xref: /spdk/lib/blobfs/blobfs.c (revision f86f10757912918b8ba7b4b3bfdab1cd4c2d180c)
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 
55 static uint64_t g_fs_cache_size = BLOBFS_CACHE_SIZE;
56 static struct spdk_mempool *g_cache_pool;
57 static TAILQ_HEAD(, spdk_file) g_caches;
58 static int g_fs_count = 0;
59 static pthread_mutex_t g_cache_init_lock = PTHREAD_MUTEX_INITIALIZER;
60 static pthread_spinlock_t g_caches_lock;
61 
62 static void
63 __sem_post(void *arg, int bserrno)
64 {
65 	sem_t *sem = arg;
66 
67 	sem_post(sem);
68 }
69 
70 void
71 spdk_cache_buffer_free(struct cache_buffer *cache_buffer)
72 {
73 	spdk_mempool_put(g_cache_pool, cache_buffer->buf);
74 	free(cache_buffer);
75 }
76 
77 #define CACHE_READAHEAD_THRESHOLD	(128 * 1024)
78 
79 struct spdk_file {
80 	struct spdk_filesystem	*fs;
81 	struct spdk_blob	*blob;
82 	char			*name;
83 	uint64_t		length;
84 	bool                    is_deleted;
85 	bool			open_for_writing;
86 	uint64_t		length_flushed;
87 	uint64_t		append_pos;
88 	uint64_t		seq_byte_count;
89 	uint64_t		next_seq_offset;
90 	uint32_t		priority;
91 	TAILQ_ENTRY(spdk_file)	tailq;
92 	spdk_blob_id		blobid;
93 	uint32_t		ref_count;
94 	pthread_spinlock_t	lock;
95 	struct cache_buffer	*last;
96 	struct cache_tree	*tree;
97 	TAILQ_HEAD(open_requests_head, spdk_fs_request) open_requests;
98 	TAILQ_HEAD(sync_requests_head, spdk_fs_request) sync_requests;
99 	TAILQ_ENTRY(spdk_file)	cache_tailq;
100 };
101 
102 struct spdk_deleted_file {
103 	spdk_blob_id	id;
104 	TAILQ_ENTRY(spdk_deleted_file)	tailq;
105 };
106 
107 struct spdk_filesystem {
108 	struct spdk_blob_store	*bs;
109 	TAILQ_HEAD(, spdk_file)	files;
110 	struct spdk_bs_opts	bs_opts;
111 	struct spdk_bs_dev	*bdev;
112 	fs_send_request_fn	send_request;
113 
114 	struct {
115 		uint32_t		max_ops;
116 		struct spdk_io_channel	*sync_io_channel;
117 		struct spdk_fs_channel	*sync_fs_channel;
118 	} sync_target;
119 
120 	struct {
121 		uint32_t		max_ops;
122 		struct spdk_io_channel	*md_io_channel;
123 		struct spdk_fs_channel	*md_fs_channel;
124 	} md_target;
125 
126 	struct {
127 		uint32_t		max_ops;
128 	} io_target;
129 };
130 
131 struct spdk_fs_cb_args {
132 	union {
133 		spdk_fs_op_with_handle_complete		fs_op_with_handle;
134 		spdk_fs_op_complete			fs_op;
135 		spdk_file_op_with_handle_complete	file_op_with_handle;
136 		spdk_file_op_complete			file_op;
137 		spdk_file_stat_op_complete		stat_op;
138 	} fn;
139 	void *arg;
140 	sem_t *sem;
141 	struct spdk_filesystem *fs;
142 	struct spdk_file *file;
143 	int rc;
144 	bool from_request;
145 	union {
146 		struct {
147 			TAILQ_HEAD(, spdk_deleted_file)	deleted_files;
148 		} fs_load;
149 		struct {
150 			uint64_t	length;
151 		} truncate;
152 		struct {
153 			struct spdk_io_channel	*channel;
154 			void		*user_buf;
155 			void		*pin_buf;
156 			int		is_read;
157 			off_t		offset;
158 			size_t		length;
159 			uint64_t	start_page;
160 			uint64_t	num_pages;
161 			uint32_t	blocklen;
162 		} rw;
163 		struct {
164 			const char	*old_name;
165 			const char	*new_name;
166 		} rename;
167 		struct {
168 			struct cache_buffer	*cache_buffer;
169 			uint64_t		length;
170 		} flush;
171 		struct {
172 			struct cache_buffer	*cache_buffer;
173 			uint64_t		length;
174 			uint64_t		offset;
175 		} readahead;
176 		struct {
177 			uint64_t			offset;
178 			TAILQ_ENTRY(spdk_fs_request)	tailq;
179 			bool				xattr_in_progress;
180 		} sync;
181 		struct {
182 			uint32_t			num_clusters;
183 		} resize;
184 		struct {
185 			const char	*name;
186 			uint32_t	flags;
187 			TAILQ_ENTRY(spdk_fs_request)	tailq;
188 		} open;
189 		struct {
190 			const char	*name;
191 		} create;
192 		struct {
193 			const char	*name;
194 		} delete;
195 		struct {
196 			const char	*name;
197 		} stat;
198 	} op;
199 };
200 
201 static void cache_free_buffers(struct spdk_file *file);
202 
203 static void
204 __initialize_cache(void)
205 {
206 	assert(g_cache_pool == NULL);
207 
208 	g_cache_pool = spdk_mempool_create("spdk_fs_cache",
209 					   g_fs_cache_size / CACHE_BUFFER_SIZE,
210 					   CACHE_BUFFER_SIZE,
211 					   SPDK_MEMPOOL_DEFAULT_CACHE_SIZE,
212 					   SPDK_ENV_SOCKET_ID_ANY);
213 	TAILQ_INIT(&g_caches);
214 	pthread_spin_init(&g_caches_lock, 0);
215 }
216 
217 static void
218 __free_cache(void)
219 {
220 	assert(g_cache_pool != NULL);
221 
222 	spdk_mempool_free(g_cache_pool);
223 	g_cache_pool = NULL;
224 }
225 
226 static uint64_t
227 __file_get_blob_size(struct spdk_file *file)
228 {
229 	uint64_t cluster_sz;
230 
231 	cluster_sz = file->fs->bs_opts.cluster_sz;
232 	return cluster_sz * spdk_blob_get_num_clusters(file->blob);
233 }
234 
235 struct spdk_fs_request {
236 	struct spdk_fs_cb_args		args;
237 	TAILQ_ENTRY(spdk_fs_request)	link;
238 	struct spdk_fs_channel		*channel;
239 };
240 
241 struct spdk_fs_channel {
242 	struct spdk_fs_request		*req_mem;
243 	TAILQ_HEAD(, spdk_fs_request)	reqs;
244 	sem_t				sem;
245 	struct spdk_filesystem		*fs;
246 	struct spdk_io_channel		*bs_channel;
247 	fs_send_request_fn		send_request;
248 	bool				sync;
249 	pthread_spinlock_t		lock;
250 };
251 
252 static struct spdk_fs_request *
253 alloc_fs_request(struct spdk_fs_channel *channel)
254 {
255 	struct spdk_fs_request *req;
256 
257 	if (channel->sync) {
258 		pthread_spin_lock(&channel->lock);
259 	}
260 
261 	req = TAILQ_FIRST(&channel->reqs);
262 	if (req) {
263 		TAILQ_REMOVE(&channel->reqs, req, link);
264 	}
265 
266 	if (channel->sync) {
267 		pthread_spin_unlock(&channel->lock);
268 	}
269 
270 	if (req == NULL) {
271 		return NULL;
272 	}
273 	memset(req, 0, sizeof(*req));
274 	req->channel = channel;
275 	req->args.from_request = true;
276 
277 	return req;
278 }
279 
280 static void
281 free_fs_request(struct spdk_fs_request *req)
282 {
283 	struct spdk_fs_channel *channel = req->channel;
284 
285 	if (channel->sync) {
286 		pthread_spin_lock(&channel->lock);
287 	}
288 
289 	TAILQ_INSERT_HEAD(&req->channel->reqs, req, link);
290 
291 	if (channel->sync) {
292 		pthread_spin_unlock(&channel->lock);
293 	}
294 }
295 
296 static int
297 _spdk_fs_channel_create(struct spdk_filesystem *fs, struct spdk_fs_channel *channel,
298 			uint32_t max_ops)
299 {
300 	uint32_t i;
301 
302 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_fs_request));
303 	if (!channel->req_mem) {
304 		return -1;
305 	}
306 
307 	TAILQ_INIT(&channel->reqs);
308 	sem_init(&channel->sem, 0, 0);
309 
310 	for (i = 0; i < max_ops; i++) {
311 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
312 	}
313 
314 	channel->fs = fs;
315 
316 	return 0;
317 }
318 
319 static int
320 _spdk_fs_md_channel_create(void *io_device, void *ctx_buf)
321 {
322 	struct spdk_filesystem		*fs;
323 	struct spdk_fs_channel		*channel = ctx_buf;
324 
325 	fs = SPDK_CONTAINEROF(io_device, struct spdk_filesystem, md_target);
326 
327 	return _spdk_fs_channel_create(fs, channel, fs->md_target.max_ops);
328 }
329 
330 static int
331 _spdk_fs_sync_channel_create(void *io_device, void *ctx_buf)
332 {
333 	struct spdk_filesystem		*fs;
334 	struct spdk_fs_channel		*channel = ctx_buf;
335 
336 	fs = SPDK_CONTAINEROF(io_device, struct spdk_filesystem, sync_target);
337 
338 	return _spdk_fs_channel_create(fs, channel, fs->sync_target.max_ops);
339 }
340 
341 static int
342 _spdk_fs_io_channel_create(void *io_device, void *ctx_buf)
343 {
344 	struct spdk_filesystem		*fs;
345 	struct spdk_fs_channel		*channel = ctx_buf;
346 
347 	fs = SPDK_CONTAINEROF(io_device, struct spdk_filesystem, io_target);
348 
349 	return _spdk_fs_channel_create(fs, channel, fs->io_target.max_ops);
350 }
351 
352 static void
353 _spdk_fs_channel_destroy(void *io_device, void *ctx_buf)
354 {
355 	struct spdk_fs_channel *channel = ctx_buf;
356 
357 	free(channel->req_mem);
358 	if (channel->bs_channel != NULL) {
359 		spdk_bs_free_io_channel(channel->bs_channel);
360 	}
361 }
362 
363 static void
364 __send_request_direct(fs_request_fn fn, void *arg)
365 {
366 	fn(arg);
367 }
368 
369 static void
370 common_fs_bs_init(struct spdk_filesystem *fs, struct spdk_blob_store *bs)
371 {
372 	fs->bs = bs;
373 	fs->bs_opts.cluster_sz = spdk_bs_get_cluster_size(bs);
374 	fs->md_target.md_fs_channel->bs_channel = spdk_bs_alloc_io_channel(fs->bs);
375 	fs->md_target.md_fs_channel->send_request = __send_request_direct;
376 	fs->sync_target.sync_fs_channel->bs_channel = spdk_bs_alloc_io_channel(fs->bs);
377 	fs->sync_target.sync_fs_channel->send_request = __send_request_direct;
378 
379 	pthread_mutex_lock(&g_cache_init_lock);
380 	if (g_fs_count == 0) {
381 		__initialize_cache();
382 	}
383 	g_fs_count++;
384 	pthread_mutex_unlock(&g_cache_init_lock);
385 }
386 
387 static void
388 init_cb(void *ctx, struct spdk_blob_store *bs, int bserrno)
389 {
390 	struct spdk_fs_request *req = ctx;
391 	struct spdk_fs_cb_args *args = &req->args;
392 	struct spdk_filesystem *fs = args->fs;
393 
394 	if (bserrno == 0) {
395 		common_fs_bs_init(fs, bs);
396 	} else {
397 		free(fs);
398 		fs = NULL;
399 	}
400 
401 	args->fn.fs_op_with_handle(args->arg, fs, bserrno);
402 	free_fs_request(req);
403 }
404 
405 static void
406 fs_conf_parse(void)
407 {
408 	struct spdk_conf_section *sp;
409 
410 	sp = spdk_conf_find_section(NULL, "Blobfs");
411 	if (sp == NULL) {
412 		g_fs_cache_buffer_shift = CACHE_BUFFER_SHIFT_DEFAULT;
413 		return;
414 	}
415 
416 	g_fs_cache_buffer_shift = spdk_conf_section_get_intval(sp, "CacheBufferShift");
417 	if (g_fs_cache_buffer_shift <= 0) {
418 		g_fs_cache_buffer_shift = CACHE_BUFFER_SHIFT_DEFAULT;
419 	}
420 }
421 
422 static struct spdk_filesystem *
423 fs_alloc(struct spdk_bs_dev *dev, fs_send_request_fn send_request_fn)
424 {
425 	struct spdk_filesystem *fs;
426 
427 	fs = calloc(1, sizeof(*fs));
428 	if (fs == NULL) {
429 		return NULL;
430 	}
431 
432 	fs->bdev = dev;
433 	fs->send_request = send_request_fn;
434 	TAILQ_INIT(&fs->files);
435 
436 	fs->md_target.max_ops = 512;
437 	spdk_io_device_register(&fs->md_target, _spdk_fs_md_channel_create, _spdk_fs_channel_destroy,
438 				sizeof(struct spdk_fs_channel));
439 	fs->md_target.md_io_channel = spdk_get_io_channel(&fs->md_target);
440 	fs->md_target.md_fs_channel = spdk_io_channel_get_ctx(fs->md_target.md_io_channel);
441 
442 	fs->sync_target.max_ops = 512;
443 	spdk_io_device_register(&fs->sync_target, _spdk_fs_sync_channel_create, _spdk_fs_channel_destroy,
444 				sizeof(struct spdk_fs_channel));
445 	fs->sync_target.sync_io_channel = spdk_get_io_channel(&fs->sync_target);
446 	fs->sync_target.sync_fs_channel = spdk_io_channel_get_ctx(fs->sync_target.sync_io_channel);
447 
448 	fs->io_target.max_ops = 512;
449 	spdk_io_device_register(&fs->io_target, _spdk_fs_io_channel_create, _spdk_fs_channel_destroy,
450 				sizeof(struct spdk_fs_channel));
451 
452 	return fs;
453 }
454 
455 void
456 spdk_fs_init(struct spdk_bs_dev *dev, fs_send_request_fn send_request_fn,
457 	     spdk_fs_op_with_handle_complete cb_fn, void *cb_arg)
458 {
459 	struct spdk_filesystem *fs;
460 	struct spdk_fs_request *req;
461 	struct spdk_fs_cb_args *args;
462 	struct spdk_bs_opts opts = {};
463 
464 	fs = fs_alloc(dev, send_request_fn);
465 	if (fs == NULL) {
466 		cb_fn(cb_arg, NULL, -ENOMEM);
467 		return;
468 	}
469 
470 	fs_conf_parse();
471 
472 	req = alloc_fs_request(fs->md_target.md_fs_channel);
473 	if (req == NULL) {
474 		spdk_put_io_channel(fs->md_target.md_io_channel);
475 		spdk_io_device_unregister(&fs->md_target, NULL);
476 		spdk_put_io_channel(fs->sync_target.sync_io_channel);
477 		spdk_io_device_unregister(&fs->sync_target, NULL);
478 		spdk_io_device_unregister(&fs->io_target, NULL);
479 		free(fs);
480 		cb_fn(cb_arg, NULL, -ENOMEM);
481 		return;
482 	}
483 
484 	args = &req->args;
485 	args->fn.fs_op_with_handle = cb_fn;
486 	args->arg = cb_arg;
487 	args->fs = fs;
488 
489 	spdk_bs_opts_init(&opts);
490 	strncpy(opts.bstype.bstype, "BLOBFS", SPDK_BLOBSTORE_TYPE_LENGTH);
491 
492 	spdk_bs_init(dev, &opts, init_cb, req);
493 }
494 
495 static struct spdk_file *
496 file_alloc(struct spdk_filesystem *fs)
497 {
498 	struct spdk_file *file;
499 
500 	file = calloc(1, sizeof(*file));
501 	if (file == NULL) {
502 		return NULL;
503 	}
504 
505 	file->tree = calloc(1, sizeof(*file->tree));
506 	if (file->tree == NULL) {
507 		free(file);
508 		return NULL;
509 	}
510 
511 	file->fs = fs;
512 	TAILQ_INIT(&file->open_requests);
513 	TAILQ_INIT(&file->sync_requests);
514 	pthread_spin_init(&file->lock, 0);
515 	TAILQ_INSERT_TAIL(&fs->files, file, tailq);
516 	file->priority = SPDK_FILE_PRIORITY_LOW;
517 	return file;
518 }
519 
520 static void iter_delete_cb(void *ctx, int bserrno);
521 
522 static int
523 _handle_deleted_files(struct spdk_fs_request *req)
524 {
525 	struct spdk_fs_cb_args *args = &req->args;
526 	struct spdk_filesystem *fs = args->fs;
527 
528 	if (!TAILQ_EMPTY(&args->op.fs_load.deleted_files)) {
529 		struct spdk_deleted_file *deleted_file;
530 
531 		deleted_file = TAILQ_FIRST(&args->op.fs_load.deleted_files);
532 		TAILQ_REMOVE(&args->op.fs_load.deleted_files, deleted_file, tailq);
533 		spdk_bs_delete_blob(fs->bs, deleted_file->id, iter_delete_cb, req);
534 		free(deleted_file);
535 		return 0;
536 	}
537 
538 	return 1;
539 }
540 
541 static void
542 iter_delete_cb(void *ctx, int bserrno)
543 {
544 	struct spdk_fs_request *req = ctx;
545 	struct spdk_fs_cb_args *args = &req->args;
546 	struct spdk_filesystem *fs = args->fs;
547 
548 	if (_handle_deleted_files(req) == 0) {
549 		return;
550 	}
551 
552 	args->fn.fs_op_with_handle(args->arg, fs, 0);
553 	free_fs_request(req);
554 
555 }
556 
557 static void
558 iter_cb(void *ctx, struct spdk_blob *blob, int rc)
559 {
560 	struct spdk_fs_request *req = ctx;
561 	struct spdk_fs_cb_args *args = &req->args;
562 	struct spdk_filesystem *fs = args->fs;
563 	uint64_t *length;
564 	const char *name;
565 	uint32_t *is_deleted;
566 	size_t value_len;
567 
568 	if (rc == -ENOENT) {
569 		/* Finished iterating */
570 		if (_handle_deleted_files(req) == 0) {
571 			return;
572 		}
573 		args->fn.fs_op_with_handle(args->arg, fs, 0);
574 		free_fs_request(req);
575 		return;
576 	} else if (rc < 0) {
577 		args->fn.fs_op_with_handle(args->arg, fs, rc);
578 		free_fs_request(req);
579 		return;
580 	}
581 
582 	rc = spdk_blob_get_xattr_value(blob, "name", (const void **)&name, &value_len);
583 	if (rc < 0) {
584 		args->fn.fs_op_with_handle(args->arg, fs, rc);
585 		free_fs_request(req);
586 		return;
587 	}
588 
589 	rc = spdk_blob_get_xattr_value(blob, "length", (const void **)&length, &value_len);
590 	if (rc < 0) {
591 		args->fn.fs_op_with_handle(args->arg, fs, rc);
592 		free_fs_request(req);
593 		return;
594 	}
595 
596 	assert(value_len == 8);
597 
598 	/* This file could be deleted last time without close it, then app crashed, so we delete it now */
599 	rc = spdk_blob_get_xattr_value(blob, "is_deleted", (const void **)&is_deleted, &value_len);
600 	if (rc < 0) {
601 		struct spdk_file *f;
602 
603 		f = file_alloc(fs);
604 		if (f == NULL) {
605 			args->fn.fs_op_with_handle(args->arg, fs, -ENOMEM);
606 			free_fs_request(req);
607 			return;
608 		}
609 
610 		f->name = strdup(name);
611 		f->blobid = spdk_blob_get_id(blob);
612 		f->length = *length;
613 		f->length_flushed = *length;
614 		f->append_pos = *length;
615 		SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "added file %s length=%ju\n", f->name, f->length);
616 	} else {
617 		struct spdk_deleted_file *deleted_file;
618 
619 		deleted_file = calloc(1, sizeof(*deleted_file));
620 		if (deleted_file == NULL) {
621 			args->fn.fs_op_with_handle(args->arg, fs, -ENOMEM);
622 			free_fs_request(req);
623 			return;
624 		}
625 		deleted_file->id = spdk_blob_get_id(blob);
626 		TAILQ_INSERT_TAIL(&args->op.fs_load.deleted_files, deleted_file, tailq);
627 	}
628 
629 	spdk_bs_iter_next(fs->bs, blob, iter_cb, req);
630 }
631 
632 static void
633 load_cb(void *ctx, struct spdk_blob_store *bs, int bserrno)
634 {
635 	struct spdk_fs_request *req = ctx;
636 	struct spdk_fs_cb_args *args = &req->args;
637 	struct spdk_filesystem *fs = args->fs;
638 	struct spdk_bs_type bstype;
639 	static const struct spdk_bs_type blobfs_type = {"BLOBFS"};
640 	static const struct spdk_bs_type zeros;
641 
642 	if (bserrno != 0) {
643 		args->fn.fs_op_with_handle(args->arg, NULL, bserrno);
644 		free_fs_request(req);
645 		free(fs);
646 		return;
647 	}
648 
649 	bstype = spdk_bs_get_bstype(bs);
650 
651 	if (!memcmp(&bstype, &zeros, sizeof(bstype))) {
652 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "assigning bstype\n");
653 		spdk_bs_set_bstype(bs, blobfs_type);
654 	} else if (memcmp(&bstype, &blobfs_type, sizeof(bstype))) {
655 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "not blobfs\n");
656 		SPDK_TRACEDUMP(SPDK_LOG_BLOB, "bstype", &bstype, sizeof(bstype));
657 		args->fn.fs_op_with_handle(args->arg, NULL, bserrno);
658 		free_fs_request(req);
659 		free(fs);
660 		return;
661 	}
662 
663 	common_fs_bs_init(fs, bs);
664 	spdk_bs_iter_first(fs->bs, iter_cb, req);
665 }
666 
667 void
668 spdk_fs_load(struct spdk_bs_dev *dev, fs_send_request_fn send_request_fn,
669 	     spdk_fs_op_with_handle_complete cb_fn, void *cb_arg)
670 {
671 	struct spdk_filesystem *fs;
672 	struct spdk_fs_cb_args *args;
673 	struct spdk_fs_request *req;
674 	struct spdk_bs_opts opts = {};
675 
676 	fs = fs_alloc(dev, send_request_fn);
677 	if (fs == NULL) {
678 		cb_fn(cb_arg, NULL, -ENOMEM);
679 		return;
680 	}
681 
682 	fs_conf_parse();
683 
684 	req = alloc_fs_request(fs->md_target.md_fs_channel);
685 	if (req == NULL) {
686 		spdk_put_io_channel(fs->md_target.md_io_channel);
687 		spdk_io_device_unregister(&fs->md_target, NULL);
688 		spdk_put_io_channel(fs->sync_target.sync_io_channel);
689 		spdk_io_device_unregister(&fs->sync_target, NULL);
690 		spdk_io_device_unregister(&fs->io_target, NULL);
691 		free(fs);
692 		cb_fn(cb_arg, NULL, -ENOMEM);
693 		return;
694 	}
695 
696 	args = &req->args;
697 	args->fn.fs_op_with_handle = cb_fn;
698 	args->arg = cb_arg;
699 	args->fs = fs;
700 	TAILQ_INIT(&args->op.fs_load.deleted_files);
701 
702 	spdk_bs_opts_init(&opts);
703 
704 	spdk_bs_load(dev, &opts, load_cb, req);
705 }
706 
707 static void
708 unload_cb(void *ctx, int bserrno)
709 {
710 	struct spdk_fs_request *req = ctx;
711 	struct spdk_fs_cb_args *args = &req->args;
712 	struct spdk_filesystem *fs = args->fs;
713 
714 	pthread_mutex_lock(&g_cache_init_lock);
715 	g_fs_count--;
716 	if (g_fs_count == 0) {
717 		__free_cache();
718 	}
719 	pthread_mutex_unlock(&g_cache_init_lock);
720 
721 	args->fn.fs_op(args->arg, bserrno);
722 	free(req);
723 
724 	spdk_io_device_unregister(&fs->io_target, NULL);
725 	spdk_io_device_unregister(&fs->sync_target, NULL);
726 	spdk_io_device_unregister(&fs->md_target, NULL);
727 
728 	free(fs);
729 }
730 
731 void
732 spdk_fs_unload(struct spdk_filesystem *fs, spdk_fs_op_complete cb_fn, void *cb_arg)
733 {
734 	struct spdk_fs_request *req;
735 	struct spdk_fs_cb_args *args;
736 
737 	/*
738 	 * We must free the md_channel before unloading the blobstore, so just
739 	 *  allocate this request from the general heap.
740 	 */
741 	req = calloc(1, sizeof(*req));
742 	if (req == NULL) {
743 		cb_fn(cb_arg, -ENOMEM);
744 		return;
745 	}
746 
747 	args = &req->args;
748 	args->fn.fs_op = cb_fn;
749 	args->arg = cb_arg;
750 	args->fs = fs;
751 
752 	spdk_fs_free_io_channel(fs->md_target.md_io_channel);
753 	spdk_fs_free_io_channel(fs->sync_target.sync_io_channel);
754 	spdk_bs_unload(fs->bs, unload_cb, req);
755 }
756 
757 static struct spdk_file *
758 fs_find_file(struct spdk_filesystem *fs, const char *name)
759 {
760 	struct spdk_file *file;
761 
762 	TAILQ_FOREACH(file, &fs->files, tailq) {
763 		if (!strncmp(name, file->name, SPDK_FILE_NAME_MAX)) {
764 			return file;
765 		}
766 	}
767 
768 	return NULL;
769 }
770 
771 void
772 spdk_fs_file_stat_async(struct spdk_filesystem *fs, const char *name,
773 			spdk_file_stat_op_complete cb_fn, void *cb_arg)
774 {
775 	struct spdk_file_stat stat;
776 	struct spdk_file *f = NULL;
777 
778 	if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
779 		cb_fn(cb_arg, NULL, -ENAMETOOLONG);
780 		return;
781 	}
782 
783 	f = fs_find_file(fs, name);
784 	if (f != NULL) {
785 		stat.blobid = f->blobid;
786 		stat.size = f->length;
787 		cb_fn(cb_arg, &stat, 0);
788 		return;
789 	}
790 
791 	cb_fn(cb_arg, NULL, -ENOENT);
792 }
793 
794 static void
795 __copy_stat(void *arg, struct spdk_file_stat *stat, int fserrno)
796 {
797 	struct spdk_fs_request *req = arg;
798 	struct spdk_fs_cb_args *args = &req->args;
799 
800 	args->rc = fserrno;
801 	if (fserrno == 0) {
802 		memcpy(args->arg, stat, sizeof(*stat));
803 	}
804 	sem_post(args->sem);
805 }
806 
807 static void
808 __file_stat(void *arg)
809 {
810 	struct spdk_fs_request *req = arg;
811 	struct spdk_fs_cb_args *args = &req->args;
812 
813 	spdk_fs_file_stat_async(args->fs, args->op.stat.name,
814 				args->fn.stat_op, req);
815 }
816 
817 int
818 spdk_fs_file_stat(struct spdk_filesystem *fs, struct spdk_io_channel *_channel,
819 		  const char *name, struct spdk_file_stat *stat)
820 {
821 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
822 	struct spdk_fs_request *req;
823 	int rc;
824 
825 	req = alloc_fs_request(channel);
826 	assert(req != NULL);
827 
828 	req->args.fs = fs;
829 	req->args.op.stat.name = name;
830 	req->args.fn.stat_op = __copy_stat;
831 	req->args.arg = stat;
832 	req->args.sem = &channel->sem;
833 	channel->send_request(__file_stat, req);
834 	sem_wait(&channel->sem);
835 
836 	rc = req->args.rc;
837 	free_fs_request(req);
838 
839 	return rc;
840 }
841 
842 static void
843 fs_create_blob_close_cb(void *ctx, int bserrno)
844 {
845 	struct spdk_fs_request *req = ctx;
846 	struct spdk_fs_cb_args *args = &req->args;
847 
848 	args->fn.file_op(args->arg, bserrno);
849 	free_fs_request(req);
850 }
851 
852 static void
853 fs_create_blob_open_cb(void *ctx, struct spdk_blob *blob, int bserrno)
854 {
855 	struct spdk_fs_request *req = ctx;
856 	struct spdk_fs_cb_args *args = &req->args;
857 	struct spdk_file *f = args->file;
858 	uint64_t length = 0;
859 
860 	spdk_blob_resize(blob, 1);
861 	spdk_blob_set_xattr(blob, "name", f->name, strlen(f->name) + 1);
862 	spdk_blob_set_xattr(blob, "length", &length, sizeof(length));
863 
864 	spdk_blob_close(blob, fs_create_blob_close_cb, args);
865 }
866 
867 static void
868 fs_create_blob_create_cb(void *ctx, spdk_blob_id blobid, int bserrno)
869 {
870 	struct spdk_fs_request *req = ctx;
871 	struct spdk_fs_cb_args *args = &req->args;
872 	struct spdk_file *f = args->file;
873 
874 	f->blobid = blobid;
875 	spdk_bs_open_blob(f->fs->bs, blobid, fs_create_blob_open_cb, req);
876 }
877 
878 void
879 spdk_fs_create_file_async(struct spdk_filesystem *fs, const char *name,
880 			  spdk_file_op_complete cb_fn, void *cb_arg)
881 {
882 	struct spdk_file *file;
883 	struct spdk_fs_request *req;
884 	struct spdk_fs_cb_args *args;
885 
886 	if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
887 		cb_fn(cb_arg, -ENAMETOOLONG);
888 		return;
889 	}
890 
891 	file = fs_find_file(fs, name);
892 	if (file != NULL) {
893 		cb_fn(cb_arg, -EEXIST);
894 		return;
895 	}
896 
897 	file = file_alloc(fs);
898 	if (file == NULL) {
899 		cb_fn(cb_arg, -ENOMEM);
900 		return;
901 	}
902 
903 	req = alloc_fs_request(fs->md_target.md_fs_channel);
904 	if (req == NULL) {
905 		cb_fn(cb_arg, -ENOMEM);
906 		return;
907 	}
908 
909 	args = &req->args;
910 	args->file = file;
911 	args->fn.file_op = cb_fn;
912 	args->arg = cb_arg;
913 
914 	file->name = strdup(name);
915 	spdk_bs_create_blob(fs->bs, fs_create_blob_create_cb, args);
916 }
917 
918 static void
919 __fs_create_file_done(void *arg, int fserrno)
920 {
921 	struct spdk_fs_request *req = arg;
922 	struct spdk_fs_cb_args *args = &req->args;
923 
924 	args->rc = fserrno;
925 	sem_post(args->sem);
926 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", args->op.create.name);
927 }
928 
929 static void
930 __fs_create_file(void *arg)
931 {
932 	struct spdk_fs_request *req = arg;
933 	struct spdk_fs_cb_args *args = &req->args;
934 
935 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", args->op.create.name);
936 	spdk_fs_create_file_async(args->fs, args->op.create.name, __fs_create_file_done, req);
937 }
938 
939 int
940 spdk_fs_create_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel, const char *name)
941 {
942 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
943 	struct spdk_fs_request *req;
944 	struct spdk_fs_cb_args *args;
945 	int rc;
946 
947 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", name);
948 
949 	req = alloc_fs_request(channel);
950 	assert(req != NULL);
951 
952 	args = &req->args;
953 	args->fs = fs;
954 	args->op.create.name = name;
955 	args->sem = &channel->sem;
956 	fs->send_request(__fs_create_file, req);
957 	sem_wait(&channel->sem);
958 	rc = args->rc;
959 	free_fs_request(req);
960 
961 	return rc;
962 }
963 
964 static void
965 fs_open_blob_done(void *ctx, struct spdk_blob *blob, int bserrno)
966 {
967 	struct spdk_fs_request *req = ctx;
968 	struct spdk_fs_cb_args *args = &req->args;
969 	struct spdk_file *f = args->file;
970 
971 	f->blob = blob;
972 	while (!TAILQ_EMPTY(&f->open_requests)) {
973 		req = TAILQ_FIRST(&f->open_requests);
974 		args = &req->args;
975 		TAILQ_REMOVE(&f->open_requests, req, args.op.open.tailq);
976 		args->fn.file_op_with_handle(args->arg, f, bserrno);
977 		free_fs_request(req);
978 	}
979 }
980 
981 static void
982 fs_open_blob_create_cb(void *ctx, int bserrno)
983 {
984 	struct spdk_fs_request *req = ctx;
985 	struct spdk_fs_cb_args *args = &req->args;
986 	struct spdk_file *file = args->file;
987 	struct spdk_filesystem *fs = args->fs;
988 
989 	if (file == NULL) {
990 		/*
991 		 * This is from an open with CREATE flag - the file
992 		 *  is now created so look it up in the file list for this
993 		 *  filesystem.
994 		 */
995 		file = fs_find_file(fs, args->op.open.name);
996 		assert(file != NULL);
997 		args->file = file;
998 	}
999 
1000 	file->ref_count++;
1001 	TAILQ_INSERT_TAIL(&file->open_requests, req, args.op.open.tailq);
1002 	if (file->ref_count == 1) {
1003 		assert(file->blob == NULL);
1004 		spdk_bs_open_blob(fs->bs, file->blobid, fs_open_blob_done, req);
1005 	} else if (file->blob != NULL) {
1006 		fs_open_blob_done(req, file->blob, 0);
1007 	} else {
1008 		/*
1009 		 * The blob open for this file is in progress due to a previous
1010 		 *  open request.  When that open completes, it will invoke the
1011 		 *  open callback for this request.
1012 		 */
1013 	}
1014 }
1015 
1016 void
1017 spdk_fs_open_file_async(struct spdk_filesystem *fs, const char *name, uint32_t flags,
1018 			spdk_file_op_with_handle_complete cb_fn, void *cb_arg)
1019 {
1020 	struct spdk_file *f = NULL;
1021 	struct spdk_fs_request *req;
1022 	struct spdk_fs_cb_args *args;
1023 
1024 	if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
1025 		cb_fn(cb_arg, NULL, -ENAMETOOLONG);
1026 		return;
1027 	}
1028 
1029 	f = fs_find_file(fs, name);
1030 	if (f == NULL && !(flags & SPDK_BLOBFS_OPEN_CREATE)) {
1031 		cb_fn(cb_arg, NULL, -ENOENT);
1032 		return;
1033 	}
1034 
1035 	if (f != NULL && f->is_deleted == true) {
1036 		cb_fn(cb_arg, NULL, -ENOENT);
1037 		return;
1038 	}
1039 
1040 	req = alloc_fs_request(fs->md_target.md_fs_channel);
1041 	if (req == NULL) {
1042 		cb_fn(cb_arg, NULL, -ENOMEM);
1043 		return;
1044 	}
1045 
1046 	args = &req->args;
1047 	args->fn.file_op_with_handle = cb_fn;
1048 	args->arg = cb_arg;
1049 	args->file = f;
1050 	args->fs = fs;
1051 	args->op.open.name = name;
1052 
1053 	if (f == NULL) {
1054 		spdk_fs_create_file_async(fs, name, fs_open_blob_create_cb, req);
1055 	} else {
1056 		fs_open_blob_create_cb(req, 0);
1057 	}
1058 }
1059 
1060 static void
1061 __fs_open_file_done(void *arg, struct spdk_file *file, int bserrno)
1062 {
1063 	struct spdk_fs_request *req = arg;
1064 	struct spdk_fs_cb_args *args = &req->args;
1065 
1066 	args->file = file;
1067 	args->rc = bserrno;
1068 	sem_post(args->sem);
1069 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", args->op.open.name);
1070 }
1071 
1072 static void
1073 __fs_open_file(void *arg)
1074 {
1075 	struct spdk_fs_request *req = arg;
1076 	struct spdk_fs_cb_args *args = &req->args;
1077 
1078 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", args->op.open.name);
1079 	spdk_fs_open_file_async(args->fs, args->op.open.name, args->op.open.flags,
1080 				__fs_open_file_done, req);
1081 }
1082 
1083 int
1084 spdk_fs_open_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel,
1085 		  const char *name, uint32_t flags, struct spdk_file **file)
1086 {
1087 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1088 	struct spdk_fs_request *req;
1089 	struct spdk_fs_cb_args *args;
1090 	int rc;
1091 
1092 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", name);
1093 
1094 	req = alloc_fs_request(channel);
1095 	assert(req != NULL);
1096 
1097 	args = &req->args;
1098 	args->fs = fs;
1099 	args->op.open.name = name;
1100 	args->op.open.flags = flags;
1101 	args->sem = &channel->sem;
1102 	fs->send_request(__fs_open_file, req);
1103 	sem_wait(&channel->sem);
1104 	rc = args->rc;
1105 	if (rc == 0) {
1106 		*file = args->file;
1107 	} else {
1108 		*file = NULL;
1109 	}
1110 	free_fs_request(req);
1111 
1112 	return rc;
1113 }
1114 
1115 static void
1116 fs_rename_blob_close_cb(void *ctx, int bserrno)
1117 {
1118 	struct spdk_fs_request *req = ctx;
1119 	struct spdk_fs_cb_args *args = &req->args;
1120 
1121 	args->fn.fs_op(args->arg, bserrno);
1122 	free_fs_request(req);
1123 }
1124 
1125 static void
1126 fs_rename_blob_open_cb(void *ctx, struct spdk_blob *blob, int bserrno)
1127 {
1128 	struct spdk_fs_request *req = ctx;
1129 	struct spdk_fs_cb_args *args = &req->args;
1130 	const char *new_name = args->op.rename.new_name;
1131 
1132 	spdk_blob_set_xattr(blob, "name", new_name, strlen(new_name) + 1);
1133 	spdk_blob_close(blob, fs_rename_blob_close_cb, req);
1134 }
1135 
1136 static void
1137 __spdk_fs_md_rename_file(struct spdk_fs_request *req)
1138 {
1139 	struct spdk_fs_cb_args *args = &req->args;
1140 	struct spdk_file *f;
1141 
1142 	f = fs_find_file(args->fs, args->op.rename.old_name);
1143 	if (f == NULL) {
1144 		args->fn.fs_op(args->arg, -ENOENT);
1145 		free_fs_request(req);
1146 		return;
1147 	}
1148 
1149 	free(f->name);
1150 	f->name = strdup(args->op.rename.new_name);
1151 	args->file = f;
1152 	spdk_bs_open_blob(args->fs->bs, f->blobid, fs_rename_blob_open_cb, req);
1153 }
1154 
1155 static void
1156 fs_rename_delete_done(void *arg, int fserrno)
1157 {
1158 	__spdk_fs_md_rename_file(arg);
1159 }
1160 
1161 void
1162 spdk_fs_rename_file_async(struct spdk_filesystem *fs,
1163 			  const char *old_name, const char *new_name,
1164 			  spdk_file_op_complete cb_fn, void *cb_arg)
1165 {
1166 	struct spdk_file *f;
1167 	struct spdk_fs_request *req;
1168 	struct spdk_fs_cb_args *args;
1169 
1170 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "old=%s new=%s\n", old_name, new_name);
1171 	if (strnlen(new_name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
1172 		cb_fn(cb_arg, -ENAMETOOLONG);
1173 		return;
1174 	}
1175 
1176 	req = alloc_fs_request(fs->md_target.md_fs_channel);
1177 	if (req == NULL) {
1178 		cb_fn(cb_arg, -ENOMEM);
1179 		return;
1180 	}
1181 
1182 	args = &req->args;
1183 	args->fn.fs_op = cb_fn;
1184 	args->fs = fs;
1185 	args->arg = cb_arg;
1186 	args->op.rename.old_name = old_name;
1187 	args->op.rename.new_name = new_name;
1188 
1189 	f = fs_find_file(fs, new_name);
1190 	if (f == NULL) {
1191 		__spdk_fs_md_rename_file(req);
1192 		return;
1193 	}
1194 
1195 	/*
1196 	 * The rename overwrites an existing file.  So delete the existing file, then
1197 	 *  do the actual rename.
1198 	 */
1199 	spdk_fs_delete_file_async(fs, new_name, fs_rename_delete_done, req);
1200 }
1201 
1202 static void
1203 __fs_rename_file_done(void *arg, int fserrno)
1204 {
1205 	struct spdk_fs_request *req = arg;
1206 	struct spdk_fs_cb_args *args = &req->args;
1207 
1208 	args->rc = fserrno;
1209 	sem_post(args->sem);
1210 }
1211 
1212 static void
1213 __fs_rename_file(void *arg)
1214 {
1215 	struct spdk_fs_request *req = arg;
1216 	struct spdk_fs_cb_args *args = &req->args;
1217 
1218 	spdk_fs_rename_file_async(args->fs, args->op.rename.old_name, args->op.rename.new_name,
1219 				  __fs_rename_file_done, req);
1220 }
1221 
1222 int
1223 spdk_fs_rename_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel,
1224 		    const char *old_name, const char *new_name)
1225 {
1226 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1227 	struct spdk_fs_request *req;
1228 	struct spdk_fs_cb_args *args;
1229 	int rc;
1230 
1231 	req = alloc_fs_request(channel);
1232 	assert(req != NULL);
1233 
1234 	args = &req->args;
1235 
1236 	args->fs = fs;
1237 	args->op.rename.old_name = old_name;
1238 	args->op.rename.new_name = new_name;
1239 	args->sem = &channel->sem;
1240 	fs->send_request(__fs_rename_file, req);
1241 	sem_wait(&channel->sem);
1242 	rc = args->rc;
1243 	free_fs_request(req);
1244 	return rc;
1245 }
1246 
1247 static void
1248 blob_delete_cb(void *ctx, int bserrno)
1249 {
1250 	struct spdk_fs_request *req = ctx;
1251 	struct spdk_fs_cb_args *args = &req->args;
1252 
1253 	args->fn.file_op(args->arg, bserrno);
1254 	free_fs_request(req);
1255 }
1256 
1257 void
1258 spdk_fs_delete_file_async(struct spdk_filesystem *fs, const char *name,
1259 			  spdk_file_op_complete cb_fn, void *cb_arg)
1260 {
1261 	struct spdk_file *f;
1262 	spdk_blob_id blobid;
1263 	struct spdk_fs_request *req;
1264 	struct spdk_fs_cb_args *args;
1265 
1266 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s\n", name);
1267 
1268 	if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
1269 		cb_fn(cb_arg, -ENAMETOOLONG);
1270 		return;
1271 	}
1272 
1273 	f = fs_find_file(fs, name);
1274 	if (f == NULL) {
1275 		cb_fn(cb_arg, -ENOENT);
1276 		return;
1277 	}
1278 
1279 	req = alloc_fs_request(fs->md_target.md_fs_channel);
1280 	if (req == NULL) {
1281 		cb_fn(cb_arg, -ENOMEM);
1282 		return;
1283 	}
1284 
1285 	args = &req->args;
1286 	args->fn.file_op = cb_fn;
1287 	args->arg = cb_arg;
1288 
1289 	if (f->ref_count > 0) {
1290 		/* If the ref > 0, we mark the file as deleted and delete it when we close it. */
1291 		f->is_deleted = true;
1292 		spdk_blob_set_xattr(f->blob, "is_deleted", &f->is_deleted, sizeof(bool));
1293 		spdk_blob_sync_md(f->blob, blob_delete_cb, args);
1294 		return;
1295 	}
1296 
1297 	TAILQ_REMOVE(&fs->files, f, tailq);
1298 
1299 	cache_free_buffers(f);
1300 
1301 	blobid = f->blobid;
1302 
1303 	free(f->name);
1304 	free(f->tree);
1305 	free(f);
1306 
1307 	spdk_bs_delete_blob(fs->bs, blobid, blob_delete_cb, req);
1308 }
1309 
1310 static void
1311 __fs_delete_file_done(void *arg, int fserrno)
1312 {
1313 	struct spdk_fs_request *req = arg;
1314 	struct spdk_fs_cb_args *args = &req->args;
1315 
1316 	args->rc = fserrno;
1317 	sem_post(args->sem);
1318 }
1319 
1320 static void
1321 __fs_delete_file(void *arg)
1322 {
1323 	struct spdk_fs_request *req = arg;
1324 	struct spdk_fs_cb_args *args = &req->args;
1325 
1326 	spdk_fs_delete_file_async(args->fs, args->op.delete.name, __fs_delete_file_done, req);
1327 }
1328 
1329 int
1330 spdk_fs_delete_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel,
1331 		    const char *name)
1332 {
1333 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1334 	struct spdk_fs_request *req;
1335 	struct spdk_fs_cb_args *args;
1336 	int rc;
1337 
1338 	req = alloc_fs_request(channel);
1339 	assert(req != NULL);
1340 
1341 	args = &req->args;
1342 	args->fs = fs;
1343 	args->op.delete.name = name;
1344 	args->sem = &channel->sem;
1345 	fs->send_request(__fs_delete_file, req);
1346 	sem_wait(&channel->sem);
1347 	rc = args->rc;
1348 	free_fs_request(req);
1349 
1350 	return rc;
1351 }
1352 
1353 spdk_fs_iter
1354 spdk_fs_iter_first(struct spdk_filesystem *fs)
1355 {
1356 	struct spdk_file *f;
1357 
1358 	f = TAILQ_FIRST(&fs->files);
1359 	return f;
1360 }
1361 
1362 spdk_fs_iter
1363 spdk_fs_iter_next(spdk_fs_iter iter)
1364 {
1365 	struct spdk_file *f = iter;
1366 
1367 	if (f == NULL) {
1368 		return NULL;
1369 	}
1370 
1371 	f = TAILQ_NEXT(f, tailq);
1372 	return f;
1373 }
1374 
1375 const char *
1376 spdk_file_get_name(struct spdk_file *file)
1377 {
1378 	return file->name;
1379 }
1380 
1381 uint64_t
1382 spdk_file_get_length(struct spdk_file *file)
1383 {
1384 	assert(file != NULL);
1385 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s length=0x%jx\n", file->name, file->length);
1386 	return file->length;
1387 }
1388 
1389 static void
1390 fs_truncate_complete_cb(void *ctx, int bserrno)
1391 {
1392 	struct spdk_fs_request *req = ctx;
1393 	struct spdk_fs_cb_args *args = &req->args;
1394 
1395 	args->fn.file_op(args->arg, bserrno);
1396 	free_fs_request(req);
1397 }
1398 
1399 static uint64_t
1400 __bytes_to_clusters(uint64_t length, uint64_t cluster_sz)
1401 {
1402 	return (length + cluster_sz - 1) / cluster_sz;
1403 }
1404 
1405 void
1406 spdk_file_truncate_async(struct spdk_file *file, uint64_t length,
1407 			 spdk_file_op_complete cb_fn, void *cb_arg)
1408 {
1409 	struct spdk_filesystem *fs;
1410 	size_t num_clusters;
1411 	struct spdk_fs_request *req;
1412 	struct spdk_fs_cb_args *args;
1413 
1414 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s old=0x%jx new=0x%jx\n", file->name, file->length, length);
1415 	if (length == file->length) {
1416 		cb_fn(cb_arg, 0);
1417 		return;
1418 	}
1419 
1420 	req = alloc_fs_request(file->fs->md_target.md_fs_channel);
1421 	if (req == NULL) {
1422 		cb_fn(cb_arg, -ENOMEM);
1423 		return;
1424 	}
1425 
1426 	args = &req->args;
1427 	args->fn.file_op = cb_fn;
1428 	args->arg = cb_arg;
1429 	args->file = file;
1430 	fs = file->fs;
1431 
1432 	num_clusters = __bytes_to_clusters(length, fs->bs_opts.cluster_sz);
1433 
1434 	spdk_blob_resize(file->blob, num_clusters);
1435 	spdk_blob_set_xattr(file->blob, "length", &length, sizeof(length));
1436 
1437 	file->length = length;
1438 	if (file->append_pos > file->length) {
1439 		file->append_pos = file->length;
1440 	}
1441 
1442 	spdk_blob_sync_md(file->blob, fs_truncate_complete_cb, args);
1443 }
1444 
1445 static void
1446 __truncate(void *arg)
1447 {
1448 	struct spdk_fs_request *req = arg;
1449 	struct spdk_fs_cb_args *args = &req->args;
1450 
1451 	spdk_file_truncate_async(args->file, args->op.truncate.length,
1452 				 args->fn.file_op, args->arg);
1453 }
1454 
1455 void
1456 spdk_file_truncate(struct spdk_file *file, struct spdk_io_channel *_channel,
1457 		   uint64_t length)
1458 {
1459 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1460 	struct spdk_fs_request *req;
1461 	struct spdk_fs_cb_args *args;
1462 
1463 	req = alloc_fs_request(channel);
1464 	assert(req != NULL);
1465 
1466 	args = &req->args;
1467 
1468 	args->file = file;
1469 	args->op.truncate.length = length;
1470 	args->fn.file_op = __sem_post;
1471 	args->arg = &channel->sem;
1472 
1473 	channel->send_request(__truncate, req);
1474 	sem_wait(&channel->sem);
1475 	free_fs_request(req);
1476 }
1477 
1478 static void
1479 __rw_done(void *ctx, int bserrno)
1480 {
1481 	struct spdk_fs_request *req = ctx;
1482 	struct spdk_fs_cb_args *args = &req->args;
1483 
1484 	spdk_dma_free(args->op.rw.pin_buf);
1485 	args->fn.file_op(args->arg, bserrno);
1486 	free_fs_request(req);
1487 }
1488 
1489 static void
1490 __read_done(void *ctx, int bserrno)
1491 {
1492 	struct spdk_fs_request *req = ctx;
1493 	struct spdk_fs_cb_args *args = &req->args;
1494 
1495 	if (args->op.rw.is_read) {
1496 		memcpy(args->op.rw.user_buf,
1497 		       args->op.rw.pin_buf + (args->op.rw.offset & 0xFFF),
1498 		       args->op.rw.length);
1499 		__rw_done(req, 0);
1500 	} else {
1501 		memcpy(args->op.rw.pin_buf + (args->op.rw.offset & 0xFFF),
1502 		       args->op.rw.user_buf,
1503 		       args->op.rw.length);
1504 		spdk_bs_io_write_blob(args->file->blob, args->op.rw.channel,
1505 				      args->op.rw.pin_buf,
1506 				      args->op.rw.start_page, args->op.rw.num_pages,
1507 				      __rw_done, req);
1508 	}
1509 }
1510 
1511 static void
1512 __do_blob_read(void *ctx, int fserrno)
1513 {
1514 	struct spdk_fs_request *req = ctx;
1515 	struct spdk_fs_cb_args *args = &req->args;
1516 
1517 	spdk_bs_io_read_blob(args->file->blob, args->op.rw.channel,
1518 			     args->op.rw.pin_buf,
1519 			     args->op.rw.start_page, args->op.rw.num_pages,
1520 			     __read_done, req);
1521 }
1522 
1523 static void
1524 __get_page_parameters(struct spdk_file *file, uint64_t offset, uint64_t length,
1525 		      uint64_t *start_page, uint32_t *page_size, uint64_t *num_pages)
1526 {
1527 	uint64_t end_page;
1528 
1529 	*page_size = spdk_bs_get_page_size(file->fs->bs);
1530 	*start_page = offset / *page_size;
1531 	end_page = (offset + length - 1) / *page_size;
1532 	*num_pages = (end_page - *start_page + 1);
1533 }
1534 
1535 static void
1536 __readwrite(struct spdk_file *file, struct spdk_io_channel *_channel,
1537 	    void *payload, uint64_t offset, uint64_t length,
1538 	    spdk_file_op_complete cb_fn, void *cb_arg, int is_read)
1539 {
1540 	struct spdk_fs_request *req;
1541 	struct spdk_fs_cb_args *args;
1542 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1543 	uint64_t start_page, num_pages, pin_buf_length;
1544 	uint32_t page_size;
1545 
1546 	if (is_read && offset + length > file->length) {
1547 		cb_fn(cb_arg, -EINVAL);
1548 		return;
1549 	}
1550 
1551 	req = alloc_fs_request(channel);
1552 	if (req == NULL) {
1553 		cb_fn(cb_arg, -ENOMEM);
1554 		return;
1555 	}
1556 
1557 	args = &req->args;
1558 	args->fn.file_op = cb_fn;
1559 	args->arg = cb_arg;
1560 	args->file = file;
1561 	args->op.rw.channel = channel->bs_channel;
1562 	args->op.rw.user_buf = payload;
1563 	args->op.rw.is_read = is_read;
1564 	args->op.rw.offset = offset;
1565 	args->op.rw.length = length;
1566 
1567 	__get_page_parameters(file, offset, length, &start_page, &page_size, &num_pages);
1568 	pin_buf_length = num_pages * page_size;
1569 	args->op.rw.pin_buf = spdk_dma_malloc(pin_buf_length, 4096, NULL);
1570 
1571 	args->op.rw.start_page = start_page;
1572 	args->op.rw.num_pages = num_pages;
1573 
1574 	if (!is_read && file->length < offset + length) {
1575 		spdk_file_truncate_async(file, offset + length, __do_blob_read, req);
1576 	} else {
1577 		__do_blob_read(req, 0);
1578 	}
1579 }
1580 
1581 void
1582 spdk_file_write_async(struct spdk_file *file, struct spdk_io_channel *channel,
1583 		      void *payload, uint64_t offset, uint64_t length,
1584 		      spdk_file_op_complete cb_fn, void *cb_arg)
1585 {
1586 	__readwrite(file, channel, payload, offset, length, cb_fn, cb_arg, 0);
1587 }
1588 
1589 void
1590 spdk_file_read_async(struct spdk_file *file, struct spdk_io_channel *channel,
1591 		     void *payload, uint64_t offset, uint64_t length,
1592 		     spdk_file_op_complete cb_fn, void *cb_arg)
1593 {
1594 	SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "file=%s offset=%jx length=%jx\n",
1595 		      file->name, offset, length);
1596 	__readwrite(file, channel, payload, offset, length, cb_fn, cb_arg, 1);
1597 }
1598 
1599 struct spdk_io_channel *
1600 spdk_fs_alloc_io_channel(struct spdk_filesystem *fs)
1601 {
1602 	struct spdk_io_channel *io_channel;
1603 	struct spdk_fs_channel *fs_channel;
1604 
1605 	io_channel = spdk_get_io_channel(&fs->io_target);
1606 	fs_channel = spdk_io_channel_get_ctx(io_channel);
1607 	fs_channel->bs_channel = spdk_bs_alloc_io_channel(fs->bs);
1608 	fs_channel->send_request = __send_request_direct;
1609 
1610 	return io_channel;
1611 }
1612 
1613 struct spdk_io_channel *
1614 spdk_fs_alloc_io_channel_sync(struct spdk_filesystem *fs)
1615 {
1616 	struct spdk_io_channel *io_channel;
1617 	struct spdk_fs_channel *fs_channel;
1618 
1619 	io_channel = spdk_get_io_channel(&fs->io_target);
1620 	fs_channel = spdk_io_channel_get_ctx(io_channel);
1621 	fs_channel->send_request = fs->send_request;
1622 	fs_channel->sync = 1;
1623 	pthread_spin_init(&fs_channel->lock, 0);
1624 
1625 	return io_channel;
1626 }
1627 
1628 void
1629 spdk_fs_free_io_channel(struct spdk_io_channel *channel)
1630 {
1631 	spdk_put_io_channel(channel);
1632 }
1633 
1634 void
1635 spdk_fs_set_cache_size(uint64_t size_in_mb)
1636 {
1637 	g_fs_cache_size = size_in_mb * 1024 * 1024;
1638 }
1639 
1640 uint64_t
1641 spdk_fs_get_cache_size(void)
1642 {
1643 	return g_fs_cache_size / (1024 * 1024);
1644 }
1645 
1646 static void __file_flush(void *_args);
1647 
1648 static void *
1649 alloc_cache_memory_buffer(struct spdk_file *context)
1650 {
1651 	struct spdk_file *file;
1652 	void *buf;
1653 
1654 	buf = spdk_mempool_get(g_cache_pool);
1655 	if (buf != NULL) {
1656 		return buf;
1657 	}
1658 
1659 	pthread_spin_lock(&g_caches_lock);
1660 	TAILQ_FOREACH(file, &g_caches, cache_tailq) {
1661 		if (!file->open_for_writing &&
1662 		    file->priority == SPDK_FILE_PRIORITY_LOW &&
1663 		    file != context) {
1664 			break;
1665 		}
1666 	}
1667 	pthread_spin_unlock(&g_caches_lock);
1668 	if (file != NULL) {
1669 		cache_free_buffers(file);
1670 		buf = spdk_mempool_get(g_cache_pool);
1671 		if (buf != NULL) {
1672 			return buf;
1673 		}
1674 	}
1675 
1676 	pthread_spin_lock(&g_caches_lock);
1677 	TAILQ_FOREACH(file, &g_caches, cache_tailq) {
1678 		if (!file->open_for_writing && file != context) {
1679 			break;
1680 		}
1681 	}
1682 	pthread_spin_unlock(&g_caches_lock);
1683 	if (file != NULL) {
1684 		cache_free_buffers(file);
1685 		buf = spdk_mempool_get(g_cache_pool);
1686 		if (buf != NULL) {
1687 			return buf;
1688 		}
1689 	}
1690 
1691 	pthread_spin_lock(&g_caches_lock);
1692 	TAILQ_FOREACH(file, &g_caches, cache_tailq) {
1693 		if (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 	return NULL;
1707 }
1708 
1709 static struct cache_buffer *
1710 cache_insert_buffer(struct spdk_file *file, uint64_t offset)
1711 {
1712 	struct cache_buffer *buf;
1713 	int count = 0;
1714 
1715 	buf = calloc(1, sizeof(*buf));
1716 	if (buf == NULL) {
1717 		SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "calloc failed\n");
1718 		return NULL;
1719 	}
1720 
1721 	buf->buf = alloc_cache_memory_buffer(file);
1722 	while (buf->buf == NULL) {
1723 		/*
1724 		 * TODO: alloc_cache_memory_buffer() should eventually free
1725 		 *  some buffers.  Need a more sophisticated check here, instead
1726 		 *  of just bailing if 100 tries does not result in getting a
1727 		 *  free buffer.  This will involve using the sync channel's
1728 		 *  semaphore to block until a buffer becomes available.
1729 		 */
1730 		if (count++ == 100) {
1731 			SPDK_ERRLOG("could not allocate cache buffer\n");
1732 			assert(false);
1733 			free(buf);
1734 			return NULL;
1735 		}
1736 		buf->buf = alloc_cache_memory_buffer(file);
1737 	}
1738 
1739 	buf->buf_size = CACHE_BUFFER_SIZE;
1740 	buf->offset = offset;
1741 
1742 	pthread_spin_lock(&g_caches_lock);
1743 	if (file->tree->present_mask == 0) {
1744 		TAILQ_INSERT_TAIL(&g_caches, file, cache_tailq);
1745 	}
1746 	file->tree = spdk_tree_insert_buffer(file->tree, buf);
1747 	pthread_spin_unlock(&g_caches_lock);
1748 
1749 	return buf;
1750 }
1751 
1752 static struct cache_buffer *
1753 cache_append_buffer(struct spdk_file *file)
1754 {
1755 	struct cache_buffer *last;
1756 
1757 	assert(file->last == NULL || file->last->bytes_filled == file->last->buf_size);
1758 	assert((file->append_pos % CACHE_BUFFER_SIZE) == 0);
1759 
1760 	last = cache_insert_buffer(file, file->append_pos);
1761 	if (last == NULL) {
1762 		SPDK_DEBUGLOG(SPDK_LOG_BLOBFS, "cache_insert_buffer failed\n");
1763 		return NULL;
1764 	}
1765 
1766 	file->last = last;
1767 
1768 	return last;
1769 }
1770 
1771 static void
1772 __wake_caller(struct spdk_fs_cb_args *args)
1773 {
1774 	sem_post(args->sem);
1775 }
1776 
1777 static void __check_sync_reqs(struct spdk_file *file);
1778 
1779 static void
1780 __file_cache_finish_sync(struct spdk_file *file)
1781 {
1782 	struct spdk_fs_request *sync_req;
1783 	struct spdk_fs_cb_args *sync_args;
1784 
1785 	pthread_spin_lock(&file->lock);
1786 	sync_req = TAILQ_FIRST(&file->sync_requests);
1787 	sync_args = &sync_req->args;
1788 	assert(sync_args->op.sync.offset <= file->length_flushed);
1789 	BLOBFS_TRACE(file, "sync done offset=%jx\n", sync_args->op.sync.offset);
1790 	TAILQ_REMOVE(&file->sync_requests, sync_req, args.op.sync.tailq);
1791 	pthread_spin_unlock(&file->lock);
1792 
1793 	sync_args->fn.file_op(sync_args->arg, 0);
1794 	__check_sync_reqs(file);
1795 
1796 	pthread_spin_lock(&file->lock);
1797 	free_fs_request(sync_req);
1798 	pthread_spin_unlock(&file->lock);
1799 }
1800 
1801 static void
1802 __file_cache_finish_sync_bs_cb(void *ctx, int bserrno)
1803 {
1804 	struct spdk_file *file = ctx;
1805 
1806 	__file_cache_finish_sync(file);
1807 }
1808 
1809 static void
1810 __free_args(struct spdk_fs_cb_args *args)
1811 {
1812 	struct spdk_fs_request *req;
1813 
1814 	if (!args->from_request) {
1815 		free(args);
1816 	} else {
1817 		/* Depends on args being at the start of the spdk_fs_request structure. */
1818 		req = (struct spdk_fs_request *)args;
1819 		free_fs_request(req);
1820 	}
1821 }
1822 
1823 static void
1824 __check_sync_reqs(struct spdk_file *file)
1825 {
1826 	struct spdk_fs_request *sync_req;
1827 
1828 	pthread_spin_lock(&file->lock);
1829 
1830 	TAILQ_FOREACH(sync_req, &file->sync_requests, args.op.sync.tailq) {
1831 		if (sync_req->args.op.sync.offset <= file->length_flushed) {
1832 			break;
1833 		}
1834 	}
1835 
1836 	if (sync_req != NULL && !sync_req->args.op.sync.xattr_in_progress) {
1837 		BLOBFS_TRACE(file, "set xattr length 0x%jx\n", file->length_flushed);
1838 		sync_req->args.op.sync.xattr_in_progress = true;
1839 		spdk_blob_set_xattr(file->blob, "length", &file->length_flushed,
1840 				    sizeof(file->length_flushed));
1841 
1842 		pthread_spin_unlock(&file->lock);
1843 		spdk_blob_sync_md(file->blob, __file_cache_finish_sync_bs_cb, file);
1844 	} else {
1845 		pthread_spin_unlock(&file->lock);
1846 	}
1847 }
1848 
1849 static void
1850 __file_flush_done(void *arg, int bserrno)
1851 {
1852 	struct spdk_fs_cb_args *args = arg;
1853 	struct spdk_file *file = args->file;
1854 	struct cache_buffer *next = args->op.flush.cache_buffer;
1855 
1856 	BLOBFS_TRACE(file, "length=%jx\n", args->op.flush.length);
1857 
1858 	pthread_spin_lock(&file->lock);
1859 	next->in_progress = false;
1860 	next->bytes_flushed += args->op.flush.length;
1861 	file->length_flushed += args->op.flush.length;
1862 	if (file->length_flushed > file->length) {
1863 		file->length = file->length_flushed;
1864 	}
1865 	if (next->bytes_flushed == next->buf_size) {
1866 		BLOBFS_TRACE(file, "write buffer fully flushed 0x%jx\n", file->length_flushed);
1867 		next = spdk_tree_find_buffer(file->tree, file->length_flushed);
1868 	}
1869 
1870 	/*
1871 	 * Assert that there is no cached data that extends past the end of the underlying
1872 	 *  blob.
1873 	 */
1874 	assert(next == NULL || next->offset < __file_get_blob_size(file) ||
1875 	       next->bytes_filled == 0);
1876 
1877 	pthread_spin_unlock(&file->lock);
1878 
1879 	__check_sync_reqs(file);
1880 
1881 	__file_flush(args);
1882 }
1883 
1884 static void
1885 __file_flush(void *_args)
1886 {
1887 	struct spdk_fs_cb_args *args = _args;
1888 	struct spdk_file *file = args->file;
1889 	struct cache_buffer *next;
1890 	uint64_t offset, length, start_page, num_pages;
1891 	uint32_t page_size;
1892 
1893 	pthread_spin_lock(&file->lock);
1894 	next = spdk_tree_find_buffer(file->tree, file->length_flushed);
1895 	if (next == NULL || next->in_progress) {
1896 		/*
1897 		 * There is either no data to flush, or a flush I/O is already in
1898 		 *  progress.  So return immediately - if a flush I/O is in
1899 		 *  progress we will flush more data after that is completed.
1900 		 */
1901 		__free_args(args);
1902 		pthread_spin_unlock(&file->lock);
1903 		return;
1904 	}
1905 
1906 	offset = next->offset + next->bytes_flushed;
1907 	length = next->bytes_filled - next->bytes_flushed;
1908 	if (length == 0) {
1909 		__free_args(args);
1910 		pthread_spin_unlock(&file->lock);
1911 		return;
1912 	}
1913 	args->op.flush.length = length;
1914 	args->op.flush.cache_buffer = next;
1915 
1916 	__get_page_parameters(file, offset, length, &start_page, &page_size, &num_pages);
1917 
1918 	next->in_progress = true;
1919 	BLOBFS_TRACE(file, "offset=%jx length=%jx page start=%jx num=%jx\n",
1920 		     offset, length, start_page, num_pages);
1921 	pthread_spin_unlock(&file->lock);
1922 	spdk_bs_io_write_blob(file->blob, file->fs->sync_target.sync_fs_channel->bs_channel,
1923 			      next->buf + (start_page * page_size) - next->offset,
1924 			      start_page, num_pages,
1925 			      __file_flush_done, args);
1926 }
1927 
1928 static void
1929 __file_extend_done(void *arg, int bserrno)
1930 {
1931 	struct spdk_fs_cb_args *args = arg;
1932 
1933 	__wake_caller(args);
1934 }
1935 
1936 static void
1937 __file_extend_blob(void *_args)
1938 {
1939 	struct spdk_fs_cb_args *args = _args;
1940 	struct spdk_file *file = args->file;
1941 
1942 	spdk_blob_resize(file->blob, args->op.resize.num_clusters);
1943 
1944 	spdk_blob_sync_md(file->blob, __file_extend_done, args);
1945 }
1946 
1947 static void
1948 __rw_from_file_done(void *arg, int bserrno)
1949 {
1950 	struct spdk_fs_cb_args *args = arg;
1951 
1952 	__wake_caller(args);
1953 	__free_args(args);
1954 }
1955 
1956 static void
1957 __rw_from_file(void *_args)
1958 {
1959 	struct spdk_fs_cb_args *args = _args;
1960 	struct spdk_file *file = args->file;
1961 
1962 	if (args->op.rw.is_read) {
1963 		spdk_file_read_async(file, file->fs->sync_target.sync_io_channel, args->op.rw.user_buf,
1964 				     args->op.rw.offset, args->op.rw.length,
1965 				     __rw_from_file_done, args);
1966 	} else {
1967 		spdk_file_write_async(file, file->fs->sync_target.sync_io_channel, args->op.rw.user_buf,
1968 				      args->op.rw.offset, args->op.rw.length,
1969 				      __rw_from_file_done, args);
1970 	}
1971 }
1972 
1973 static int
1974 __send_rw_from_file(struct spdk_file *file, sem_t *sem, void *payload,
1975 		    uint64_t offset, uint64_t length, bool is_read)
1976 {
1977 	struct spdk_fs_cb_args *args;
1978 
1979 	args = calloc(1, sizeof(*args));
1980 	if (args == NULL) {
1981 		sem_post(sem);
1982 		return -ENOMEM;
1983 	}
1984 
1985 	args->file = file;
1986 	args->sem = sem;
1987 	args->op.rw.user_buf = payload;
1988 	args->op.rw.offset = offset;
1989 	args->op.rw.length = length;
1990 	args->op.rw.is_read = is_read;
1991 	file->fs->send_request(__rw_from_file, args);
1992 	return 0;
1993 }
1994 
1995 int
1996 spdk_file_write(struct spdk_file *file, struct spdk_io_channel *_channel,
1997 		void *payload, uint64_t offset, uint64_t length)
1998 {
1999 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
2000 	struct spdk_fs_cb_args *args;
2001 	uint64_t rem_length, copy, blob_size, cluster_sz;
2002 	uint32_t cache_buffers_filled = 0;
2003 	uint8_t *cur_payload;
2004 	struct cache_buffer *last;
2005 
2006 	BLOBFS_TRACE_RW(file, "offset=%jx length=%jx\n", offset, length);
2007 
2008 	if (length == 0) {
2009 		return 0;
2010 	}
2011 
2012 	if (offset != file->append_pos) {
2013 		BLOBFS_TRACE(file, " error offset=%jx append_pos=%jx\n", offset, file->append_pos);
2014 		return -EINVAL;
2015 	}
2016 
2017 	pthread_spin_lock(&file->lock);
2018 	file->open_for_writing = true;
2019 
2020 	if (file->last == NULL) {
2021 		if (file->append_pos % CACHE_BUFFER_SIZE == 0) {
2022 			cache_append_buffer(file);
2023 		} else {
2024 			int rc;
2025 
2026 			file->append_pos += length;
2027 			pthread_spin_unlock(&file->lock);
2028 			rc = __send_rw_from_file(file, &channel->sem, payload,
2029 						 offset, length, false);
2030 			sem_wait(&channel->sem);
2031 			return rc;
2032 		}
2033 	}
2034 
2035 	blob_size = __file_get_blob_size(file);
2036 
2037 	if ((offset + length) > blob_size) {
2038 		struct spdk_fs_cb_args extend_args = {};
2039 
2040 		cluster_sz = file->fs->bs_opts.cluster_sz;
2041 		extend_args.sem = &channel->sem;
2042 		extend_args.op.resize.num_clusters = __bytes_to_clusters((offset + length), cluster_sz);
2043 		extend_args.file = file;
2044 		BLOBFS_TRACE(file, "start resize to %u clusters\n", extend_args.op.resize.num_clusters);
2045 		pthread_spin_unlock(&file->lock);
2046 		file->fs->send_request(__file_extend_blob, &extend_args);
2047 		sem_wait(&channel->sem);
2048 	}
2049 
2050 	last = file->last;
2051 	rem_length = length;
2052 	cur_payload = payload;
2053 	while (rem_length > 0) {
2054 		copy = last->buf_size - last->bytes_filled;
2055 		if (copy > rem_length) {
2056 			copy = rem_length;
2057 		}
2058 		BLOBFS_TRACE_RW(file, "  fill offset=%jx length=%jx\n", file->append_pos, copy);
2059 		memcpy(&last->buf[last->bytes_filled], cur_payload, copy);
2060 		file->append_pos += copy;
2061 		if (file->length < file->append_pos) {
2062 			file->length = file->append_pos;
2063 		}
2064 		cur_payload += copy;
2065 		last->bytes_filled += copy;
2066 		rem_length -= copy;
2067 		if (last->bytes_filled == last->buf_size) {
2068 			cache_buffers_filled++;
2069 			last = cache_append_buffer(file);
2070 			if (last == NULL) {
2071 				BLOBFS_TRACE(file, "nomem\n");
2072 				pthread_spin_unlock(&file->lock);
2073 				return -ENOMEM;
2074 			}
2075 		}
2076 	}
2077 
2078 	if (cache_buffers_filled == 0) {
2079 		pthread_spin_unlock(&file->lock);
2080 		return 0;
2081 	}
2082 
2083 	args = calloc(1, sizeof(*args));
2084 	if (args == NULL) {
2085 		pthread_spin_unlock(&file->lock);
2086 		return -ENOMEM;
2087 	}
2088 
2089 	args->file = file;
2090 	file->fs->send_request(__file_flush, args);
2091 	pthread_spin_unlock(&file->lock);
2092 	return 0;
2093 }
2094 
2095 static void
2096 __readahead_done(void *arg, int bserrno)
2097 {
2098 	struct spdk_fs_cb_args *args = arg;
2099 	struct cache_buffer *cache_buffer = args->op.readahead.cache_buffer;
2100 	struct spdk_file *file = args->file;
2101 
2102 	BLOBFS_TRACE(file, "offset=%jx\n", cache_buffer->offset);
2103 
2104 	pthread_spin_lock(&file->lock);
2105 	cache_buffer->bytes_filled = args->op.readahead.length;
2106 	cache_buffer->bytes_flushed = args->op.readahead.length;
2107 	cache_buffer->in_progress = false;
2108 	pthread_spin_unlock(&file->lock);
2109 
2110 	__free_args(args);
2111 }
2112 
2113 static void
2114 __readahead(void *_args)
2115 {
2116 	struct spdk_fs_cb_args *args = _args;
2117 	struct spdk_file *file = args->file;
2118 	uint64_t offset, length, start_page, num_pages;
2119 	uint32_t page_size;
2120 
2121 	offset = args->op.readahead.offset;
2122 	length = args->op.readahead.length;
2123 	assert(length > 0);
2124 
2125 	__get_page_parameters(file, offset, length, &start_page, &page_size, &num_pages);
2126 
2127 	BLOBFS_TRACE(file, "offset=%jx length=%jx page start=%jx num=%jx\n",
2128 		     offset, length, start_page, num_pages);
2129 	spdk_bs_io_read_blob(file->blob, file->fs->sync_target.sync_fs_channel->bs_channel,
2130 			     args->op.readahead.cache_buffer->buf,
2131 			     start_page, num_pages,
2132 			     __readahead_done, args);
2133 }
2134 
2135 static uint64_t
2136 __next_cache_buffer_offset(uint64_t offset)
2137 {
2138 	return (offset + CACHE_BUFFER_SIZE) & ~(CACHE_TREE_LEVEL_MASK(0));
2139 }
2140 
2141 static void
2142 check_readahead(struct spdk_file *file, uint64_t offset)
2143 {
2144 	struct spdk_fs_cb_args *args;
2145 
2146 	offset = __next_cache_buffer_offset(offset);
2147 	if (spdk_tree_find_buffer(file->tree, offset) != NULL || file->length <= offset) {
2148 		return;
2149 	}
2150 
2151 	args = calloc(1, sizeof(*args));
2152 	if (args == NULL) {
2153 		return;
2154 	}
2155 
2156 	BLOBFS_TRACE(file, "offset=%jx\n", offset);
2157 
2158 	args->file = file;
2159 	args->op.readahead.offset = offset;
2160 	args->op.readahead.cache_buffer = cache_insert_buffer(file, offset);
2161 	args->op.readahead.cache_buffer->in_progress = true;
2162 	if (file->length < (offset + CACHE_BUFFER_SIZE)) {
2163 		args->op.readahead.length = file->length & (CACHE_BUFFER_SIZE - 1);
2164 	} else {
2165 		args->op.readahead.length = CACHE_BUFFER_SIZE;
2166 	}
2167 	file->fs->send_request(__readahead, args);
2168 }
2169 
2170 static int
2171 __file_read(struct spdk_file *file, void *payload, uint64_t offset, uint64_t length, sem_t *sem)
2172 {
2173 	struct cache_buffer *buf;
2174 	int rc;
2175 
2176 	buf = spdk_tree_find_filled_buffer(file->tree, offset);
2177 	if (buf == NULL) {
2178 		pthread_spin_unlock(&file->lock);
2179 		rc = __send_rw_from_file(file, sem, payload, offset, length, true);
2180 		pthread_spin_lock(&file->lock);
2181 		return rc;
2182 	}
2183 
2184 	if ((offset + length) > (buf->offset + buf->bytes_filled)) {
2185 		length = buf->offset + buf->bytes_filled - offset;
2186 	}
2187 	BLOBFS_TRACE(file, "read %p offset=%ju length=%ju\n", payload, offset, length);
2188 	memcpy(payload, &buf->buf[offset - buf->offset], length);
2189 	if ((offset + length) % CACHE_BUFFER_SIZE == 0) {
2190 		pthread_spin_lock(&g_caches_lock);
2191 		spdk_tree_remove_buffer(file->tree, buf);
2192 		if (file->tree->present_mask == 0) {
2193 			TAILQ_REMOVE(&g_caches, file, cache_tailq);
2194 		}
2195 		pthread_spin_unlock(&g_caches_lock);
2196 	}
2197 
2198 	sem_post(sem);
2199 	return 0;
2200 }
2201 
2202 int64_t
2203 spdk_file_read(struct spdk_file *file, struct spdk_io_channel *_channel,
2204 	       void *payload, uint64_t offset, uint64_t length)
2205 {
2206 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
2207 	uint64_t final_offset, final_length;
2208 	uint32_t sub_reads = 0;
2209 	int rc = 0;
2210 
2211 	pthread_spin_lock(&file->lock);
2212 
2213 	BLOBFS_TRACE_RW(file, "offset=%ju length=%ju\n", offset, length);
2214 
2215 	file->open_for_writing = false;
2216 
2217 	if (length == 0 || offset >= file->append_pos) {
2218 		pthread_spin_unlock(&file->lock);
2219 		return 0;
2220 	}
2221 
2222 	if (offset + length > file->append_pos) {
2223 		length = file->append_pos - offset;
2224 	}
2225 
2226 	if (offset != file->next_seq_offset) {
2227 		file->seq_byte_count = 0;
2228 	}
2229 	file->seq_byte_count += length;
2230 	file->next_seq_offset = offset + length;
2231 	if (file->seq_byte_count >= CACHE_READAHEAD_THRESHOLD) {
2232 		check_readahead(file, offset);
2233 		check_readahead(file, offset + CACHE_BUFFER_SIZE);
2234 	}
2235 
2236 	final_length = 0;
2237 	final_offset = offset + length;
2238 	while (offset < final_offset) {
2239 		length = NEXT_CACHE_BUFFER_OFFSET(offset) - offset;
2240 		if (length > (final_offset - offset)) {
2241 			length = final_offset - offset;
2242 		}
2243 		rc = __file_read(file, payload, offset, length, &channel->sem);
2244 		if (rc == 0) {
2245 			final_length += length;
2246 		} else {
2247 			break;
2248 		}
2249 		payload += length;
2250 		offset += length;
2251 		sub_reads++;
2252 	}
2253 	pthread_spin_unlock(&file->lock);
2254 	while (sub_reads-- > 0) {
2255 		sem_wait(&channel->sem);
2256 	}
2257 	if (rc == 0) {
2258 		return final_length;
2259 	} else {
2260 		return rc;
2261 	}
2262 }
2263 
2264 static void
2265 _file_sync(struct spdk_file *file, struct spdk_fs_channel *channel,
2266 	   spdk_file_op_complete cb_fn, void *cb_arg)
2267 {
2268 	struct spdk_fs_request *sync_req;
2269 	struct spdk_fs_request *flush_req;
2270 	struct spdk_fs_cb_args *sync_args;
2271 	struct spdk_fs_cb_args *flush_args;
2272 
2273 	BLOBFS_TRACE(file, "offset=%jx\n", file->append_pos);
2274 
2275 	pthread_spin_lock(&file->lock);
2276 	if (file->append_pos <= file->length_flushed || file->last == NULL) {
2277 		BLOBFS_TRACE(file, "done - no data to flush\n");
2278 		pthread_spin_unlock(&file->lock);
2279 		cb_fn(cb_arg, 0);
2280 		return;
2281 	}
2282 
2283 	sync_req = alloc_fs_request(channel);
2284 	assert(sync_req != NULL);
2285 	sync_args = &sync_req->args;
2286 
2287 	flush_req = alloc_fs_request(channel);
2288 	assert(flush_req != NULL);
2289 	flush_args = &flush_req->args;
2290 
2291 	sync_args->file = file;
2292 	sync_args->fn.file_op = cb_fn;
2293 	sync_args->arg = cb_arg;
2294 	sync_args->op.sync.offset = file->append_pos;
2295 	sync_args->op.sync.xattr_in_progress = false;
2296 	TAILQ_INSERT_TAIL(&file->sync_requests, sync_req, args.op.sync.tailq);
2297 	pthread_spin_unlock(&file->lock);
2298 
2299 	flush_args->file = file;
2300 	channel->send_request(__file_flush, flush_args);
2301 }
2302 
2303 int
2304 spdk_file_sync(struct spdk_file *file, struct spdk_io_channel *_channel)
2305 {
2306 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
2307 
2308 	_file_sync(file, channel, __sem_post, &channel->sem);
2309 	sem_wait(&channel->sem);
2310 
2311 	return 0;
2312 }
2313 
2314 void
2315 spdk_file_sync_async(struct spdk_file *file, struct spdk_io_channel *_channel,
2316 		     spdk_file_op_complete cb_fn, void *cb_arg)
2317 {
2318 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
2319 
2320 	_file_sync(file, channel, cb_fn, cb_arg);
2321 }
2322 
2323 void
2324 spdk_file_set_priority(struct spdk_file *file, uint32_t priority)
2325 {
2326 	BLOBFS_TRACE(file, "priority=%u\n", priority);
2327 	file->priority = priority;
2328 
2329 }
2330 
2331 /*
2332  * Close routines
2333  */
2334 
2335 static void
2336 __file_close_async_done(void *ctx, int bserrno)
2337 {
2338 	struct spdk_fs_request *req = ctx;
2339 	struct spdk_fs_cb_args *args = &req->args;
2340 	struct spdk_file *file = args->file;
2341 
2342 	if (file->is_deleted) {
2343 		spdk_fs_delete_file_async(file->fs, file->name, blob_delete_cb, ctx);
2344 		return;
2345 	}
2346 	args->fn.file_op(args->arg, bserrno);
2347 	free_fs_request(req);
2348 }
2349 
2350 static void
2351 __file_close_async(struct spdk_file *file, struct spdk_fs_request *req)
2352 {
2353 	struct spdk_blob *blob;
2354 
2355 	pthread_spin_lock(&file->lock);
2356 	if (file->ref_count == 0) {
2357 		pthread_spin_unlock(&file->lock);
2358 		__file_close_async_done(req, -EBADF);
2359 		return;
2360 	}
2361 
2362 	file->ref_count--;
2363 	if (file->ref_count > 0) {
2364 		pthread_spin_unlock(&file->lock);
2365 		__file_close_async_done(req, 0);
2366 		return;
2367 	}
2368 
2369 	pthread_spin_unlock(&file->lock);
2370 
2371 	blob = file->blob;
2372 	file->blob = NULL;
2373 	spdk_blob_close(blob, __file_close_async_done, req);
2374 }
2375 
2376 static void
2377 __file_close_async__sync_done(void *arg, int fserrno)
2378 {
2379 	struct spdk_fs_request *req = arg;
2380 	struct spdk_fs_cb_args *args = &req->args;
2381 
2382 	__file_close_async(args->file, req);
2383 }
2384 
2385 void
2386 spdk_file_close_async(struct spdk_file *file, spdk_file_op_complete cb_fn, void *cb_arg)
2387 {
2388 	struct spdk_fs_request *req;
2389 	struct spdk_fs_cb_args *args;
2390 
2391 	req = alloc_fs_request(file->fs->md_target.md_fs_channel);
2392 	if (req == NULL) {
2393 		cb_fn(cb_arg, -ENOMEM);
2394 		return;
2395 	}
2396 
2397 	args = &req->args;
2398 	args->file = file;
2399 	args->fn.file_op = cb_fn;
2400 	args->arg = cb_arg;
2401 
2402 	spdk_file_sync_async(file, file->fs->md_target.md_io_channel, __file_close_async__sync_done, req);
2403 }
2404 
2405 static void
2406 __file_close_done(void *arg, int fserrno)
2407 {
2408 	struct spdk_fs_cb_args *args = arg;
2409 
2410 	args->rc = fserrno;
2411 	sem_post(args->sem);
2412 }
2413 
2414 static void
2415 __file_close(void *arg)
2416 {
2417 	struct spdk_fs_request *req = arg;
2418 	struct spdk_fs_cb_args *args = &req->args;
2419 	struct spdk_file *file = args->file;
2420 
2421 	__file_close_async(file, req);
2422 }
2423 
2424 int
2425 spdk_file_close(struct spdk_file *file, struct spdk_io_channel *_channel)
2426 {
2427 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
2428 	struct spdk_fs_request *req;
2429 	struct spdk_fs_cb_args *args;
2430 
2431 	req = alloc_fs_request(channel);
2432 	assert(req != NULL);
2433 
2434 	args = &req->args;
2435 
2436 	spdk_file_sync(file, _channel);
2437 	BLOBFS_TRACE(file, "name=%s\n", file->name);
2438 	args->file = file;
2439 	args->sem = &channel->sem;
2440 	args->fn.file_op = __file_close_done;
2441 	args->arg = req;
2442 	channel->send_request(__file_close, req);
2443 	sem_wait(&channel->sem);
2444 
2445 	return args->rc;
2446 }
2447 
2448 static void
2449 cache_free_buffers(struct spdk_file *file)
2450 {
2451 	BLOBFS_TRACE(file, "free=%s\n", file->name);
2452 	pthread_spin_lock(&file->lock);
2453 	pthread_spin_lock(&g_caches_lock);
2454 	if (file->tree->present_mask == 0) {
2455 		pthread_spin_unlock(&g_caches_lock);
2456 		pthread_spin_unlock(&file->lock);
2457 		return;
2458 	}
2459 	spdk_tree_free_buffers(file->tree);
2460 
2461 	TAILQ_REMOVE(&g_caches, file, cache_tailq);
2462 	/* If not freed, put it in the end of the queue */
2463 	if (file->tree->present_mask != 0) {
2464 		TAILQ_INSERT_TAIL(&g_caches, file, cache_tailq);
2465 	}
2466 	file->last = NULL;
2467 	pthread_spin_unlock(&g_caches_lock);
2468 	pthread_spin_unlock(&file->lock);
2469 }
2470 
2471 SPDK_LOG_REGISTER_COMPONENT("blobfs", SPDK_LOG_BLOBFS)
2472 SPDK_LOG_REGISTER_COMPONENT("blobfs_rw", SPDK_LOG_BLOBFS_RW)
2473