alloc(dev, send_request_fn); 474 if (fs == NULL) { 475 cb_fn(cb_arg, NULL, -ENOMEM); 476 return; 477 } 478 479 fs_conf_parse(); 480 481 req = alloc_fs_request(fs->md_target.md_fs_channel); 482 if (req == NULL) { 483 spdk_put_io_channel(fs->md_target.md_io_channel); 484 spdk_io_device_unregister(&fs->md_target, NULL); 485 spdk_put_io_channel(fs->sync_target.sync_io_channel); 486 spdk_io_device_unregister(&fs->sync_target, NULL); 487 spdk_io_device_unregister(&fs->io_target, NULL); 488 free(fs); 489 cb_fn(cb_arg, NULL, -ENOMEM); 490 return; 491 } 492 493 args = &req->args; 494 args->fn.fs_op_with_handle = cb_fn; 495 args->arg = cb_arg; 496 args->fs = fs; 497 498 spdk_bs_opts_init(&opts); 499 snprintf(opts.bstype.bstype, sizeof(opts.bstype.bstype), "BLOBFS"); 500 if (opt) { 501 opts.cluster_sz = opt->cluster_sz; 502 } 503 spdk_bs_init(dev, &opts, init_cb, req); 504 } 505 506 static struct spdk_file * 507 file_alloc(struct spdk_filesystem *fs) 508 { 509 struct spdk_file *file; 510 511 file = calloc(1, sizeof(*file)); 512 if (file == NULL) { 513 return NULL; 514 } 515 516 file->tree = calloc(1, sizeof(*file->tree)); 517 if (file->tree == NULL) { 518 free(file); 519 return NULL; 520 } 521 522 file->fs = fs; 523 TAILQ_INIT(&file->open_requests); 524 TAILQ_INIT(&file->sync_requests); 525 pthread_spin_init(&file->lock, 0); 526 TAILQ_INSERT_TAIL(&fs->files, file, tailq); 527 file->priority = SPDK_FILE_PRIORITY_LOW; 528 return file; 529 } 530 531 static void fs_load_done(void *ctx, int bserrno); 532 533 static int 534 _handle_deleted_files(struct spdk_fs_request *req) 535 { 536 struct spdk_fs_cb_args *args = &req->args; 537 struct spdk_filesystem *fs = args->fs; 538 539 if (!TAILQ_EMPTY(&args->op.fs_load.deleted_files)) { 540 struct spdk_deleted_file *deleted_file; 541 542 deleted_file = TAILQ_FIRST(&args->op.fs_load.deleted_files); 543 TAILQ_REMOVE(&args->op.fs_load.deleted_files, deleted_file, tailq); 544 spdk_bs_delete_blob(fs->bs, deleted_file->id, fs_load_done, req); 545 free(deleted_file); 546 return 0; 547 } 548 549 return 1; 550 } 551 552 static void 553 fs_load_done(void *ctx, int bserrno) 554 { 555 struct spdk_fs_request *req = ctx; 556 struct spdk_fs_cb_args *args = &req->args; 557 struct spdk_filesystem *fs = args->fs; 558 559 /* The filesystem has been loaded. Now check if there are any files that 560 * were marked for deletion before last unload. Do not complete the 561 * fs_load callback until all of them have been deleted on disk. 562 */ 563 if (_handle_deleted_files(req) == 0) { 564 /* We found a file that's been marked for deleting but not actually 565 * deleted yet. This function will get called again once the delete 566 * operation is completed. 567 */ 568 return; 569 } 570 571 args->fn.fs_op_with_handle(args->arg, fs, 0); 572 free_fs_request(req); 573 574 } 575 576 static void 577 iter_cb(void *ctx, struct spdk_blob *blob, int rc) 578 { 579 struct spdk_fs_request *req = ctx; 580 struct spdk_fs_cb_args *args = &req->args; 581 struct spdk_filesystem *fs = args->fs; 582 uint64_t *length; 583 const char *name; 584 uint32_t *is_deleted; 585 size_t value_len; 586 587 if (rc < 0) { 588 args->fn.fs_op_with_handle(args->arg, fs, rc); 589 free_fs_request(req); 590 return; 591 } 592 593 rc = spdk_blob_get_xattr_value(blob, "name", (const void **)&name, &value_len); 594 if (rc < 0) { 595 args->fn.fs_op_with_handle(args->arg, fs, rc); 596 free_fs_request(req); 597 return; 598 } 599 600 rc = spdk_blob_get_xattr_value(blob, "length", (const void **)&length, &value_len); 601 if (rc < 0) { 602 args->fn.fs_op_with_handle(args->arg, fs, rc); 603 free_fs_request(req); 604 return; 605 } 606 607 assert(value_len == 8); 608 609 /* This file could be deleted last time without close it, then app crashed, so we delete it now */ 610 rc = spdk_blob_get_xattr_value(blob, "is_deleted", (const void **)&is_deleted, &value_len); 611 if (rc < 0) { 612 struct spdk_file *f; 613 614 f = file_alloc(fs); 615 if (f == NULL) { 616 args->fn.fs_op_with_handle(args->arg, fs, -ENOMEM); 617 free_fs_request(req); 618 return; 619 } 620 621 f->name = strdup(name); 622 f->blobid = spdk_blob_get_id(blob); 623 f->length = *length; 624 f->length_flushed = *length; 625 f->append_pos = *length; 626 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "added file %s length=%ju\n", f->name, f->length); 627 } else { 628 struct spdk_deleted_file *deleted_file; 629 630 deleted_file = calloc(1, sizeof(*deleted_file)); 631 if (deleted_file == NULL) { 632 args->fn.fs_op_with_handle(args->arg, fs, -ENOMEM); 633 free_fs_request(req); 634 return; 635 } 636 deleted_file->id = spdk_blob_get_id(blob); 637 TAILQ_INSERT_TAIL(&args->op.fs_load.deleted_files, deleted_file, tailq); 638 } 639 } 640 641 static void 642 load_cb(void *ctx, struct spdk_blob_store *bs, int bserrno) 643 { 644 struct spdk_fs_request *req = ctx; 645 struct spdk_fs_cb_args *args = &req->args; 646 struct spdk_filesystem *fs = args->fs; 647 struct spdk_bs_type bstype; 648 static const struct spdk_bs_type blobfs_type = {"BLOBFS"}; 649 static const struct spdk_bs_type zeros; 650 651 if (bserrno != 0) { 652 args->fn.fs_op_with_handle(args->arg, NULL, bserrno); 653 free_fs_request(req); 654 free(fs); 655 return; 656 } 657 658 bstype = spdk_bs_get_bstype(bs); 659 660 if (!memcmp(&bstype, &zeros, sizeof(bstype))) { 661 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "assigning bstype\n"); 662 spdk_bs_set_bstype(bs, blobfs_type); 663 } else if (memcmp(&bstype, &blobfs_type, sizeof(bstype))) { 664 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "not blobfs\n"); 665 SPDK_TRACEDUMP(SPDK_LOG_BLOB, "bstype", &bstype, sizeof(bstype)); 666 args->fn.fs_op_with_handle(args->arg, NULL, bserrno); 667 free_fs_request(req); 668 free(fs); 669 return; 670 } 671 672 common_fs_bs_init(fs, bs); 673 fs_load_done(req, 0); 674 } 675 676 void 677 spdk_fs_load(struct spdk_bs_dev *dev, fs_send_request_fn send_request_fn, 678 spdk_fs_op_with_handle_complete cb_fn, void *cb_arg) 679 { 680 struct spdk_filesystem *fs; 681 struct spdk_fs_cb_args *args; 682 struct spdk_fs_request *req; 683 struct spdk_bs_opts bs_opts; 684 685 fs = fs_alloc(dev, send_request_fn); 686 if (fs == NULL) { 687 cb_fn(cb_arg, NULL, -ENOMEM); 688 return; 689 } 690 691 fs_conf_parse(); 692 693 req = alloc_fs_request(fs->md_target.md_fs_channel); 694 if (req == NULL) { 695 spdk_put_io_channel(fs->md_target.md_io_channel); 696 spdk_io_device_unregister(&fs->md_target, NULL); 697 spdk_put_io_channel(fs->sync_target.sync_io_channel); 698 spdk_io_device_unregister(&fs->sync_target, NULL); 699 spdk_io_device_unregister(&fs->io_target, NULL); 700 free(fs); 701 cb_fn(cb_arg, NULL, -ENOMEM); 702 return; 703 } 704 705 args = &req->args; 706 args->fn.fs_op_with_handle = cb_fn; 707 args->arg = cb_arg; 708 args->fs = fs; 709 TAILQ_INIT(&args->op.fs_load.deleted_files); 710 spdk_bs_opts_init(&bs_opts); 711 bs_opts.iter_cb_fn = iter_cb; 712 bs_opts.iter_cb_arg = req; 713 spdk_bs_load(dev, &bs_opts, load_cb, req); 714 } 715 716 static void 717 unload_cb(void *ctx, int bserrno) 718 { 719 struct spdk_fs_request *req = ctx; 720 struct spdk_fs_cb_args *args = &req->args; 721 struct spdk_filesystem *fs = args->fs; 722 723 pthread_mutex_lock(&g_cache_init_lock); 724 g_fs_count--; 725 if (g_fs_count == 0) { 726 __free_cache(); 727 } 728 pthread_mutex_unlock(&g_cache_init_lock); 729 730 args->fn.fs_op(args->arg, bserrno); 731 free(req); 732 733 spdk_io_device_unregister(&fs->io_target, NULL); 734 spdk_io_device_unregister(&fs->sync_target, NULL); 735 spdk_io_device_unregister(&fs->md_target, NULL); 736 737 free(fs); 738 } 739 740 void 741 spdk_fs_unload(struct spdk_filesystem *fs, spdk_fs_op_complete cb_fn, void *cb_arg) 742 { 743 struct spdk_fs_request *req; 744 struct spdk_fs_cb_args *args; 745 746 /* 747 * We must free the md_channel before unloading the blobstore, so just 748 * allocate this request from the general heap. 749 */ 750 req = calloc(1, sizeof(*req)); 751 if (req == NULL) { 752 cb_fn(cb_arg, -ENOMEM); 753 return; 754 } 755 756 args = &req->args; 757 args->fn.fs_op = cb_fn; 758 args->arg = cb_arg; 759 args->fs = fs; 760 761 spdk_fs_free_io_channel(fs->md_target.md_io_channel); 762 spdk_fs_free_io_channel(fs->sync_target.sync_io_channel); 763 spdk_bs_unload(fs->bs, unload_cb, req); 764 } 765 766 static struct spdk_file * 767 fs_find_file(struct spdk_filesystem *fs, const char *name) 768 { 769 struct spdk_file *file; 770 771 TAILQ_FOREACH(file, &fs->files, tailq) { 772 if (!strncmp(name, file->name, SPDK_FILE_NAME_MAX)) { 773 return file; 774 } 775 } 776 777 return NULL; 778 } 779 780 void 781 spdk_fs_file_stat_async(struct spdk_filesystem *fs, const char *name, 782 spdk_file_stat_op_complete cb_fn, void *cb_arg) 783 { 784 struct spdk_file_stat stat; 785 struct spdk_file *f = NULL; 786 787 if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) { 788 cb_fn(cb_arg, NULL, -ENAMETOOLONG); 789 return; 790 } 791 792 f = fs_find_file(fs, name); 793 if (f != NULL) { 794 stat.blobid = f->blobid; 795 stat.size = f->append_pos >= f->length ? f->append_pos : f->length; 796 cb_fn(cb_arg, &stat, 0); 797 return; 798 } 799 800 cb_fn(cb_arg, NULL, -ENOENT); 801 } 802 803 static void 804 __copy_stat(void *arg, struct spdk_file_stat *stat, int fserrno) 805 { 806 struct spdk_fs_request *req = arg; 807 struct spdk_fs_cb_args *args = &req->args; 808 809 args->rc = fserrno; 810 if (fserrno == 0) { 811 memcpy(args->arg, stat, sizeof(*stat)); 812 } 813 sem_post(args->sem); 814 } 815 816 static void 817 __file_stat(void *arg) 818 { 819 struct spdk_fs_request *req = arg; 820 struct spdk_fs_cb_args *args = &req->args; 821 822 spdk_fs_file_stat_async(args->fs, args->op.stat.name, 823 args->fn.stat_op, req); 824 } 825 826 int 827 spdk_fs_file_stat(struct spdk_filesystem *fs, struct spdk_io_channel *_channel, 828 const char *name, struct spdk_file_stat *stat) 829 { 830 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 831 struct spdk_fs_request *req; 832 int rc; 833 834 req = alloc_fs_request(channel); 835 assert(req != NULL); 836 837 req->args.fs = fs; 838 req->args.op.stat.name = name; 839 req->args.fn.stat_op = __copy_stat; 840 req->args.arg = stat; 841 req->args.sem = &channel->sem; 842 channel->send_request(__file_stat, req); 843 sem_wait(&channel->sem); 844 845 rc = req->args.rc; 846 free_fs_request(req); 847 848 return rc; 849 } 850 851 static void 852 fs_create_blob_close_cb(void *ctx, int bserrno) 853 { 854 struct spdk_fs_request *req = ctx; 855 struct spdk_fs_cb_args *args = &req->args; 856 857 args->fn.file_op(args->arg, bserrno); 858 free_fs_request(req); 859 } 860 861 static void 862 fs_create_blob_resize_cb(void *ctx, int bserrno) 863 { 864 struct spdk_fs_request *req = ctx; 865 struct spdk_fs_cb_args *args = &req->args; 866 struct spdk_file *f = args->file; 867 struct spdk_blob *blob = args->op.create.blob; 868 uint64_t length = 0; 869 870 spdk_blob_set_xattr(blob, "name", f->name, strlen(f->name) + 1); 871 spdk_blob_set_xattr(blob, "length", &length, sizeof(length)); 872 873 spdk_blob_close(blob, fs_create_blob_close_cb, args); 874 } 875 876 static void 877 fs_create_blob_open_cb(void *ctx, struct spdk_blob *blob, int bserrno) 878 { 879 struct spdk_fs_request *req = ctx; 880 struct spdk_fs_cb_args *args = &req->args; 881 882 args->op.create.blob = blob; 883 spdk_blob_resize(blob, 1, fs_create_blob_resize_cb, req); 884 } 885 886 static void 887 fs_create_blob_create_cb(void *ctx, spdk_blob_id blobid, int bserrno) 888 { 889 struct spdk_fs_request *req = ctx; 890 struct spdk_fs_cb_args *args = &req->args; 891 struct spdk_file *f = args->file; 892 893 f->blobid = blobid; 894 spdk_bs_open_blob(f->fs->bs, blobid, fs_create_blob_open_cb, req); 895 } 896 897 void 898 spdk_fs_create_file_async(struct spdk_filesystem *fs, const char *name, 899 spdk_file_op_complete cb_fn, void *cb_arg) 900 { 901 struct spdk_file *file; 902 struct spdk_fs_request *req; 903 struct spdk_fs_cb_args *args; 904 905 if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) { 906 cb_fn(cb_arg, -ENAMETOOLONG); 907 return; 908 } 909 910 file = fs_find_file(fs, name); 911 if (file != NULL) { 912 cb_fn(cb_arg, -EEXIST); 913 return; 914 } 915 916 file = file_alloc(fs); 917 if (file == NULL) { 918 cb_fn(cb_arg, -ENOMEM); 919 return; 920 } 921 922 req = alloc_fs_request(fs->md_target.md_fs_channel); 923 if (req == NULL) { 924 cb_fn(cb_arg, -ENOMEM); 925 return; 926 } 927 928 args = &req->args; 929 args->file = file; 930 args->fn.file_op = cb_fn; 931 args->arg = cb_arg; 932 933 file->name = strdup(name); 934 spdk_bs_create_blob(fs->bs, fs_create_blob_create_cb, args); 935 } 936 937 static void 938 __fs_create_file_done(void *arg, int fserrno) 939 { 940 struct spdk_fs_request *req = arg; 941 struct spdk_fs_cb_args *args = &req->args; 942 943 args->rc = fserrno; 944 sem_post(args->sem); 945 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", args->op.create.name); 946 } 947 948 static void 949 __fs_create_file(void *arg) 950 { 951 struct spdk_fs_request *req = arg; 952 struct spdk_fs_cb_args *args = &req->args; 953 954 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", args->op.create.name); 955 spdk_fs_create_file_async(args->fs, args->op.create.name, __fs_create_file_done, req); 956 } 957 958 int 959 spdk_fs_create_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel, const char *name) 960 { 961 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 962 struct spdk_fs_request *req; 963 struct spdk_fs_cb_args *args; 964 int rc; 965 966 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", name); 967 968 req = alloc_fs_request(channel); 969 assert(req != NULL); 970 971 args = &req->args; 972 args->fs = fs; 973 args->op.create.name = name; 974 args->sem = &channel->sem; 975 fs->send_request(__fs_create_file, req); 976 sem_wait(&channel->sem); 977 rc = args->rc; 978 free_fs_request(req); 979 980 return rc; 981 } 982 983 static void 984 fs_open_blob_done(void *ctx, struct spdk_blob *blob, int bserrno) 985 { 986 struct spdk_fs_request *req = ctx; 987 struct spdk_fs_cb_args *args = &req->args; 988 struct spdk_file *f = args->file; 989 990 f->blob = blob; 991 while (!TAILQ_EMPTY(&f->open_requests)) { 992 req = TAILQ_FIRST(&f->open_requests); 993 args = &req->args; 994 TAILQ_REMOVE(&f->open_requests, req, args.op.open.tailq); 995 args->fn.file_op_with_handle(args->arg, f, bserrno); 996 free_fs_request(req); 997 } 998 } 999 1000 static void 1001 fs_open_blob_create_cb(void *ctx, int bserrno) 1002 { 1003 struct spdk_fs_request *req = ctx; 1004 struct spdk_fs_cb_args *args = &req->args; 1005 struct spdk_file *file = args->file; 1006 struct spdk_filesystem *fs = args->fs; 1007 1008 if (file == NULL) { 1009 /* 1010 * This is from an open with CREATE flag - the file 1011 * is now created so look it up in the file list for this 1012 * filesystem. 1013 */ 1014 file = fs_find_file(fs, args->op.open.name); 1015 assert(file != NULL); 1016 args->file = file; 1017 } 1018 1019 file->ref_count++; 1020 TAILQ_INSERT_TAIL(&file->open_requests, req, args.op.open.tailq); 1021 if (file->ref_count == 1) { 1022 assert(file->blob == NULL); 1023 spdk_bs_open_blob(fs->bs, file->blobid, fs_open_blob_done, req); 1024 } else if (file->blob != NULL) { 1025 fs_open_blob_done(req, file->blob, 0); 1026 } else { 1027 /* 1028 * The blob open for this file is in progress due to a previous 1029 * open request. When that open completes, it will invoke the 1030 * open callback for this request. 1031 */ 1032 } 1033 } 1034 1035 void 1036 spdk_fs_open_file_async(struct spdk_filesystem *fs, const char *name, uint32_t flags, 1037 spdk_file_op_with_handle_complete cb_fn, void *cb_arg) 1038 { 1039 struct spdk_file *f = NULL; 1040 struct spdk_fs_request *req; 1041 struct spdk_fs_cb_args *args; 1042 1043 if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) { 1044 cb_fn(cb_arg, NULL, -ENAMETOOLONG); 1045 return; 1046 } 1047 1048 f = fs_find_file(fs, name); 1049 if (f == NULL && !(flags & SPDK_BLOBFS_OPEN_CREATE)) { 1050 cb_fn(cb_arg, NULL, -ENOENT); 1051 return; 1052 } 1053 1054 if (f != NULL && f->is_deleted == true) { 1055 cb_fn(cb_arg, NULL, -ENOENT); 1056 return; 1057 } 1058 1059 req = alloc_fs_request(fs->md_target.md_fs_channel); 1060 if (req == NULL) { 1061 cb_fn(cb_arg, NULL, -ENOMEM); 1062 return; 1063 } 1064 1065 args = &req->args; 1066 args->fn.file_op_with_handle = cb_fn; 1067 args->arg = cb_arg; 1068 args->file = f; 1069 args->fs = fs; 1070 args->op.open.name = name; 1071 1072 if (f == NULL) { 1073 spdk_fs_create_file_async(fs, name, fs_open_blob_create_cb, req); 1074 } else { 1075 fs_open_blob_create_cb(req, 0); 1076 } 1077 } 1078 1079 static void 1080 __fs_open_file_done(void *arg, struct spdk_file *file, int bserrno) 1081 { 1082 struct spdk_fs_request *req = arg; 1083 struct spdk_fs_cb_args *args = &req->args; 1084 1085 args->file = file; 1086 args->rc = bserrno; 1087 sem_post(args->sem); 1088 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", args->op.open.name); 1089 } 1090 1091 static void 1092 __fs_open_file(void *arg) 1093 { 1094 struct spdk_fs_request *req = arg; 1095 struct spdk_fs_cb_args *args = &req->args; 1096 1097 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", args->op.open.name); 1098 spdk_fs_open_file_async(args->fs, args->op.open.name, args->op.open.flags, 1099 __fs_open_file_done, req); 1100 } 1101 1102 int 1103 spdk_fs_open_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel, 1104 const char *name, uint32_t flags, struct spdk_file **file) 1105 { 1106 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 1107 struct spdk_fs_request *req; 1108 struct spdk_fs_cb_args *args; 1109 int rc; 1110 1111 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", name); 1112 1113 req = alloc_fs_request(channel); 1114 assert(req != NULL); 1115 1116 args = &req->args; 1117 args->fs = fs; 1118 args->op.open.name = name; 1119 args->op.open.flags = flags; 1120 args->sem = &channel->sem; 1121 fs->send_request(__fs_open_file, req); 1122 sem_wait(&channel->sem); 1123 rc = args->rc; 1124 if (rc == 0) { 1125 *file = args->file; 1126 } else { 1127 *file = NULL; 1128 } 1129 free_fs_request(req); 1130 1131 return rc; 1132 } 1133 1134 static void 1135 fs_rename_blob_close_cb(void *ctx, int bserrno) 1136 { 1137 struct spdk_fs_request *req = ctx; 1138 struct spdk_fs_cb_args *args = &req->args; 1139 1140 args->fn.fs_op(args->arg, bserrno); 1141 free_fs_request(req); 1142 } 1143 1144 static void 1145 fs_rename_blob_open_cb(void *ctx, struct spdk_blob *blob, int bserrno) 1146 { 1147 struct spdk_fs_request *req = ctx; 1148 struct spdk_fs_cb_args *args = &req->args; 1149 const char *new_name = args->op.rename.new_name; 1150 1151 spdk_blob_set_xattr(blob, "name", new_name, strlen(new_name) + 1); 1152 spdk_blob_close(blob, fs_rename_blob_close_cb, req); 1153 } 1154 1155 static void 1156 __spdk_fs_md_rename_file(struct spdk_fs_request *req) 1157 { 1158 struct spdk_fs_cb_args *args = &req->args; 1159 struct spdk_file *f; 1160 1161 f = fs_find_file(args->fs, args->op.rename.old_name); 1162 if (f == NULL) { 1163 args->fn.fs_op(args->arg, -ENOENT); 1164 free_fs_request(req); 1165 return; 1166 } 1167 1168 free(f->name); 1169 f->name = strdup(args->op.rename.new_name); 1170 args->file = f; 1171 spdk_bs_open_blob(args->fs->bs, f->blobid, fs_rename_blob_open_cb, req); 1172 } 1173 1174 static void 1175 fs_rename_delete_done(void *arg, int fserrno) 1176 { 1177 __spdk_fs_md_rename_file(arg); 1178 } 1179 1180 void 1181 spdk_fs_rename_file_async(struct spdk_filesystem *fs, 1182 const char *old_name, const char *new_name, 1183 spdk_file_op_complete cb_fn, void *cb_arg) 1184 { 1185 struct spdk_file *f; 1186 struct spdk_fs_request *req; 1187 struct spdk_fs_cb_args *args; 1188 1189 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "old=%s new=%s\n", old_name, new_name); 1190 if (strnlen(new_name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) { 1191 cb_fn(cb_arg, -ENAMETOOLONG); 1192 return; 1193 } 1194 1195 req = alloc_fs_request(fs->md_target.md_fs_channel); 1196 if (req == NULL) { 1197 cb_fn(cb_arg, -ENOMEM); 1198 return; 1199 } 1200 1201 args = &req->args; 1202 args->fn.fs_op = cb_fn; 1203 args->fs = fs; 1204 args->arg = cb_arg; 1205 args->op.rename.old_name = old_name; 1206 args->op.rename.new_name = new_name; 1207 1208 f = fs_find_file(fs, new_name); 1209 if (f == NULL) { 1210 __spdk_fs_md_rename_file(req); 1211 return; 1212 } 1213 1214 /* 1215 * The rename overwrites an existing file. So delete the existing file, then 1216 * do the actual rename. 1217 */ 1218 spdk_fs_delete_file_async(fs, new_name, fs_rename_delete_done, req); 1219 } 1220 1221 static void 1222 __fs_rename_file_done(void *arg, int fserrno) 1223 { 1224 struct spdk_fs_request *req = arg; 1225 struct spdk_fs_cb_args *args = &req->args; 1226 1227 args->rc = fserrno; 1228 sem_post(args->sem); 1229 } 1230 1231 static void 1232 __fs_rename_file(void *arg) 1233 { 1234 struct spdk_fs_request *req = arg; 1235 struct spdk_fs_cb_args *args = &req->args; 1236 1237 spdk_fs_rename_file_async(args->fs, args->op.rename.old_name, args->op.rename.new_name, 1238 __fs_rename_file_done, req); 1239 } 1240 1241 int 1242 spdk_fs_rename_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel, 1243 const char *old_name, const char *new_name) 1244 { 1245 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 1246 struct spdk_fs_request *req; 1247 struct spdk_fs_cb_args *args; 1248 int rc; 1249 1250 req = alloc_fs_request(channel); 1251 assert(req != NULL); 1252 1253 args = &req->args; 1254 1255 args->fs = fs; 1256 args->op.rename.old_name = old_name; 1257 args->op.rename.new_name = new_name; 1258 args->sem = &channel->sem; 1259 fs->send_request(__fs_rename_file, req); 1260 sem_wait(&channel->sem); 1261 rc = args->rc; 1262 free_fs_request(req); 1263 return rc; 1264 } 1265 1266 static void 1267 blob_delete_cb(void *ctx, int bserrno) 1268 { 1269 struct spdk_fs_request *req = ctx; 1270 struct spdk_fs_cb_args *args = &req->args; 1271 1272 args->fn.file_op(args->arg, bserrno); 1273 free_fs_request(req); 1274 } 1275 1276 void 1277 spdk_fs_delete_file_async(struct spdk_filesystem *fs, const char *name, 1278 spdk_file_op_complete cb_fn, void *cb_arg) 1279 { 1280 struct spdk_file *f; 1281 spdk_blob_id blobid; 1282 struct spdk_fs_request *req; 1283 struct spdk_fs_cb_args *args; 1284 1285 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", name); 1286 1287 if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) { 1288 cb_fn(cb_arg, -ENAMETOOLONG); 1289 return; 1290 } 1291 1292 f = fs_find_file(fs, name); 1293 if (f == NULL) { 1294 cb_fn(cb_arg, -ENOENT); 1295 return; 1296 } 1297 1298 req = alloc_fs_request(fs->md_target.md_fs_channel); 1299 if (req == NULL) { 1300 cb_fn(cb_arg, -ENOMEM); 1301 return; 1302 } 1303 1304 args = &req->args; 1305 args->fn.file_op = cb_fn; 1306 args->arg = cb_arg; 1307 1308 if (f->ref_count > 0) { 1309 /* If the ref > 0, we mark the file as deleted and delete it when we close it. */ 1310 f->is_deleted = true; 1311 spdk_blob_set_xattr(f->blob, "is_deleted", &f->is_deleted, sizeof(bool)); 1312 spdk_blob_sync_md(f->blob, blob_delete_cb, args); 1313 return; 1314 } 1315 1316 TAILQ_REMOVE(&fs->files, f, tailq); 1317 1318 cache_free_buffers(f); 1319 1320 blobid = f->blobid; 1321 1322 free(f->name); 1323 free(f->tree); 1324 free(f); 1325 1326 spdk_bs_delete_blob(fs->bs, blobid, blob_delete_cb, req); 1327 } 1328 1329 static void 1330 __fs_delete_file_done(void *arg, int fserrno) 1331 { 1332 struct spdk_fs_request *req = arg; 1333 struct spdk_fs_cb_args *args = &req->args; 1334 1335 args->rc = fserrno; 1336 sem_post(args->sem); 1337 } 1338 1339 static void 1340 __fs_delete_file(void *arg) 1341 { 1342 struct spdk_fs_request *req = arg; 1343 struct spdk_fs_cb_args *args = &req->args; 1344 1345 spdk_fs_delete_file_async(args->fs, args->op.delete.name, __fs_delete_file_done, req); 1346 } 1347 1348 int 1349 spdk_fs_delete_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel, 1350 const char *name) 1351 { 1352 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 1353 struct spdk_fs_request *req; 1354 struct spdk_fs_cb_args *args; 1355 int rc; 1356 1357 req = alloc_fs_request(channel); 1358 assert(req != NULL); 1359 1360 args = &req->args; 1361 args->fs = fs; 1362 args->op.delete.name = name; 1363 args->sem = &channel->sem; 1364 fs->send_request(__fs_delete_file, req); 1365 sem_wait(&channel->sem); 1366 rc = args->rc; 1367 free_fs_request(req); 1368 1369 return rc; 1370 } 1371 1372 spdk_fs_iter 1373 spdk_fs_iter_first(struct spdk_filesystem *fs) 1374 { 1375 struct spdk_file *f; 1376 1377 f = TAILQ_FIRST(&fs->files); 1378 return f; 1379 } 1380 1381 spdk_fs_iter 1382 spdk_fs_iter_next(spdk_fs_iter iter) 1383 { 1384 struct spdk_file *f = iter; 1385 1386 if (f == NULL) { 1387 return NULL; 1388 } 1389 1390 f = TAILQ_NEXT(f, tailq); 1391 return f; 1392 } 1393 1394 const char * 1395 spdk_file_get_name(struct spdk_file *file) 1396 { 1397 return file->name; 1398 } 1399 1400 uint64_t 1401 spdk_file_get_length(struct spdk_file *file) 1402 { 1403 assert(file != NULL); 1404 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s length=0x%jx\n", file->name, file->length); 1405 return file->length; 1406 } 1407 1408 static void 1409 fs_truncate_complete_cb(void *ctx, int bserrno) 1410 { 1411 struct spdk_fs_request *req = ctx; 1412 struct spdk_fs_cb_args *args = &req->args; 1413 1414 args->fn.file_op(args->arg, bserrno); 1415 free_fs_request(req); 1416 } 1417 1418 static void 1419 fs_truncate_resize_cb(void *ctx, int bserrno) 1420 { 1421 struct spdk_fs_request *req = ctx; 1422 struct spdk_fs_cb_args *args = &req->args; 1423 struct spdk_file *file = args->file; 1424 uint64_t *length = &args->op.truncate.length; 1425 1426 spdk_blob_set_xattr(file->blob, "length", length, sizeof(*length)); 1427 1428 file->length = *length; 1429 if (file->append_pos > file->length) { 1430 file->append_pos = file->length; 1431 } 1432 1433 spdk_blob_sync_md(file->blob, fs_truncate_complete_cb, args); 1434 } 1435 1436 static uint64_t 1437 __bytes_to_clusters(uint64_t length, uint64_t cluster_sz) 1438 { 1439 return (length + cluster_sz - 1) / cluster_sz; 1440 } 1441 1442 void 1443 spdk_file_truncate_async(struct spdk_file *file, uint64_t length, 1444 spdk_file_op_complete cb_fn, void *cb_arg) 1445 { 1446 struct spdk_filesystem *fs; 1447 size_t num_clusters; 1448 struct spdk_fs_request *req; 1449 struct spdk_fs_cb_args *args; 1450 1451 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s old=0x%jx new=0x%jx\n", file->name, file->length, length); 1452 if (length == file->length) { 1453 cb_fn(cb_arg, 0); 1454 return; 1455 } 1456 1457 req = alloc_fs_request(file->fs->md_target.md_fs_channel); 1458 if (req == NULL) { 1459 cb_fn(cb_arg, -ENOMEM); 1460 return; 1461 } 1462 1463 args = &req->args; 1464 args->fn.file_op = cb_fn; 1465 args->arg = cb_arg; 1466 args->file = file; 1467 args->op.truncate.length = length; 1468 fs = file->fs; 1469 1470 num_clusters = __bytes_to_clusters(length, fs->bs_opts.cluster_sz); 1471 1472 spdk_blob_resize(file->blob, num_clusters, fs_truncate_resize_cb, req); 1473 } 1474 1475 static void 1476 __truncate(void *arg) 1477 { 1478 struct spdk_fs_request *req = arg; 1479 struct spdk_fs_cb_args *args = &req->args; 1480 1481 spdk_file_truncate_async(args->file, args->op.truncate.length, 1482 args->fn.file_op, args->arg); 1483 } 1484 1485 void 1486 spdk_file_truncate(struct spdk_file *file, struct spdk_io_channel *_channel, 1487 uint64_t length) 1488 { 1489 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 1490 struct spdk_fs_request *req; 1491 struct spdk_fs_cb_args *args; 1492 1493 req = alloc_fs_request(channel); 1494 assert(req != NULL); 1495 1496 args = &req->args; 1497 1498 args->file = file; 1499 args->op.truncate.length = length; 1500 args->fn.file_op = __sem_post; 1501 args->arg = &channel->sem; 1502 1503 channel->send_request(__truncate, req); 1504 sem_wait(&channel->sem); 1505 free_fs_request(req); 1506 } 1507 1508 static void 1509 __rw_done(void *ctx, int bserrno) 1510 { 1511 struct spdk_fs_request *req = ctx; 1512 struct spdk_fs_cb_args *args = &req->args; 1513 1514 spdk_dma_free(args->op.rw.pin_buf); 1515 args->fn.file_op(args->arg, bserrno); 1516 free_fs_request(req); 1517 } 1518 1519 static void 1520 __read_done(void *ctx, int bserrno) 1521 { 1522 struct spdk_fs_request *req = ctx; 1523 struct spdk_fs_cb_args *args = &req->args; 1524 1525 if (args->op.rw.is_read) { 1526 memcpy(args->op.rw.user_buf, 1527 args->op.rw.pin_buf + (args->op.rw.offset & 0xFFF), 1528 args->op.rw.length); 1529 __rw_done(req, 0); 1530 } else { 1531 memcpy(args->op.rw.pin_buf + (args->op.rw.offset & 0xFFF), 1532 args->op.rw.user_buf, 1533 args->op.rw.length); 1534 spdk_blob_io_write(args->file->blob, args->op.rw.channel, 1535 args->op.rw.pin_buf, 1536 args->op.rw.start_page, args->op.rw.num_pages, 1537 __rw_done, req); 1538 } 1539 } 1540 1541 static void 1542 __do_blob_read(void *ctx, int fserrno) 1543 { 1544 struct spdk_fs_request *req = ctx; 1545 struct spdk_fs_cb_args *args = &req->args; 1546 1547 spdk_blob_io_read(args->file->blob, args->op.rw.channel, 1548 args->op.rw.pin_buf, 1549 args->op.rw.start_page, args->op.rw.num_pages, 1550 __read_done, req); 1551 } 1552 1553 static void 1554 __get_page_parameters(struct spdk_file *file, uint64_t offset, uint64_t length, 1555 uint64_t *start_page, uint32_t *page_size, uint64_t *num_pages) 1556 { 1557 uint64_t end_page; 1558 1559 *page_size = spdk_bs_get_page_size(file->fs->bs); 1560 *start_page = offset / *page_size; 1561 end_page = (offset + length - 1) / *page_size; 1562 *num_pages = (end_page - *start_page + 1); 1563 } 1564 1565 static void 1566 __readwrite(struct spdk_file *file, struct spdk_io_channel *_channel, 1567 void *payload, uint64_t offset, uint64_t length, 1568 spdk_file_op_complete cb_fn, void *cb_arg, int is_read) 1569 { 1570 struct spdk_fs_request *req; 1571 struct spdk_fs_cb_args *args; 1572 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 1573 uint64_t start_page, num_pages, pin_buf_length; 1574 uint32_t page_size; 1575 1576 if (is_read && offset + length > file->length) { 1577 cb_fn(cb_arg, -EINVAL); 1578 return; 1579 } 1580 1581 req = alloc_fs_request(channel); 1582 if (req == NULL) { 1583 cb_fn(cb_arg, -ENOMEM); 1584 return; 1585 } 1586 1587 args = &req->args; 1588 args->fn.file_op = cb_fn; 1589 args->arg = cb_arg; 1590 args->file = file; 1591 args->op.rw.channel = channel->bs_channel; 1592 args->op.rw.user_buf = payload; 1593 args->op.rw.is_read = is_read; 1594 args->op.rw.offset = offset; 1595 args->op.rw.length = length; 1596 1597 __get_page_parameters(file, offset, length, &start_page, &page_size, &num_pages); 1598 pin_buf_length = num_pages * page_size; 1599 args->op.rw.pin_buf = spdk_dma_malloc(pin_buf_length, 4096, NULL); 1600 1601 args->op.rw.start_page = start_page; 1602 args->op.rw.num_pages = num_pages; 1603 1604 if (!is_read && file->length < offset + length) { 1605 spdk_file_truncate_async(file, offset + length, __do_blob_read, req); 1606 } else { 1607 __do_blob_read(req, 0); 1608 } 1609 } 1610 1611 void 1612 spdk_file_write_async(struct spdk_file *file, struct spdk_io_channel *channel, 1613 void *payload, uint64_t offset, uint64_t length, 1614 spdk_file_op_complete cb_fn, void *cb_arg) 1615 { 1616 __readwrite(file, channel, payload, offset, length, cb_fn, cb_arg, 0); 1617 } 1618 1619 void 1620 spdk_file_read_async(struct spdk_file *file, struct spdk_io_channel *channel, 1621 void *payload, uint64_t offset, uint64_t length, 1622 spdk_file_op_complete cb_fn, void *cb_arg) 1623 { 1624 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s offset=%jx length=%jx\n", 1625 file->name, offset, length); 1626 __readwrite(file, channel, payload, offset, length, cb_fn, cb_arg, 1); 1627 } 1628 1629 struct spdk_io_channel * 1630 spdk_fs_alloc_io_channel(struct spdk_filesystem *fs) 1631 { 1632 struct spdk_io_channel *io_channel; 1633 struct spdk_fs_channel *fs_channel; 1634 1635 io_channel = spdk_get_io_channel(&fs->io_target); 1636 fs_channel = spdk_io_channel_get_ctx(io_channel); 1637 fs_channel->bs_channel = spdk_bs_alloc_io_channel(fs->bs); 1638 fs_channel->send_request = __send_request_direct; 1639 1640 return io_channel; 1641 } 1642 1643 struct spdk_io_channel * 1644 spdk_fs_alloc_io_channel_sync(struct spdk_filesystem *fs) 1645 { 1646 struct spdk_io_channel *io_channel; 1647 struct spdk_fs_channel *fs_channel; 1648 1649 io_channel = spdk_get_io_channel(&fs->io_target); 1650 fs_channel = spdk_io_channel_get_ctx(io_channel); 1651 fs_channel->send_request = fs->send_request; 1652 fs_channel->sync = 1; 1653 pthread_spin_init(&fs_channel->lock, 0); 1654 1655 return io_channel; 1656 } 1657 1658 void 1659 spdk_fs_free_io_channel(struct spdk_io_channel *channel) 1660 { 1661 spdk_put_io_channel(channel); 1662 } 1663 1664 void 1665 spdk_fs_set_cache_size(uint64_t size_in_mb) 1666 { 1667 g_fs_cache_size = size_in_mb * 1024 * 1024; 1668 } 1669 1670 uint64_t 1671 spdk_fs_get_cache_size(void) 1672 { 1673 return g_fs_cache_size / (1024 * 1024); 1674 } 1675 1676 static void __file_flush(void *_args); 1677 1678 static void * 1679 alloc_cache_memory_buffer(struct spdk_file *context) 1680 { 1681 struct spdk_file *file; 1682 void *buf; 1683 1684 buf = spdk_mempool_get(g_cache_pool); 1685 if (buf != NULL) { 1686 return buf; 1687 } 1688 1689 pthread_spin_lock(&g_caches_lock); 1690 TAILQ_FOREACH(file, &g_caches, cache_tailq) { 1691 if (!file->open_for_writing && 1692 file->priority == SPDK_FILE_PRIORITY_LOW && 1693 file != context) { 1694 break; 1695 } 1696 } 1697 pthread_spin_unlock(&g_caches_lock); 1698 if (file != NULL) { 1699 cache_free_buffers(file); 1700 buf = spdk_mempool_get(g_cache_pool); 1701 if (buf != NULL) { 1702 return buf; 1703 } 1704 } 1705 1706 pthread_spin_lock(&g_caches_lock); 1707 TAILQ_FOREACH(file, &g_caches, cache_tailq) { 1708 if (!file->open_for_writing && file != context) { 1709 break; 1710 } 1711 } 1712 pthread_spin_unlock(&g_caches_lock); 1713 if (file != NULL) { 1714 cache_free_buffers(file); 1715 buf = spdk_mempool_get(g_cache_pool); 1716 if (buf != NULL) { 1717 return buf; 1718 } 1719 } 1720 1721 pthread_spin_lock(&g_caches_lock); 1722 TAILQ_FOREACH(file, &g_caches, cache_tailq) { 1723 if (file != context) { 1724 break; 1725 } 1726 } 1727 pthread_spin_unlock(&g_caches_lock); 1728 if (file != NULL) { 1729 cache_free_buffers(file); 1730 buf = spdk_mempool_get(g_cache_pool); 1731 if (buf != NULL) { 1732 return buf; 1733 } 1734 } 1735 1736 return NULL; 1737 } 1738 1739 static struct cache_buffer * 1740 cache_insert_buffer(struct spdk_file *file, uint64_t offset) 1741 { 1742 struct cache_buffer *buf; 1743 int count = 0; 1744 1745 buf = calloc(1, sizeof(*buf)); 1746 if (buf == NULL) { 1747 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "calloc failed\n"); 1748 return NULL; 1749 } 1750 1751 buf->buf = alloc_cache_memory_buffer(file); 1752 while (buf->buf == NULL) { 1753 /* 1754 * TODO: alloc_cache_memory_buffer() should eventually free 1755 * some buffers. Need a more sophisticated check here, instead 1756 * of just bailing if 100 tries does not result in getting a 1757 * free buffer. This will involve using the sync channel's 1758 * semaphore to block until a buffer becomes available. 1759 */ 1760 if (count++ == 100) { 1761 SPDK_ERRLOG("could not allocate cache buffer\n"); 1762 assert(false); 1763 free(buf); 1764 return NULL; 1765 } 1766 buf->buf = alloc_cache_memory_buffer(file); 1767 } 1768 1769 buf->buf_size = CACHE_BUFFER_SIZE; 1770 buf->offset = offset; 1771 1772 pthread_spin_lock(&g_caches_lock); 1773 if (file->tree->present_mask == 0) { 1774 TAILQ_INSERT_TAIL(&g_caches, file, cache_tailq); 1775 } 1776 file->tree = spdk_tree_insert_buffer(file->tree, buf); 1777 pthread_spin_unlock(&g_caches_lock); 1778 1779 return buf; 1780 } 1781 1782 static struct cache_buffer * 1783 cache_append_buffer(struct spdk_file *file) 1784 { 1785 struct cache_buffer *last; 1786 1787 assert(file->last == NULL || file->last->bytes_filled == file->last->buf_size); 1788 assert((file->append_pos % CACHE_BUFFER_SIZE) == 0); 1789 1790 last = cache_insert_buffer(file, file->append_pos); 1791 if (last == NULL) { 1792 SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "cache_insert_buffer failed\n"); 1793 return NULL; 1794 } 1795 1796 file->last = last; 1797 1798 return last; 1799 } 1800 1801 static void 1802 __wake_caller(struct spdk_fs_cb_args *args) 1803 { 1804 sem_post(args->sem); 1805 } 1806 1807 static void __check_sync_reqs(struct spdk_file *file); 1808 1809 static void 1810 __file_cache_finish_sync(struct spdk_file *file) 1811 { 1812 struct spdk_fs_request *sync_req; 1813 struct spdk_fs_cb_args *sync_args; 1814 1815 pthread_spin_lock(&file->lock); 1816 sync_req = TAILQ_FIRST(&file->sync_requests); 1817 sync_args = &sync_req->args; 1818 assert(sync_args->op.sync.offset <= file->length_flushed); 1819 BLOBFS_TRACE(file, "sync done offset=%jx\n", sync_args->op.sync.offset); 1820 TAILQ_REMOVE(&file->sync_requests, sync_req, args.op.sync.tailq); 1821 pthread_spin_unlock(&file->lock); 1822 1823 sync_args->fn.file_op(sync_args->arg, 0); 1824 __check_sync_reqs(file); 1825 1826 pthread_spin_lock(&file->lock); 1827 free_fs_request(sync_req); 1828 pthread_spin_unlock(&file->lock); 1829 } 1830 1831 static void 1832 __file_cache_finish_sync_bs_cb(void *ctx, int bserrno) 1833 { 1834 struct spdk_file *file = ctx; 1835 1836 __file_cache_finish_sync(file); 1837 } 1838 1839 static void 1840 __free_args(struct spdk_fs_cb_args *args) 1841 { 1842 struct spdk_fs_request *req; 1843 1844 if (!args->from_request) { 1845 free(args); 1846 } else { 1847 /* Depends on args being at the start of the spdk_fs_request structure. */ 1848 req = (struct spdk_fs_request *)args; 1849 free_fs_request(req); 1850 } 1851 } 1852 1853 static void 1854 __check_sync_reqs(struct spdk_file *file) 1855 { 1856 struct spdk_fs_request *sync_req; 1857 1858 pthread_spin_lock(&file->lock); 1859 1860 TAILQ_FOREACH(sync_req, &file->sync_requests, args.op.sync.tailq) { 1861 if (sync_req->args.op.sync.offset <= file->length_flushed) { 1862 break; 1863 } 1864 } 1865 1866 if (sync_req != NULL && !sync_req->args.op.sync.xattr_in_progress) { 1867 BLOBFS_TRACE(file, "set xattr length 0x%jx\n", file->length_flushed); 1868 sync_req->args.op.sync.xattr_in_progress = true; 1869 spdk_blob_set_xattr(file->blob, "length", &file->length_flushed, 1870 sizeof(file->length_flushed)); 1871 1872 pthread_spin_unlock(&file->lock); 1873 spdk_blob_sync_md(file->blob, __file_cache_finish_sync_bs_cb, file); 1874 } else { 1875 pthread_spin_unlock(&file->lock); 1876 } 1877 } 1878 1879 static void 1880 __file_flush_done(void *arg, int bserrno) 1881 { 1882 struct spdk_fs_cb_args *args = arg; 1883 struct spdk_file *file = args->file; 1884 struct cache_buffer *next = args->op.flush.cache_buffer; 1885 1886 BLOBFS_TRACE(file, "length=%jx\n", args->op.flush.length); 1887 1888 pthread_spin_lock(&file->lock); 1889 next->in_progress = false; 1890 next->bytes_flushed += args->op.flush.length; 1891 file->length_flushed += args->op.flush.length; 1892 if (file->length_flushed > file->length) { 1893 file->length = file->length_flushed; 1894 } 1895 if (next->bytes_flushed == next->buf_size) { 1896 BLOBFS_TRACE(file, "write buffer fully flushed 0x%jx\n", file->length_flushed); 1897 next = spdk_tree_find_buffer(file->tree, file->length_flushed); 1898 } 1899 1900 /* 1901 * Assert that there is no cached data that extends past the end of the underlying 1902 * blob. 1903 */ 1904 assert(next == NULL || next->offset < __file_get_blob_size(file) || 1905 next->bytes_filled == 0); 1906 1907 pthread_spin_unlock(&file->lock); 1908 1909 __check_sync_reqs(file); 1910 1911 __file_flush(args); 1912 } 1913 1914 static void 1915 __file_flush(void *_args) 1916 { 1917 struct spdk_fs_cb_args *args = _args; 1918 struct spdk_file *file = args->file; 1919 struct cache_buffer *next; 1920 uint64_t offset, length, start_page, num_pages; 1921 uint32_t page_size; 1922 1923 pthread_spin_lock(&file->lock); 1924 next = spdk_tree_find_buffer(file->tree, file->length_flushed); 1925 if (next == NULL || next->in_progress) { 1926 /* 1927 * There is either no data to flush, or a flush I/O is already in 1928 * progress. So return immediately - if a flush I/O is in 1929 * progress we will flush more data after that is completed. 1930 */ 1931 __free_args(args); 1932 if (next == NULL) { 1933 /* 1934 * For cases where a file's cache was evicted, and then the 1935 * file was later appended, we will write the data directly 1936 * to disk and bypass cache. So just update length_flushed 1937 * here to reflect that all data was already written to disk. 1938 */ 1939 file->length_flushed = file->append_pos; 1940 } 1941 pthread_spin_unlock(&file->lock); 1942 if (next == NULL) { 1943 /* 1944 * There is no data to flush, but we still need to check for any 1945 * outstanding sync requests to make sure metadata gets updated. 1946 */ 1947 __check_sync_reqs(file); 1948 } 1949 return; 1950 } 1951 1952 offset = next->offset + next->bytes_flushed; 1953 length = next->bytes_filled - next->bytes_flushed; 1954 if (length == 0) { 1955 __free_args(args); 1956 pthread_spin_unlock(&file->lock); 1957 return; 1958 } 1959 args->op.flush.length = length; 1960 args->op.flush.cache_buffer = next; 1961 1962 __get_page_parameters(file, offset, length, &start_page, &page_size, &num_pages); 1963 1964 next->in_progress = true; 1965 BLOBFS_TRACE(file, "offset=%jx length=%jx page start=%jx num=%jx\n", 1966 offset, length, start_page, num_pages); 1967 pthread_spin_unlock(&file->lock); 1968 spdk_blob_io_write(file->blob, file->fs->sync_target.sync_fs_channel->bs_channel, 1969 next->buf + (start_page * page_size) - next->offset, 1970 start_page, num_pages, __file_flush_done, args); 1971 } 1972 1973 static void 1974 __file_extend_done(void *arg, int bserrno) 1975 { 1976 struct spdk_fs_cb_args *args = arg; 1977 1978 __wake_caller(args); 1979 } 1980 1981 static void 1982 __file_extend_resize_cb(void *_args, int bserrno) 1983 { 1984 struct spdk_fs_cb_args *args = _args; 1985 struct spdk_file *file = args->file; 1986 1987 spdk_blob_sync_md(file->blob, __file_extend_done, args); 1988 } 1989 1990 static void 1991 __file_extend_blob(void *_args) 1992 { 1993 struct spdk_fs_cb_args *args = _args; 1994 struct spdk_file *file = args->file; 1995 1996 spdk_blob_resize(file->blob, args->op.resize.num_clusters, __file_extend_resize_cb, args); 1997 } 1998 1999 static void 2000 __rw_from_file_done(void *arg, int bserrno) 2001 { 2002 struct spdk_fs_cb_args *args = arg; 2003 2004 __wake_caller(args); 2005 __free_args(args); 2006 } 2007 2008 static void 2009 __rw_from_file(void *_args) 2010 { 2011 struct spdk_fs_cb_args *args = _args; 2012 struct spdk_file *file = args->file; 2013 2014 if (args->op.rw.is_read) { 2015 spdk_file_read_async(file, file->fs->sync_target.sync_io_channel, args->op.rw.user_buf, 2016 args->op.rw.offset, args->op.rw.length, 2017 __rw_from_file_done, args); 2018 } else { 2019 spdk_file_write_async(file, file->fs->sync_target.sync_io_channel, args->op.rw.user_buf, 2020 args->op.rw.offset, args->op.rw.length, 2021 __rw_from_file_done, args); 2022 } 2023 } 2024 2025 static int 2026 __send_rw_from_file(struct spdk_file *file, sem_t *sem, void *payload, 2027 uint64_t offset, uint64_t length, bool is_read) 2028 { 2029 struct spdk_fs_cb_args *args; 2030 2031 args = calloc(1, sizeof(*args)); 2032 if (args == NULL) { 2033 sem_post(sem); 2034 return -ENOMEM; 2035 } 2036 2037 args->file = file; 2038 args->sem = sem; 2039 args->op.rw.user_buf = payload; 2040 args->op.rw.offset = offset; 2041 args->op.rw.length = length; 2042 args->op.rw.is_read = is_read; 2043 file->fs->send_request(__rw_from_file, args); 2044 return 0; 2045 } 2046 2047 int 2048 spdk_file_write(struct spdk_file *file, struct spdk_io_channel *_channel, 2049 void *payload, uint64_t offset, uint64_t length) 2050 { 2051 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 2052 struct spdk_fs_cb_args *args; 2053 uint64_t rem_length, copy, blob_size, cluster_sz; 2054 uint32_t cache_buffers_filled = 0; 2055 uint8_t *cur_payload; 2056 struct cache_buffer *last; 2057 2058 BLOBFS_TRACE_RW(file, "offset=%jx length=%jx\n", offset, length); 2059 2060 if (length == 0) { 2061 return 0; 2062 } 2063 2064 if (offset != file->append_pos) { 2065 BLOBFS_TRACE(file, " error offset=%jx append_pos=%jx\n", offset, file->append_pos); 2066 return -EINVAL; 2067 } 2068 2069 pthread_spin_lock(&file->lock); 2070 file->open_for_writing = true; 2071 2072 if (file->last == NULL) { 2073 if (file->append_pos % CACHE_BUFFER_SIZE == 0) { 2074 cache_append_buffer(file); 2075 } else { 2076 int rc; 2077 2078 file->append_pos += length; 2079 pthread_spin_unlock(&file->lock); 2080 rc = __send_rw_from_file(file, &channel->sem, payload, 2081 offset, length, false); 2082 sem_wait(&channel->sem); 2083 return rc; 2084 } 2085 } 2086 2087 blob_size = __file_get_blob_size(file); 2088 2089 if ((offset + length) > blob_size) { 2090 struct spdk_fs_cb_args extend_args = {}; 2091 2092 cluster_sz = file->fs->bs_opts.cluster_sz; 2093 extend_args.sem = &channel->sem; 2094 extend_args.op.resize.num_clusters = __bytes_to_clusters((offset + length), cluster_sz); 2095 extend_args.file = file; 2096 BLOBFS_TRACE(file, "start resize to %u clusters\n", extend_args.op.resize.num_clusters); 2097 pthread_spin_unlock(&file->lock); 2098 file->fs->send_request(__file_extend_blob, &extend_args); 2099 sem_wait(&channel->sem); 2100 } 2101 2102 last = file->last; 2103 rem_length = length; 2104 cur_payload = payload; 2105 while (rem_length > 0) { 2106 copy = last->buf_size - last->bytes_filled; 2107 if (copy > rem_length) { 2108 copy = rem_length; 2109 } 2110 BLOBFS_TRACE_RW(file, " fill offset=%jx length=%jx\n", file->append_pos, copy); 2111 memcpy(&last->buf[last->bytes_filled], cur_payload, copy); 2112 file->append_pos += copy; 2113 if (file->length < file->append_pos) { 2114 file->length = file->append_pos; 2115 } 2116 cur_payload += copy; 2117 last->bytes_filled += copy; 2118 rem_length -= copy; 2119 if (last->bytes_filled == last->buf_size) { 2120 cache_buffers_filled++; 2121 last = cache_append_buffer(file); 2122 if (last == NULL) { 2123 BLOBFS_TRACE(file, "nomem\n"); 2124 pthread_spin_unlock(&file->lock); 2125 return -ENOMEM; 2126 } 2127 } 2128 } 2129 2130 pthread_spin_unlock(&file->lock); 2131 2132 if (cache_buffers_filled == 0) { 2133 return 0; 2134 } 2135 2136 args = calloc(1, sizeof(*args)); 2137 if (args == NULL) { 2138 return -ENOMEM; 2139 } 2140 2141 args->file = file; 2142 file->fs->send_request(__file_flush, args); 2143 return 0; 2144 } 2145 2146 static void 2147 __readahead_done(void *arg, int bserrno) 2148 { 2149 struct spdk_fs_cb_args *args = arg; 2150 struct cache_buffer *cache_buffer = args->op.readahead.cache_buffer; 2151 struct spdk_file *file = args->file; 2152 2153 BLOBFS_TRACE(file, "offset=%jx\n", cache_buffer->offset); 2154 2155 pthread_spin_lock(&file->lock); 2156 cache_buffer->bytes_filled = args->op.readahead.length; 2157 cache_buffer->bytes_flushed = args->op.readahead.length; 2158 cache_buffer->in_progress = false; 2159 pthread_spin_unlock(&file->lock); 2160 2161 __free_args(args); 2162 } 2163 2164 static void 2165 __readahead(void *_args) 2166 { 2167 struct spdk_fs_cb_args *args = _args; 2168 struct spdk_file *file = args->file; 2169 uint64_t offset, length, start_page, num_pages; 2170 uint32_t page_size; 2171 2172 offset = args->op.readahead.offset; 2173 length = args->op.readahead.length; 2174 assert(length > 0); 2175 2176 __get_page_parameters(file, offset, length, &start_page, &page_size, &num_pages); 2177 2178 BLOBFS_TRACE(file, "offset=%jx length=%jx page start=%jx num=%jx\n", 2179 offset, length, start_page, num_pages); 2180 spdk_blob_io_read(file->blob, file->fs->sync_target.sync_fs_channel->bs_channel, 2181 args->op.readahead.cache_buffer->buf, 2182 start_page, num_pages, __readahead_done, args); 2183 } 2184 2185 static uint64_t 2186 __next_cache_buffer_offset(uint64_t offset) 2187 { 2188 return (offset + CACHE_BUFFER_SIZE) & ~(CACHE_TREE_LEVEL_MASK(0)); 2189 } 2190 2191 static void 2192 check_readahead(struct spdk_file *file, uint64_t offset) 2193 { 2194 struct spdk_fs_cb_args *args; 2195 2196 offset = __next_cache_buffer_offset(offset); 2197 if (spdk_tree_find_buffer(file->tree, offset) != NULL || file->length <= offset) { 2198 return; 2199 } 2200 2201 args = calloc(1, sizeof(*args)); 2202 if (args == NULL) { 2203 return; 2204 } 2205 2206 BLOBFS_TRACE(file, "offset=%jx\n", offset); 2207 2208 args->file = file; 2209 args->op.readahead.offset = offset; 2210 args->op.readahead.cache_buffer = cache_insert_buffer(file, offset); 2211 args->op.readahead.cache_buffer->in_progress = true; 2212 if (file->length < (offset + CACHE_BUFFER_SIZE)) { 2213 args->op.readahead.length = file->length & (CACHE_BUFFER_SIZE - 1); 2214 } else { 2215 args->op.readahead.length = CACHE_BUFFER_SIZE; 2216 } 2217 file->fs->send_request(__readahead, args); 2218 } 2219 2220 static int 2221 __file_read(struct spdk_file *file, void *payload, uint64_t offset, uint64_t length, sem_t *sem) 2222 { 2223 struct cache_buffer *buf; 2224 int rc; 2225 2226 buf = spdk_tree_find_filled_buffer(file->tree, offset); 2227 if (buf == NULL) { 2228 pthread_spin_unlock(&file->lock); 2229 rc = __send_rw_from_file(file, sem, payload, offset, length, true); 2230 pthread_spin_lock(&file->lock); 2231 return rc; 2232 } 2233 2234 if ((offset + length) > (buf->offset + buf->bytes_filled)) { 2235 length = buf->offset + buf->bytes_filled - offset; 2236 } 2237 BLOBFS_TRACE(file, "read %p offset=%ju length=%ju\n", payload, offset, length); 2238 memcpy(payload, &buf->buf[offset - buf->offset], length); 2239 if ((offset + length) % CACHE_BUFFER_SIZE == 0) { 2240 pthread_spin_lock(&g_caches_lock); 2241 spdk_tree_remove_buffer(file->tree, buf); 2242 if (file->tree->present_mask == 0) { 2243 TAILQ_REMOVE(&g_caches, file, cache_tailq); 2244 } 2245 pthread_spin_unlock(&g_caches_lock); 2246 } 2247 2248 sem_post(sem); 2249 return 0; 2250 } 2251 2252 int64_t 2253 spdk_file_read(struct spdk_file *file, struct spdk_io_channel *_channel, 2254 void *payload, uint64_t offset, uint64_t length) 2255 { 2256 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 2257 uint64_t final_offset, final_length; 2258 uint32_t sub_reads = 0; 2259 int rc = 0; 2260 2261 pthread_spin_lock(&file->lock); 2262 2263 BLOBFS_TRACE_RW(file, "offset=%ju length=%ju\n", offset, length); 2264 2265 file->open_for_writing = false; 2266 2267 if (length == 0 || offset >= file->append_pos) { 2268 pthread_spin_unlock(&file->lock); 2269 return 0; 2270 } 2271 2272 if (offset + length > file->append_pos) { 2273 length = file->append_pos - offset; 2274 } 2275 2276 if (offset != file->next_seq_offset) { 2277 file->seq_byte_count = 0; 2278 } 2279 file->seq_byte_count += length; 2280 file->next_seq_offset = offset + length; 2281 if (file->seq_byte_count >= CACHE_READAHEAD_THRESHOLD) { 2282 check_readahead(file, offset); 2283 check_readahead(file, offset + CACHE_BUFFER_SIZE); 2284 } 2285 2286 final_length = 0; 2287 final_offset = offset + length; 2288 while (offset < final_offset) { 2289 length = NEXT_CACHE_BUFFER_OFFSET(offset) - offset; 2290 if (length > (final_offset - offset)) { 2291 length = final_offset - offset; 2292 } 2293 rc = __file_read(file, payload, offset, length, &channel->sem); 2294 if (rc == 0) { 2295 final_length += length; 2296 } else { 2297 break; 2298 } 2299 payload += length; 2300 offset += length; 2301 sub_reads++; 2302 } 2303 pthread_spin_unlock(&file->lock); 2304 while (sub_reads-- > 0) { 2305 sem_wait(&channel->sem); 2306 } 2307 if (rc == 0) { 2308 return final_length; 2309 } else { 2310 return rc; 2311 } 2312 } 2313 2314 static void 2315 _file_sync(struct spdk_file *file, struct spdk_fs_channel *channel, 2316 spdk_file_op_complete cb_fn, void *cb_arg) 2317 { 2318 struct spdk_fs_request *sync_req; 2319 struct spdk_fs_request *flush_req; 2320 struct spdk_fs_cb_args *sync_args; 2321 struct spdk_fs_cb_args *flush_args; 2322 2323 BLOBFS_TRACE(file, "offset=%jx\n", file->append_pos); 2324 2325 pthread_spin_lock(&file->lock); 2326 if (file->append_pos <= file->length_flushed) { 2327 BLOBFS_TRACE(file, "done - no data to flush\n"); 2328 pthread_spin_unlock(&file->lock); 2329 cb_fn(cb_arg, 0); 2330 return; 2331 } 2332 2333 sync_req = alloc_fs_request(channel); 2334 assert(sync_req != NULL); 2335 sync_args = &sync_req->args; 2336 2337 flush_req = alloc_fs_request(channel); 2338 assert(flush_req != NULL); 2339 flush_args = &flush_req->args; 2340 2341 sync_args->file = file; 2342 sync_args->fn.file_op = cb_fn; 2343 sync_args->arg = cb_arg; 2344 sync_args->op.sync.offset = file->append_pos; 2345 sync_args->op.sync.xattr_in_progress = false; 2346 TAILQ_INSERT_TAIL(&file->sync_requests, sync_req, args.op.sync.tailq); 2347 pthread_spin_unlock(&file->lock); 2348 2349 flush_args->file = file; 2350 channel->send_request(__file_flush, flush_args); 2351 } 2352 2353 int 2354 spdk_file_sync(struct spdk_file *file, struct spdk_io_channel *_channel) 2355 { 2356 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 2357 2358 _file_sync(file, channel, __sem_post, &channel->sem); 2359 sem_wait(&channel->sem); 2360 2361 return 0; 2362 } 2363 2364 void 2365 spdk_file_sync_async(struct spdk_file *file, struct spdk_io_channel *_channel, 2366 spdk_file_op_complete cb_fn, void *cb_arg) 2367 { 2368 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 2369 2370 _file_sync(file, channel, cb_fn, cb_arg); 2371 } 2372 2373 void 2374 spdk_file_set_priority(struct spdk_file *file, uint32_t priority) 2375 { 2376 BLOBFS_TRACE(file, "priority=%u\n", priority); 2377 file->priority = priority; 2378 2379 } 2380 2381 /* 2382 * Close routines 2383 */ 2384 2385 static void 2386 __file_close_async_done(void *ctx, int bserrno) 2387 { 2388 struct spdk_fs_request *req = ctx; 2389 struct spdk_fs_cb_args *args = &req->args; 2390 struct spdk_file *file = args->file; 2391 2392 if (file->is_deleted) { 2393 spdk_fs_delete_file_async(file->fs, file->name, blob_delete_cb, ctx); 2394 return; 2395 } 2396 2397 args->fn.file_op(args->arg, bserrno); 2398 free_fs_request(req); 2399 } 2400 2401 static void 2402 __file_close_async(struct spdk_file *file, struct spdk_fs_request *req) 2403 { 2404 struct spdk_blob *blob; 2405 2406 pthread_spin_lock(&file->lock); 2407 if (file->ref_count == 0) { 2408 pthread_spin_unlock(&file->lock); 2409 __file_close_async_done(req, -EBADF); 2410 return; 2411 } 2412 2413 file->ref_count--; 2414 if (file->ref_count > 0) { 2415 pthread_spin_unlock(&file->lock); 2416 req->args.fn.file_op(req->args.arg, 0); 2417 free_fs_request(req); 2418 return; 2419 } 2420 2421 pthread_spin_unlock(&file->lock); 2422 2423 blob = file->blob; 2424 file->blob = NULL; 2425 spdk_blob_close(blob, __file_close_async_done, req); 2426 } 2427 2428 static void 2429 __file_close_async__sync_done(void *arg, int fserrno) 2430 { 2431 struct spdk_fs_request *req = arg; 2432 struct spdk_fs_cb_args *args = &req->args; 2433 2434 __file_close_async(args->file, req); 2435 } 2436 2437 void 2438 spdk_file_close_async(struct spdk_file *file, spdk_file_op_complete cb_fn, void *cb_arg) 2439 { 2440 struct spdk_fs_request *req; 2441 struct spdk_fs_cb_args *args; 2442 2443 req = alloc_fs_request(file->fs->md_target.md_fs_channel); 2444 if (req == NULL) { 2445 cb_fn(cb_arg, -ENOMEM); 2446 return; 2447 } 2448 2449 args = &req->args; 2450 args->file = file; 2451 args->fn.file_op = cb_fn; 2452 args->arg = cb_arg; 2453 2454 spdk_file_sync_async(file, file->fs->md_target.md_io_channel, __file_close_async__sync_done, req); 2455 } 2456 2457 static void 2458 __file_close_done(void *arg, int fserrno) 2459 { 2460 struct spdk_fs_cb_args *args = arg; 2461 2462 args->rc = fserrno; 2463 sem_post(args->sem); 2464 } 2465 2466 static void 2467 __file_close(void *arg) 2468 { 2469 struct spdk_fs_request *req = arg; 2470 struct spdk_fs_cb_args *args = &req->args; 2471 struct spdk_file *file = args->file; 2472 2473 __file_close_async(file, req); 2474 } 2475 2476 int 2477 spdk_file_close(struct spdk_file *file, struct spdk_io_channel *_channel) 2478 { 2479 struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel); 2480 struct spdk_fs_request *req; 2481 struct spdk_fs_cb_args *args; 2482 2483 req = alloc_fs_request(channel); 2484 assert(req != NULL); 2485 2486 args = &req->args; 2487 2488 spdk_file_sync(file, _channel); 2489 BLOBFS_TRACE(file, "name=%s\n", file->name); 2490 args->file = file; 2491 args->sem = &channel->sem; 2492 args->fn.file_op = __file_close_done; 2493 args->arg = req; 2494 channel->send_request(__file_close, req); 2495 sem_wait(&channel->sem); 2496 2497 return args->rc; 2498 } 2499 2500 static void 2501 cache_free_buffers(struct spdk_file *file) 2502 { 2503 BLOBFS_TRACE(file, "free=%s\n", file->name); 2504 pthread_spin_lock(&file->lock); 2505 pthread_spin_lock(&g_caches_lock); 2506 if (file->tree->present_mask == 0) { 2507 pthread_spin_unlock(&g_caches_lock); 2508 pthread_spin_unlock(&file->lock); 2509 return; 2510 } 2511 spdk_tree_free_buffers(file->tree); 2512 2513 TAILQ_REMOVE(&g_caches, file, cache_tailq); 2514 /* If not freed, put it in the end of the queue */ 2515 if (file->tree->present_mask != 0) { 2516 TAILQ_INSERT_TAIL(&g_caches, file, cache_tailq); 2517 } 2518 file->last = NULL; 2519 pthread_spin_unlock(&g_caches_lock); 2520 pthread_spin_unlock(&file->lock); 2521 } 2522 2523 SPDK_LOG_REGISTER_COMPONENT("blobfs", SPDK_LOG_BLOBFS) 2524 SPDK_LOG_REGISTER_COMPONENT("blobfs_rw", SPDK_LOG_BLOBFS_RW) 2525