xref: /spdk/lib/blobfs/blobfs.c (revision a6014eb2adf0c95816b23ef94a911005fa047511)
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 "blobfs_internal.h"
38 
39 #include "spdk/queue.h"
40 #include "spdk/io_channel.h"
41 #include "spdk/assert.h"
42 #include "spdk/env.h"
43 #include "spdk/util.h"
44 #include "spdk_internal/log.h"
45 
46 #define BLOBFS_TRACE(file, str, args...) \
47 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s " str, file->name, ##args)
48 
49 #define BLOBFS_TRACE_RW(file, str, args...) \
50 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS_RW, "file=%s " str, file->name, ##args)
51 
52 #define BLOBFS_CACHE_SIZE (4ULL * 1024 * 1024 * 1024)
53 
54 static uint64_t g_fs_cache_size = BLOBFS_CACHE_SIZE;
55 static struct spdk_mempool *g_cache_pool;
56 static TAILQ_HEAD(, spdk_file) g_caches;
57 static int g_fs_count = 0;
58 static pthread_mutex_t g_cache_init_lock = PTHREAD_MUTEX_INITIALIZER;
59 static pthread_spinlock_t g_caches_lock;
60 
61 static void
62 __sem_post(void *arg, int bserrno)
63 {
64 	sem_t *sem = arg;
65 
66 	sem_post(sem);
67 }
68 
69 void
70 spdk_cache_buffer_free(struct cache_buffer *cache_buffer)
71 {
72 	spdk_mempool_put(g_cache_pool, cache_buffer->buf);
73 	free(cache_buffer);
74 }
75 
76 #define CACHE_READAHEAD_THRESHOLD	(128 * 1024)
77 
78 struct spdk_file {
79 	struct spdk_filesystem	*fs;
80 	struct spdk_blob	*blob;
81 	char			*name;
82 	uint64_t		length;
83 	bool			open_for_writing;
84 	uint64_t		length_flushed;
85 	uint64_t		append_pos;
86 	uint64_t		seq_byte_count;
87 	uint64_t		next_seq_offset;
88 	uint32_t		priority;
89 	TAILQ_ENTRY(spdk_file)	tailq;
90 	spdk_blob_id		blobid;
91 	uint32_t		ref_count;
92 	pthread_spinlock_t	lock;
93 	struct cache_buffer	*last;
94 	struct cache_tree	*tree;
95 	TAILQ_HEAD(open_requests_head, spdk_fs_request) open_requests;
96 	TAILQ_HEAD(sync_requests_head, spdk_fs_request) sync_requests;
97 	TAILQ_ENTRY(spdk_file)	cache_tailq;
98 };
99 
100 struct spdk_filesystem {
101 	struct spdk_blob_store	*bs;
102 	TAILQ_HEAD(, spdk_file)	files;
103 	struct spdk_bs_opts	bs_opts;
104 	struct spdk_bs_dev	*bdev;
105 	fs_send_request_fn	send_request;
106 
107 	struct {
108 		uint32_t		max_ops;
109 		struct spdk_io_channel	*sync_io_channel;
110 		struct spdk_fs_channel	*sync_fs_channel;
111 	} sync_target;
112 
113 	struct {
114 		uint32_t		max_ops;
115 		struct spdk_io_channel	*md_io_channel;
116 		struct spdk_fs_channel	*md_fs_channel;
117 	} md_target;
118 
119 	struct {
120 		uint32_t		max_ops;
121 	} io_target;
122 };
123 
124 struct spdk_fs_cb_args {
125 	union {
126 		spdk_fs_op_with_handle_complete		fs_op_with_handle;
127 		spdk_fs_op_complete			fs_op;
128 		spdk_file_op_with_handle_complete	file_op_with_handle;
129 		spdk_file_op_complete			file_op;
130 		spdk_file_stat_op_complete		stat_op;
131 	} fn;
132 	void *arg;
133 	sem_t *sem;
134 	struct spdk_filesystem *fs;
135 	struct spdk_file *file;
136 	int rc;
137 	bool from_request;
138 	union {
139 		struct {
140 			uint64_t	length;
141 		} truncate;
142 		struct {
143 			struct spdk_io_channel	*channel;
144 			void		*user_buf;
145 			void		*pin_buf;
146 			int		is_read;
147 			off_t		offset;
148 			size_t		length;
149 			uint64_t	start_page;
150 			uint64_t	num_pages;
151 			uint32_t	blocklen;
152 		} rw;
153 		struct {
154 			const char	*old_name;
155 			const char	*new_name;
156 		} rename;
157 		struct {
158 			struct cache_buffer	*cache_buffer;
159 			uint64_t		length;
160 		} flush;
161 		struct {
162 			struct cache_buffer	*cache_buffer;
163 			uint64_t		length;
164 			uint64_t		offset;
165 		} readahead;
166 		struct {
167 			uint64_t			offset;
168 			TAILQ_ENTRY(spdk_fs_request)	tailq;
169 			bool				xattr_in_progress;
170 		} sync;
171 		struct {
172 			uint32_t			num_clusters;
173 		} resize;
174 		struct {
175 			const char	*name;
176 			uint32_t	flags;
177 			TAILQ_ENTRY(spdk_fs_request)	tailq;
178 		} open;
179 		struct {
180 			const char	*name;
181 		} create;
182 		struct {
183 			const char	*name;
184 		} delete;
185 		struct {
186 			const char	*name;
187 		} stat;
188 	} op;
189 };
190 
191 static void cache_free_buffers(struct spdk_file *file);
192 
193 static void
194 __initialize_cache(void)
195 {
196 	assert(g_cache_pool == NULL);
197 
198 	g_cache_pool = spdk_mempool_create("spdk_fs_cache",
199 					   g_fs_cache_size / CACHE_BUFFER_SIZE,
200 					   CACHE_BUFFER_SIZE, -1, SPDK_ENV_SOCKET_ID_ANY);
201 	TAILQ_INIT(&g_caches);
202 	pthread_spin_init(&g_caches_lock, 0);
203 }
204 
205 static void
206 __free_cache(void)
207 {
208 	assert(g_cache_pool != NULL);
209 
210 	spdk_mempool_free(g_cache_pool);
211 	g_cache_pool = NULL;
212 }
213 
214 static uint64_t
215 __file_get_blob_size(struct spdk_file *file)
216 {
217 	uint64_t cluster_sz;
218 
219 	cluster_sz = file->fs->bs_opts.cluster_sz;
220 	return cluster_sz * spdk_blob_get_num_clusters(file->blob);
221 }
222 
223 struct spdk_fs_request {
224 	struct spdk_fs_cb_args		args;
225 	TAILQ_ENTRY(spdk_fs_request)	link;
226 	struct spdk_fs_channel		*channel;
227 };
228 
229 struct spdk_fs_channel {
230 	struct spdk_fs_request		*req_mem;
231 	TAILQ_HEAD(, spdk_fs_request)	reqs;
232 	sem_t				sem;
233 	struct spdk_filesystem		*fs;
234 	struct spdk_io_channel		*bs_channel;
235 	fs_send_request_fn		send_request;
236 	bool				sync;
237 	pthread_spinlock_t		lock;
238 };
239 
240 static struct spdk_fs_request *
241 alloc_fs_request(struct spdk_fs_channel *channel)
242 {
243 	struct spdk_fs_request *req;
244 
245 	if (channel->sync) {
246 		pthread_spin_lock(&channel->lock);
247 	}
248 
249 	req = TAILQ_FIRST(&channel->reqs);
250 	if (req) {
251 		TAILQ_REMOVE(&channel->reqs, req, link);
252 	}
253 
254 	if (channel->sync) {
255 		pthread_spin_unlock(&channel->lock);
256 	}
257 
258 	if (req == NULL) {
259 		return NULL;
260 	}
261 	memset(req, 0, sizeof(*req));
262 	req->channel = channel;
263 	req->args.from_request = true;
264 
265 	return req;
266 }
267 
268 static void
269 free_fs_request(struct spdk_fs_request *req)
270 {
271 	struct spdk_fs_channel *channel = req->channel;
272 
273 	if (channel->sync) {
274 		pthread_spin_lock(&channel->lock);
275 	}
276 
277 	TAILQ_INSERT_HEAD(&req->channel->reqs, req, link);
278 
279 	if (channel->sync) {
280 		pthread_spin_unlock(&channel->lock);
281 	}
282 }
283 
284 static int
285 _spdk_fs_channel_create(struct spdk_filesystem *fs, struct spdk_fs_channel *channel,
286 			uint32_t max_ops)
287 {
288 	uint32_t i;
289 
290 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_fs_request));
291 	if (!channel->req_mem) {
292 		return -1;
293 	}
294 
295 	TAILQ_INIT(&channel->reqs);
296 	sem_init(&channel->sem, 0, 0);
297 
298 	for (i = 0; i < max_ops; i++) {
299 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
300 	}
301 
302 	channel->fs = fs;
303 
304 	return 0;
305 }
306 
307 static int
308 _spdk_fs_md_channel_create(void *io_device, void *ctx_buf)
309 {
310 	struct spdk_filesystem		*fs;
311 	struct spdk_fs_channel		*channel = ctx_buf;
312 
313 	fs = SPDK_CONTAINEROF(io_device, struct spdk_filesystem, md_target);
314 
315 	return _spdk_fs_channel_create(fs, channel, fs->md_target.max_ops);
316 }
317 
318 static int
319 _spdk_fs_sync_channel_create(void *io_device, void *ctx_buf)
320 {
321 	struct spdk_filesystem		*fs;
322 	struct spdk_fs_channel		*channel = ctx_buf;
323 
324 	fs = SPDK_CONTAINEROF(io_device, struct spdk_filesystem, sync_target);
325 
326 	return _spdk_fs_channel_create(fs, channel, fs->sync_target.max_ops);
327 }
328 
329 static int
330 _spdk_fs_io_channel_create(void *io_device, void *ctx_buf)
331 {
332 	struct spdk_filesystem		*fs;
333 	struct spdk_fs_channel		*channel = ctx_buf;
334 
335 	fs = SPDK_CONTAINEROF(io_device, struct spdk_filesystem, io_target);
336 
337 	return _spdk_fs_channel_create(fs, channel, fs->io_target.max_ops);
338 }
339 
340 static void
341 _spdk_fs_channel_destroy(void *io_device, void *ctx_buf)
342 {
343 	struct spdk_fs_channel *channel = ctx_buf;
344 
345 	free(channel->req_mem);
346 	if (channel->bs_channel != NULL) {
347 		spdk_bs_free_io_channel(channel->bs_channel);
348 	}
349 }
350 
351 static void
352 __send_request_direct(fs_request_fn fn, void *arg)
353 {
354 	fn(arg);
355 }
356 
357 static void
358 common_fs_bs_init(struct spdk_filesystem *fs, struct spdk_blob_store *bs)
359 {
360 	fs->bs = bs;
361 	fs->bs_opts.cluster_sz = spdk_bs_get_cluster_size(bs);
362 	fs->md_target.md_fs_channel->bs_channel = spdk_bs_alloc_io_channel(fs->bs);
363 	fs->md_target.md_fs_channel->send_request = __send_request_direct;
364 	fs->sync_target.sync_fs_channel->bs_channel = spdk_bs_alloc_io_channel(fs->bs);
365 	fs->sync_target.sync_fs_channel->send_request = __send_request_direct;
366 
367 	pthread_mutex_lock(&g_cache_init_lock);
368 	if (g_fs_count == 0) {
369 		__initialize_cache();
370 	}
371 	g_fs_count++;
372 	pthread_mutex_unlock(&g_cache_init_lock);
373 }
374 
375 static void
376 init_cb(void *ctx, struct spdk_blob_store *bs, int bserrno)
377 {
378 	struct spdk_fs_request *req = ctx;
379 	struct spdk_fs_cb_args *args = &req->args;
380 	struct spdk_filesystem *fs = args->fs;
381 
382 	if (bserrno == 0) {
383 		common_fs_bs_init(fs, bs);
384 	} else {
385 		free(fs);
386 		fs = NULL;
387 	}
388 
389 	args->fn.fs_op_with_handle(args->arg, fs, bserrno);
390 	free_fs_request(req);
391 }
392 
393 static struct spdk_filesystem *
394 fs_alloc(struct spdk_bs_dev *dev, fs_send_request_fn send_request_fn)
395 {
396 	struct spdk_filesystem *fs;
397 
398 	fs = calloc(1, sizeof(*fs));
399 	if (fs == NULL) {
400 		return NULL;
401 	}
402 
403 	fs->bdev = dev;
404 	fs->send_request = send_request_fn;
405 	TAILQ_INIT(&fs->files);
406 
407 	fs->md_target.max_ops = 512;
408 	spdk_io_device_register(&fs->md_target, _spdk_fs_md_channel_create, _spdk_fs_channel_destroy,
409 				sizeof(struct spdk_fs_channel));
410 	fs->md_target.md_io_channel = spdk_get_io_channel(&fs->md_target);
411 	fs->md_target.md_fs_channel = spdk_io_channel_get_ctx(fs->md_target.md_io_channel);
412 
413 	fs->sync_target.max_ops = 512;
414 	spdk_io_device_register(&fs->sync_target, _spdk_fs_sync_channel_create, _spdk_fs_channel_destroy,
415 				sizeof(struct spdk_fs_channel));
416 	fs->sync_target.sync_io_channel = spdk_get_io_channel(&fs->sync_target);
417 	fs->sync_target.sync_fs_channel = spdk_io_channel_get_ctx(fs->sync_target.sync_io_channel);
418 
419 	fs->io_target.max_ops = 512;
420 	spdk_io_device_register(&fs->io_target, _spdk_fs_io_channel_create, _spdk_fs_channel_destroy,
421 				sizeof(struct spdk_fs_channel));
422 
423 	return fs;
424 }
425 
426 void
427 spdk_fs_init(struct spdk_bs_dev *dev, fs_send_request_fn send_request_fn,
428 	     spdk_fs_op_with_handle_complete cb_fn, void *cb_arg)
429 {
430 	struct spdk_filesystem *fs;
431 	struct spdk_fs_request *req;
432 	struct spdk_fs_cb_args *args;
433 
434 	fs = fs_alloc(dev, send_request_fn);
435 	if (fs == NULL) {
436 		cb_fn(cb_arg, NULL, -ENOMEM);
437 		return;
438 	}
439 
440 	req = alloc_fs_request(fs->md_target.md_fs_channel);
441 	if (req == NULL) {
442 		cb_fn(cb_arg, NULL, -ENOMEM);
443 		return;
444 	}
445 
446 	args = &req->args;
447 	args->fn.fs_op_with_handle = cb_fn;
448 	args->arg = cb_arg;
449 	args->fs = fs;
450 
451 	spdk_bs_init(dev, NULL, init_cb, req);
452 }
453 
454 static struct spdk_file *
455 file_alloc(struct spdk_filesystem *fs)
456 {
457 	struct spdk_file *file;
458 
459 	file = calloc(1, sizeof(*file));
460 	if (file == NULL) {
461 		return NULL;
462 	}
463 
464 	file->fs = fs;
465 	TAILQ_INIT(&file->open_requests);
466 	TAILQ_INIT(&file->sync_requests);
467 	pthread_spin_init(&file->lock, 0);
468 	file->tree = calloc(1, sizeof(*file->tree));
469 	TAILQ_INSERT_TAIL(&fs->files, file, tailq);
470 	file->priority = SPDK_FILE_PRIORITY_LOW;
471 	return file;
472 }
473 
474 static void
475 iter_cb(void *ctx, struct spdk_blob *blob, int rc)
476 {
477 	struct spdk_fs_request *req = ctx;
478 	struct spdk_fs_cb_args *args = &req->args;
479 	struct spdk_filesystem *fs = args->fs;
480 	struct spdk_file *f;
481 	uint64_t *length;
482 	const char *name;
483 	size_t value_len;
484 
485 	if (rc == -ENOENT) {
486 		/* Finished iterating */
487 		args->fn.fs_op_with_handle(args->arg, fs, 0);
488 		free_fs_request(req);
489 		return;
490 	} else if (rc < 0) {
491 		args->fn.fs_op_with_handle(args->arg, fs, rc);
492 		free_fs_request(req);
493 		return;
494 	}
495 
496 	rc = spdk_bs_md_get_xattr_value(blob, "name", (const void **)&name, &value_len);
497 	if (rc < 0) {
498 		args->fn.fs_op_with_handle(args->arg, fs, rc);
499 		free_fs_request(req);
500 		return;
501 	}
502 
503 	rc = spdk_bs_md_get_xattr_value(blob, "length", (const void **)&length, &value_len);
504 	if (rc < 0) {
505 		args->fn.fs_op_with_handle(args->arg, fs, rc);
506 		free_fs_request(req);
507 		return;
508 	}
509 	assert(value_len == 8);
510 
511 	f = file_alloc(fs);
512 	if (f == NULL) {
513 		args->fn.fs_op_with_handle(args->arg, fs, -ENOMEM);
514 		free_fs_request(req);
515 		return;
516 	}
517 
518 	f->name = strdup(name);
519 	f->blobid = spdk_blob_get_id(blob);
520 	f->length = *length;
521 	f->length_flushed = *length;
522 	f->append_pos = *length;
523 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "added file %s length=%ju\n", f->name, f->length);
524 
525 	spdk_bs_md_iter_next(fs->bs, &blob, iter_cb, req);
526 }
527 
528 static void
529 load_cb(void *ctx, struct spdk_blob_store *bs, int bserrno)
530 {
531 	struct spdk_fs_request *req = ctx;
532 	struct spdk_fs_cb_args *args = &req->args;
533 	struct spdk_filesystem *fs = args->fs;
534 
535 	if (bserrno != 0) {
536 		args->fn.fs_op_with_handle(args->arg, NULL, bserrno);
537 		free_fs_request(req);
538 		free(fs);
539 		return;
540 	}
541 
542 	common_fs_bs_init(fs, bs);
543 	spdk_bs_md_iter_first(fs->bs, iter_cb, req);
544 }
545 
546 void
547 spdk_fs_load(struct spdk_bs_dev *dev, fs_send_request_fn send_request_fn,
548 	     spdk_fs_op_with_handle_complete cb_fn, void *cb_arg)
549 {
550 	struct spdk_filesystem *fs;
551 	struct spdk_fs_cb_args *args;
552 	struct spdk_fs_request *req;
553 
554 	fs = fs_alloc(dev, send_request_fn);
555 	if (fs == NULL) {
556 		cb_fn(cb_arg, NULL, -ENOMEM);
557 		return;
558 	}
559 
560 	req = alloc_fs_request(fs->md_target.md_fs_channel);
561 	if (req == NULL) {
562 		cb_fn(cb_arg, NULL, -ENOMEM);
563 		return;
564 	}
565 
566 	args = &req->args;
567 	args->fn.fs_op_with_handle = cb_fn;
568 	args->arg = cb_arg;
569 	args->fs = fs;
570 
571 	spdk_bs_load(dev, load_cb, req);
572 }
573 
574 static void
575 unload_cb(void *ctx, int bserrno)
576 {
577 	struct spdk_fs_request *req = ctx;
578 	struct spdk_fs_cb_args *args = &req->args;
579 	struct spdk_filesystem *fs = args->fs;
580 
581 	pthread_mutex_lock(&g_cache_init_lock);
582 	g_fs_count--;
583 	if (g_fs_count == 0) {
584 		__free_cache();
585 	}
586 	pthread_mutex_unlock(&g_cache_init_lock);
587 
588 	args->fn.fs_op(args->arg, bserrno);
589 	free(req);
590 
591 	spdk_io_device_unregister(&fs->io_target);
592 	spdk_io_device_unregister(&fs->sync_target);
593 	spdk_io_device_unregister(&fs->md_target);
594 
595 	free(fs);
596 }
597 
598 void
599 spdk_fs_unload(struct spdk_filesystem *fs, spdk_fs_op_complete cb_fn, void *cb_arg)
600 {
601 	struct spdk_fs_request *req;
602 	struct spdk_fs_cb_args *args;
603 
604 	/*
605 	 * We must free the md_channel before unloading the blobstore, so just
606 	 *  allocate this request from the general heap.
607 	 */
608 	req = calloc(1, sizeof(*req));
609 	if (req == NULL) {
610 		cb_fn(cb_arg, -ENOMEM);
611 		return;
612 	}
613 
614 	args = &req->args;
615 	args->fn.fs_op = cb_fn;
616 	args->arg = cb_arg;
617 	args->fs = fs;
618 
619 	spdk_fs_free_io_channel(fs->md_target.md_io_channel);
620 	spdk_fs_free_io_channel(fs->sync_target.sync_io_channel);
621 	spdk_bs_unload(fs->bs, unload_cb, req);
622 }
623 
624 static struct spdk_file *
625 fs_find_file(struct spdk_filesystem *fs, const char *name)
626 {
627 	struct spdk_file *file;
628 
629 	TAILQ_FOREACH(file, &fs->files, tailq) {
630 		if (!strncmp(name, file->name, SPDK_FILE_NAME_MAX)) {
631 			return file;
632 		}
633 	}
634 
635 	return NULL;
636 }
637 
638 void
639 spdk_fs_file_stat_async(struct spdk_filesystem *fs, const char *name,
640 			spdk_file_stat_op_complete cb_fn, void *cb_arg)
641 {
642 	struct spdk_file_stat stat;
643 	struct spdk_file *f = NULL;
644 
645 	if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
646 		cb_fn(cb_arg, NULL, -ENAMETOOLONG);
647 		return;
648 	}
649 
650 	f = fs_find_file(fs, name);
651 	if (f != NULL) {
652 		stat.blobid = f->blobid;
653 		stat.size = f->length;
654 		cb_fn(cb_arg, &stat, 0);
655 		return;
656 	}
657 
658 	cb_fn(cb_arg, NULL, -ENOENT);
659 }
660 
661 static void
662 __copy_stat(void *arg, struct spdk_file_stat *stat, int fserrno)
663 {
664 	struct spdk_fs_request *req = arg;
665 	struct spdk_fs_cb_args *args = &req->args;
666 
667 	args->rc = fserrno;
668 	if (fserrno == 0) {
669 		memcpy(args->arg, stat, sizeof(*stat));
670 	}
671 	sem_post(args->sem);
672 }
673 
674 static void
675 __file_stat(void *arg)
676 {
677 	struct spdk_fs_request *req = arg;
678 	struct spdk_fs_cb_args *args = &req->args;
679 
680 	spdk_fs_file_stat_async(args->fs, args->op.stat.name,
681 				args->fn.stat_op, req);
682 }
683 
684 int
685 spdk_fs_file_stat(struct spdk_filesystem *fs, struct spdk_io_channel *_channel,
686 		  const char *name, struct spdk_file_stat *stat)
687 {
688 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
689 	struct spdk_fs_request *req;
690 	int rc;
691 
692 	req = alloc_fs_request(channel);
693 	assert(req != NULL);
694 
695 	req->args.fs = fs;
696 	req->args.op.stat.name = name;
697 	req->args.fn.stat_op = __copy_stat;
698 	req->args.arg = stat;
699 	req->args.sem = &channel->sem;
700 	channel->send_request(__file_stat, req);
701 	sem_wait(&channel->sem);
702 
703 	rc = req->args.rc;
704 	free_fs_request(req);
705 
706 	return rc;
707 }
708 
709 static void
710 fs_create_blob_close_cb(void *ctx, int bserrno)
711 {
712 	struct spdk_fs_request *req = ctx;
713 	struct spdk_fs_cb_args *args = &req->args;
714 
715 	args->fn.file_op(args->arg, bserrno);
716 	free_fs_request(req);
717 }
718 
719 static void
720 fs_create_blob_open_cb(void *ctx, struct spdk_blob *blob, int bserrno)
721 {
722 	struct spdk_fs_request *req = ctx;
723 	struct spdk_fs_cb_args *args = &req->args;
724 	struct spdk_file *f = args->file;
725 	uint64_t length = 0;
726 
727 	f->blob = blob;
728 	spdk_bs_md_resize_blob(blob, 1);
729 	spdk_blob_md_set_xattr(blob, "name", f->name, strlen(f->name) + 1);
730 	spdk_blob_md_set_xattr(blob, "length", &length, sizeof(length));
731 
732 	spdk_bs_md_close_blob(&f->blob, fs_create_blob_close_cb, args);
733 }
734 
735 static void
736 fs_create_blob_create_cb(void *ctx, spdk_blob_id blobid, int bserrno)
737 {
738 	struct spdk_fs_request *req = ctx;
739 	struct spdk_fs_cb_args *args = &req->args;
740 	struct spdk_file *f = args->file;
741 
742 	f->blobid = blobid;
743 	spdk_bs_md_open_blob(f->fs->bs, blobid, fs_create_blob_open_cb, req);
744 }
745 
746 void
747 spdk_fs_create_file_async(struct spdk_filesystem *fs, const char *name,
748 			  spdk_file_op_complete cb_fn, void *cb_arg)
749 {
750 	struct spdk_file *file;
751 	struct spdk_fs_request *req;
752 	struct spdk_fs_cb_args *args;
753 
754 	if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
755 		cb_fn(cb_arg, -ENAMETOOLONG);
756 		return;
757 	}
758 
759 	file = fs_find_file(fs, name);
760 	if (file != NULL) {
761 		cb_fn(cb_arg, -EEXIST);
762 		return;
763 	}
764 
765 	file = file_alloc(fs);
766 	if (file == NULL) {
767 		cb_fn(cb_arg, -ENOMEM);
768 		return;
769 	}
770 
771 	req = alloc_fs_request(fs->md_target.md_fs_channel);
772 	if (req == NULL) {
773 		cb_fn(cb_arg, -ENOMEM);
774 		return;
775 	}
776 
777 	args = &req->args;
778 	args->file = file;
779 	args->fn.file_op = cb_fn;
780 	args->arg = cb_arg;
781 
782 	file->name = strdup(name);
783 	spdk_bs_md_create_blob(fs->bs, fs_create_blob_create_cb, args);
784 }
785 
786 static void
787 __fs_create_file_done(void *arg, int fserrno)
788 {
789 	struct spdk_fs_request *req = arg;
790 	struct spdk_fs_cb_args *args = &req->args;
791 
792 	args->rc = fserrno;
793 	sem_post(args->sem);
794 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s\n", args->op.create.name);
795 }
796 
797 static void
798 __fs_create_file(void *arg)
799 {
800 	struct spdk_fs_request *req = arg;
801 	struct spdk_fs_cb_args *args = &req->args;
802 
803 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s\n", args->op.create.name);
804 	spdk_fs_create_file_async(args->fs, args->op.create.name, __fs_create_file_done, req);
805 }
806 
807 int
808 spdk_fs_create_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel, const char *name)
809 {
810 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
811 	struct spdk_fs_request *req;
812 	struct spdk_fs_cb_args *args;
813 	int rc;
814 
815 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s\n", name);
816 
817 	req = alloc_fs_request(channel);
818 	assert(req != NULL);
819 
820 	args = &req->args;
821 	args->fs = fs;
822 	args->op.create.name = name;
823 	args->sem = &channel->sem;
824 	fs->send_request(__fs_create_file, req);
825 	sem_wait(&channel->sem);
826 	rc = args->rc;
827 	free_fs_request(req);
828 
829 	return rc;
830 }
831 
832 static void
833 fs_open_blob_done(void *ctx, struct spdk_blob *blob, int bserrno)
834 {
835 	struct spdk_fs_request *req = ctx;
836 	struct spdk_fs_cb_args *args = &req->args;
837 	struct spdk_file *f = args->file;
838 
839 	f->blob = blob;
840 	while (!TAILQ_EMPTY(&f->open_requests)) {
841 		req = TAILQ_FIRST(&f->open_requests);
842 		args = &req->args;
843 		TAILQ_REMOVE(&f->open_requests, req, args.op.open.tailq);
844 		args->fn.file_op_with_handle(args->arg, f, bserrno);
845 		free_fs_request(req);
846 	}
847 }
848 
849 static void
850 fs_open_blob_create_cb(void *ctx, int bserrno)
851 {
852 	struct spdk_fs_request *req = ctx;
853 	struct spdk_fs_cb_args *args = &req->args;
854 	struct spdk_file *file = args->file;
855 	struct spdk_filesystem *fs = args->fs;
856 
857 	if (file == NULL) {
858 		/*
859 		 * This is from an open with CREATE flag - the file
860 		 *  is now created so look it up in the file list for this
861 		 *  filesystem.
862 		 */
863 		file = fs_find_file(fs, args->op.open.name);
864 		assert(file != NULL);
865 		args->file = file;
866 	}
867 
868 	file->ref_count++;
869 	TAILQ_INSERT_TAIL(&file->open_requests, req, args.op.open.tailq);
870 	if (file->ref_count == 1) {
871 		assert(file->blob == NULL);
872 		spdk_bs_md_open_blob(fs->bs, file->blobid, fs_open_blob_done, req);
873 	} else if (file->blob != NULL) {
874 		fs_open_blob_done(req, file->blob, 0);
875 	} else {
876 		/*
877 		 * The blob open for this file is in progress due to a previous
878 		 *  open request.  When that open completes, it will invoke the
879 		 *  open callback for this request.
880 		 */
881 	}
882 }
883 
884 void
885 spdk_fs_open_file_async(struct spdk_filesystem *fs, const char *name, uint32_t flags,
886 			spdk_file_op_with_handle_complete cb_fn, void *cb_arg)
887 {
888 	struct spdk_file *f = NULL;
889 	struct spdk_fs_request *req;
890 	struct spdk_fs_cb_args *args;
891 
892 	if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
893 		cb_fn(cb_arg, NULL, -ENAMETOOLONG);
894 		return;
895 	}
896 
897 	f = fs_find_file(fs, name);
898 	if (f == NULL && !(flags & SPDK_BLOBFS_OPEN_CREATE)) {
899 		cb_fn(cb_arg, NULL, -ENOENT);
900 		return;
901 	}
902 
903 	req = alloc_fs_request(fs->md_target.md_fs_channel);
904 	if (req == NULL) {
905 		cb_fn(cb_arg, NULL, -ENOMEM);
906 		return;
907 	}
908 
909 	args = &req->args;
910 	args->fn.file_op_with_handle = cb_fn;
911 	args->arg = cb_arg;
912 	args->file = f;
913 	args->fs = fs;
914 	args->op.open.name = name;
915 
916 	if (f == NULL) {
917 		spdk_fs_create_file_async(fs, name, fs_open_blob_create_cb, req);
918 	} else {
919 		fs_open_blob_create_cb(req, 0);
920 	}
921 }
922 
923 static void
924 __fs_open_file_done(void *arg, struct spdk_file *file, int bserrno)
925 {
926 	struct spdk_fs_request *req = arg;
927 	struct spdk_fs_cb_args *args = &req->args;
928 
929 	args->file = file;
930 	args->rc = bserrno;
931 	sem_post(args->sem);
932 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s\n", args->op.open.name);
933 }
934 
935 static void
936 __fs_open_file(void *arg)
937 {
938 	struct spdk_fs_request *req = arg;
939 	struct spdk_fs_cb_args *args = &req->args;
940 
941 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s\n", args->op.open.name);
942 	spdk_fs_open_file_async(args->fs, args->op.open.name, args->op.open.flags,
943 				__fs_open_file_done, req);
944 }
945 
946 int
947 spdk_fs_open_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel,
948 		  const char *name, uint32_t flags, struct spdk_file **file)
949 {
950 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
951 	struct spdk_fs_request *req;
952 	struct spdk_fs_cb_args *args;
953 	int rc;
954 
955 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s\n", name);
956 
957 	req = alloc_fs_request(channel);
958 	assert(req != NULL);
959 
960 	args = &req->args;
961 	args->fs = fs;
962 	args->op.open.name = name;
963 	args->op.open.flags = flags;
964 	args->sem = &channel->sem;
965 	fs->send_request(__fs_open_file, req);
966 	sem_wait(&channel->sem);
967 	rc = args->rc;
968 	if (rc == 0) {
969 		*file = args->file;
970 	} else {
971 		*file = NULL;
972 	}
973 	free_fs_request(req);
974 
975 	return rc;
976 }
977 
978 static void
979 fs_rename_blob_close_cb(void *ctx, int bserrno)
980 {
981 	struct spdk_fs_request *req = ctx;
982 	struct spdk_fs_cb_args *args = &req->args;
983 
984 	args->fn.fs_op(args->arg, bserrno);
985 	free_fs_request(req);
986 }
987 
988 static void
989 fs_rename_blob_open_cb(void *ctx, struct spdk_blob *blob, int bserrno)
990 {
991 	struct spdk_fs_request *req = ctx;
992 	struct spdk_fs_cb_args *args = &req->args;
993 	struct spdk_file *f = args->file;
994 	const char *new_name = args->op.rename.new_name;
995 
996 	f->blob = blob;
997 	spdk_blob_md_set_xattr(blob, "name", new_name, strlen(new_name) + 1);
998 	spdk_bs_md_close_blob(&f->blob, fs_rename_blob_close_cb, req);
999 }
1000 
1001 static void
1002 __spdk_fs_md_rename_file(struct spdk_fs_request *req)
1003 {
1004 	struct spdk_fs_cb_args *args = &req->args;
1005 	struct spdk_file *f;
1006 
1007 	f = fs_find_file(args->fs, args->op.rename.old_name);
1008 	if (f == NULL) {
1009 		args->fn.fs_op(args->arg, -ENOENT);
1010 		free_fs_request(req);
1011 		return;
1012 	}
1013 
1014 	free(f->name);
1015 	f->name = strdup(args->op.rename.new_name);
1016 	args->file = f;
1017 	spdk_bs_md_open_blob(args->fs->bs, f->blobid, fs_rename_blob_open_cb, req);
1018 }
1019 
1020 static void
1021 fs_rename_delete_done(void *arg, int fserrno)
1022 {
1023 	__spdk_fs_md_rename_file(arg);
1024 }
1025 
1026 void
1027 spdk_fs_rename_file_async(struct spdk_filesystem *fs,
1028 			  const char *old_name, const char *new_name,
1029 			  spdk_file_op_complete cb_fn, void *cb_arg)
1030 {
1031 	struct spdk_file *f;
1032 	struct spdk_fs_request *req;
1033 	struct spdk_fs_cb_args *args;
1034 
1035 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "old=%s new=%s\n", old_name, new_name);
1036 	if (strnlen(new_name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
1037 		cb_fn(cb_arg, -ENAMETOOLONG);
1038 		return;
1039 	}
1040 
1041 	req = alloc_fs_request(fs->md_target.md_fs_channel);
1042 	if (req == NULL) {
1043 		cb_fn(cb_arg, -ENOMEM);
1044 		return;
1045 	}
1046 
1047 	args = &req->args;
1048 	args->fn.fs_op = cb_fn;
1049 	args->fs = fs;
1050 	args->arg = cb_arg;
1051 	args->op.rename.old_name = old_name;
1052 	args->op.rename.new_name = new_name;
1053 
1054 	f = fs_find_file(fs, new_name);
1055 	if (f == NULL) {
1056 		__spdk_fs_md_rename_file(req);
1057 		return;
1058 	}
1059 
1060 	/*
1061 	 * The rename overwrites an existing file.  So delete the existing file, then
1062 	 *  do the actual rename.
1063 	 */
1064 	spdk_fs_delete_file_async(fs, new_name, fs_rename_delete_done, req);
1065 }
1066 
1067 static void
1068 __fs_rename_file_done(void *arg, int fserrno)
1069 {
1070 	struct spdk_fs_request *req = arg;
1071 	struct spdk_fs_cb_args *args = &req->args;
1072 
1073 	args->rc = fserrno;
1074 	sem_post(args->sem);
1075 }
1076 
1077 static void
1078 __fs_rename_file(void *arg)
1079 {
1080 	struct spdk_fs_request *req = arg;
1081 	struct spdk_fs_cb_args *args = &req->args;
1082 
1083 	spdk_fs_rename_file_async(args->fs, args->op.rename.old_name, args->op.rename.new_name,
1084 				  __fs_rename_file_done, req);
1085 }
1086 
1087 int
1088 spdk_fs_rename_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel,
1089 		    const char *old_name, const char *new_name)
1090 {
1091 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1092 	struct spdk_fs_request *req;
1093 	struct spdk_fs_cb_args *args;
1094 	int rc;
1095 
1096 	req = alloc_fs_request(channel);
1097 	assert(req != NULL);
1098 
1099 	args = &req->args;
1100 
1101 	args->fs = fs;
1102 	args->op.rename.old_name = old_name;
1103 	args->op.rename.new_name = new_name;
1104 	args->sem = &channel->sem;
1105 	fs->send_request(__fs_rename_file, req);
1106 	sem_wait(&channel->sem);
1107 	rc = args->rc;
1108 	free_fs_request(req);
1109 	return rc;
1110 }
1111 
1112 static void
1113 blob_delete_cb(void *ctx, int bserrno)
1114 {
1115 	struct spdk_fs_request *req = ctx;
1116 	struct spdk_fs_cb_args *args = &req->args;
1117 
1118 	args->fn.file_op(args->arg, bserrno);
1119 	free_fs_request(req);
1120 }
1121 
1122 void
1123 spdk_fs_delete_file_async(struct spdk_filesystem *fs, const char *name,
1124 			  spdk_file_op_complete cb_fn, void *cb_arg)
1125 {
1126 	struct spdk_file *f;
1127 	spdk_blob_id blobid;
1128 	struct spdk_fs_request *req;
1129 	struct spdk_fs_cb_args *args;
1130 
1131 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s\n", name);
1132 
1133 	if (strnlen(name, SPDK_FILE_NAME_MAX + 1) == SPDK_FILE_NAME_MAX + 1) {
1134 		cb_fn(cb_arg, -ENAMETOOLONG);
1135 		return;
1136 	}
1137 
1138 	f = fs_find_file(fs, name);
1139 	if (f == NULL) {
1140 		cb_fn(cb_arg, -ENOENT);
1141 		return;
1142 	}
1143 
1144 	if (f->ref_count > 0) {
1145 		/* For now, do not allow deleting files with open references. */
1146 		cb_fn(cb_arg, -EBUSY);
1147 		return;
1148 	}
1149 
1150 	req = alloc_fs_request(fs->md_target.md_fs_channel);
1151 	if (req == NULL) {
1152 		cb_fn(cb_arg, -ENOMEM);
1153 		return;
1154 	}
1155 
1156 	TAILQ_REMOVE(&fs->files, f, tailq);
1157 
1158 	cache_free_buffers(f);
1159 
1160 	blobid = f->blobid;
1161 
1162 	free(f->name);
1163 	free(f->tree);
1164 	free(f);
1165 
1166 	args = &req->args;
1167 	args->fn.file_op = cb_fn;
1168 	args->arg = cb_arg;
1169 	spdk_bs_md_delete_blob(fs->bs, blobid, blob_delete_cb, req);
1170 }
1171 
1172 static void
1173 __fs_delete_file_done(void *arg, int fserrno)
1174 {
1175 	struct spdk_fs_request *req = arg;
1176 	struct spdk_fs_cb_args *args = &req->args;
1177 
1178 	args->rc = fserrno;
1179 	sem_post(args->sem);
1180 }
1181 
1182 static void
1183 __fs_delete_file(void *arg)
1184 {
1185 	struct spdk_fs_request *req = arg;
1186 	struct spdk_fs_cb_args *args = &req->args;
1187 
1188 	spdk_fs_delete_file_async(args->fs, args->op.delete.name, __fs_delete_file_done, req);
1189 }
1190 
1191 int
1192 spdk_fs_delete_file(struct spdk_filesystem *fs, struct spdk_io_channel *_channel,
1193 		    const char *name)
1194 {
1195 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1196 	struct spdk_fs_request *req;
1197 	struct spdk_fs_cb_args *args;
1198 	int rc;
1199 
1200 	req = alloc_fs_request(channel);
1201 	assert(req != NULL);
1202 
1203 	args = &req->args;
1204 	args->fs = fs;
1205 	args->op.delete.name = name;
1206 	args->sem = &channel->sem;
1207 	fs->send_request(__fs_delete_file, req);
1208 	sem_wait(&channel->sem);
1209 	rc = args->rc;
1210 	free_fs_request(req);
1211 
1212 	return rc;
1213 }
1214 
1215 spdk_fs_iter
1216 spdk_fs_iter_first(struct spdk_filesystem *fs)
1217 {
1218 	struct spdk_file *f;
1219 
1220 	f = TAILQ_FIRST(&fs->files);
1221 	return f;
1222 }
1223 
1224 spdk_fs_iter
1225 spdk_fs_iter_next(spdk_fs_iter iter)
1226 {
1227 	struct spdk_file *f = iter;
1228 
1229 	if (f == NULL) {
1230 		return NULL;
1231 	}
1232 
1233 	f = TAILQ_NEXT(f, tailq);
1234 	return f;
1235 }
1236 
1237 const char *
1238 spdk_file_get_name(struct spdk_file *file)
1239 {
1240 	return file->name;
1241 }
1242 
1243 uint64_t
1244 spdk_file_get_length(struct spdk_file *file)
1245 {
1246 	assert(file != NULL);
1247 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s length=0x%jx\n", file->name, file->length);
1248 	return file->length;
1249 }
1250 
1251 static void
1252 fs_truncate_complete_cb(void *ctx, int bserrno)
1253 {
1254 	struct spdk_fs_request *req = ctx;
1255 	struct spdk_fs_cb_args *args = &req->args;
1256 
1257 	args->fn.file_op(args->arg, bserrno);
1258 	free_fs_request(req);
1259 }
1260 
1261 static uint64_t
1262 __bytes_to_clusters(uint64_t length, uint64_t cluster_sz)
1263 {
1264 	return (length + cluster_sz - 1) / cluster_sz;
1265 }
1266 
1267 void
1268 spdk_file_truncate_async(struct spdk_file *file, uint64_t length,
1269 			 spdk_file_op_complete cb_fn, void *cb_arg)
1270 {
1271 	struct spdk_filesystem *fs;
1272 	size_t num_clusters;
1273 	struct spdk_fs_request *req;
1274 	struct spdk_fs_cb_args *args;
1275 
1276 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s old=0x%jx new=0x%jx\n", file->name, file->length, length);
1277 	if (length == file->length) {
1278 		cb_fn(cb_arg, 0);
1279 		return;
1280 	}
1281 
1282 	req = alloc_fs_request(file->fs->md_target.md_fs_channel);
1283 	if (req == NULL) {
1284 		cb_fn(cb_arg, -ENOMEM);
1285 		return;
1286 	}
1287 
1288 	args = &req->args;
1289 	args->fn.file_op = cb_fn;
1290 	args->arg = cb_arg;
1291 	args->file = file;
1292 	fs = file->fs;
1293 
1294 	num_clusters = __bytes_to_clusters(length, fs->bs_opts.cluster_sz);
1295 
1296 	spdk_bs_md_resize_blob(file->blob, num_clusters);
1297 	spdk_blob_md_set_xattr(file->blob, "length", &length, sizeof(length));
1298 
1299 	file->length = length;
1300 	if (file->append_pos > file->length) {
1301 		file->append_pos = file->length;
1302 	}
1303 
1304 	spdk_bs_md_sync_blob(file->blob, fs_truncate_complete_cb, args);
1305 }
1306 
1307 static void
1308 __truncate(void *arg)
1309 {
1310 	struct spdk_fs_request *req = arg;
1311 	struct spdk_fs_cb_args *args = &req->args;
1312 
1313 	spdk_file_truncate_async(args->file, args->op.truncate.length,
1314 				 args->fn.file_op, args->arg);
1315 }
1316 
1317 void
1318 spdk_file_truncate(struct spdk_file *file, struct spdk_io_channel *_channel,
1319 		   uint64_t length)
1320 {
1321 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1322 	struct spdk_fs_request *req;
1323 	struct spdk_fs_cb_args *args;
1324 
1325 	req = alloc_fs_request(channel);
1326 	assert(req != NULL);
1327 
1328 	args = &req->args;
1329 
1330 	args->file = file;
1331 	args->op.truncate.length = length;
1332 	args->fn.file_op = __sem_post;
1333 	args->arg = &channel->sem;
1334 
1335 	channel->send_request(__truncate, req);
1336 	sem_wait(&channel->sem);
1337 	free_fs_request(req);
1338 }
1339 
1340 static void
1341 __rw_done(void *ctx, int bserrno)
1342 {
1343 	struct spdk_fs_request *req = ctx;
1344 	struct spdk_fs_cb_args *args = &req->args;
1345 
1346 	spdk_dma_free(args->op.rw.pin_buf);
1347 	args->fn.file_op(args->arg, bserrno);
1348 	free_fs_request(req);
1349 }
1350 
1351 static void
1352 __read_done(void *ctx, int bserrno)
1353 {
1354 	struct spdk_fs_request *req = ctx;
1355 	struct spdk_fs_cb_args *args = &req->args;
1356 
1357 	if (args->op.rw.is_read) {
1358 		memcpy(args->op.rw.user_buf,
1359 		       args->op.rw.pin_buf + (args->op.rw.offset & 0xFFF),
1360 		       args->op.rw.length);
1361 		__rw_done(req, 0);
1362 	} else {
1363 		memcpy(args->op.rw.pin_buf + (args->op.rw.offset & 0xFFF),
1364 		       args->op.rw.user_buf,
1365 		       args->op.rw.length);
1366 		spdk_bs_io_write_blob(args->file->blob, args->op.rw.channel,
1367 				      args->op.rw.pin_buf,
1368 				      args->op.rw.start_page, args->op.rw.num_pages,
1369 				      __rw_done, req);
1370 	}
1371 }
1372 
1373 static void
1374 __do_blob_read(void *ctx, int fserrno)
1375 {
1376 	struct spdk_fs_request *req = ctx;
1377 	struct spdk_fs_cb_args *args = &req->args;
1378 
1379 	spdk_bs_io_read_blob(args->file->blob, args->op.rw.channel,
1380 			     args->op.rw.pin_buf,
1381 			     args->op.rw.start_page, args->op.rw.num_pages,
1382 			     __read_done, req);
1383 }
1384 
1385 static void
1386 __get_page_parameters(struct spdk_file *file, uint64_t offset, uint64_t length,
1387 		      uint64_t *start_page, uint32_t *page_size, uint64_t *num_pages)
1388 {
1389 	uint64_t end_page;
1390 
1391 	*page_size = spdk_bs_get_page_size(file->fs->bs);
1392 	*start_page = offset / *page_size;
1393 	end_page = (offset + length - 1) / *page_size;
1394 	*num_pages = (end_page - *start_page + 1);
1395 }
1396 
1397 static void
1398 __readwrite(struct spdk_file *file, struct spdk_io_channel *_channel,
1399 	    void *payload, uint64_t offset, uint64_t length,
1400 	    spdk_file_op_complete cb_fn, void *cb_arg, int is_read)
1401 {
1402 	struct spdk_fs_request *req;
1403 	struct spdk_fs_cb_args *args;
1404 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1405 	uint64_t start_page, num_pages, pin_buf_length;
1406 	uint32_t page_size;
1407 
1408 	if (is_read && offset + length > file->length) {
1409 		cb_fn(cb_arg, -EINVAL);
1410 		return;
1411 	}
1412 
1413 	req = alloc_fs_request(channel);
1414 	if (req == NULL) {
1415 		cb_fn(cb_arg, -ENOMEM);
1416 		return;
1417 	}
1418 
1419 	args = &req->args;
1420 	args->fn.file_op = cb_fn;
1421 	args->arg = cb_arg;
1422 	args->file = file;
1423 	args->op.rw.channel = channel->bs_channel;
1424 	args->op.rw.user_buf = payload;
1425 	args->op.rw.is_read = is_read;
1426 	args->op.rw.offset = offset;
1427 	args->op.rw.length = length;
1428 
1429 	__get_page_parameters(file, offset, length, &start_page, &page_size, &num_pages);
1430 	pin_buf_length = num_pages * page_size;
1431 	args->op.rw.pin_buf = spdk_dma_malloc(pin_buf_length, 4096, NULL);
1432 
1433 	args->op.rw.start_page = start_page;
1434 	args->op.rw.num_pages = num_pages;
1435 
1436 	if (!is_read && file->length < offset + length) {
1437 		spdk_file_truncate_async(file, offset + length, __do_blob_read, req);
1438 	} else {
1439 		__do_blob_read(req, 0);
1440 	}
1441 }
1442 
1443 void
1444 spdk_file_write_async(struct spdk_file *file, struct spdk_io_channel *channel,
1445 		      void *payload, uint64_t offset, uint64_t length,
1446 		      spdk_file_op_complete cb_fn, void *cb_arg)
1447 {
1448 	__readwrite(file, channel, payload, offset, length, cb_fn, cb_arg, 0);
1449 }
1450 
1451 void
1452 spdk_file_read_async(struct spdk_file *file, struct spdk_io_channel *channel,
1453 		     void *payload, uint64_t offset, uint64_t length,
1454 		     spdk_file_op_complete cb_fn, void *cb_arg)
1455 {
1456 	SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "file=%s offset=%jx length=%jx\n",
1457 		      file->name, offset, length);
1458 	__readwrite(file, channel, payload, offset, length, cb_fn, cb_arg, 1);
1459 }
1460 
1461 struct spdk_io_channel *
1462 spdk_fs_alloc_io_channel(struct spdk_filesystem *fs)
1463 {
1464 	struct spdk_io_channel *io_channel;
1465 	struct spdk_fs_channel *fs_channel;
1466 
1467 	io_channel = spdk_get_io_channel(&fs->io_target);
1468 	fs_channel = spdk_io_channel_get_ctx(io_channel);
1469 	fs_channel->bs_channel = spdk_bs_alloc_io_channel(fs->bs);
1470 	fs_channel->send_request = __send_request_direct;
1471 
1472 	return io_channel;
1473 }
1474 
1475 struct spdk_io_channel *
1476 spdk_fs_alloc_io_channel_sync(struct spdk_filesystem *fs)
1477 {
1478 	struct spdk_io_channel *io_channel;
1479 	struct spdk_fs_channel *fs_channel;
1480 
1481 	io_channel = spdk_get_io_channel(&fs->io_target);
1482 	fs_channel = spdk_io_channel_get_ctx(io_channel);
1483 	fs_channel->send_request = fs->send_request;
1484 	fs_channel->sync = 1;
1485 	pthread_spin_init(&fs_channel->lock, 0);
1486 
1487 	return io_channel;
1488 }
1489 
1490 void
1491 spdk_fs_free_io_channel(struct spdk_io_channel *channel)
1492 {
1493 	spdk_put_io_channel(channel);
1494 }
1495 
1496 void
1497 spdk_fs_set_cache_size(uint64_t size_in_mb)
1498 {
1499 	g_fs_cache_size = size_in_mb * 1024 * 1024;
1500 }
1501 
1502 uint64_t
1503 spdk_fs_get_cache_size(void)
1504 {
1505 	return g_fs_cache_size / (1024 * 1024);
1506 }
1507 
1508 static void __file_flush(void *_args);
1509 
1510 static void *
1511 alloc_cache_memory_buffer(struct spdk_file *context)
1512 {
1513 	struct spdk_file *file;
1514 	void *buf;
1515 
1516 	buf = spdk_mempool_get(g_cache_pool);
1517 	if (buf != NULL) {
1518 		return buf;
1519 	}
1520 
1521 	pthread_spin_lock(&g_caches_lock);
1522 	TAILQ_FOREACH(file, &g_caches, cache_tailq) {
1523 		if (!file->open_for_writing &&
1524 		    file->priority == SPDK_FILE_PRIORITY_LOW &&
1525 		    file != context) {
1526 			TAILQ_REMOVE(&g_caches, file, cache_tailq);
1527 			TAILQ_INSERT_TAIL(&g_caches, file, cache_tailq);
1528 			break;
1529 		}
1530 	}
1531 	pthread_spin_unlock(&g_caches_lock);
1532 	if (file != NULL) {
1533 		cache_free_buffers(file);
1534 		buf = spdk_mempool_get(g_cache_pool);
1535 		if (buf != NULL) {
1536 			return buf;
1537 		}
1538 	}
1539 
1540 	pthread_spin_lock(&g_caches_lock);
1541 	TAILQ_FOREACH(file, &g_caches, cache_tailq) {
1542 		if (!file->open_for_writing && file != context) {
1543 			TAILQ_REMOVE(&g_caches, file, cache_tailq);
1544 			TAILQ_INSERT_TAIL(&g_caches, file, cache_tailq);
1545 			break;
1546 		}
1547 	}
1548 	pthread_spin_unlock(&g_caches_lock);
1549 	if (file != NULL) {
1550 		cache_free_buffers(file);
1551 		buf = spdk_mempool_get(g_cache_pool);
1552 		if (buf != NULL) {
1553 			return buf;
1554 		}
1555 	}
1556 
1557 	pthread_spin_lock(&g_caches_lock);
1558 	TAILQ_FOREACH(file, &g_caches, cache_tailq) {
1559 		if (file != context) {
1560 			TAILQ_REMOVE(&g_caches, file, cache_tailq);
1561 			TAILQ_INSERT_TAIL(&g_caches, file, cache_tailq);
1562 			break;
1563 		}
1564 	}
1565 	pthread_spin_unlock(&g_caches_lock);
1566 	if (file != NULL) {
1567 		cache_free_buffers(file);
1568 		buf = spdk_mempool_get(g_cache_pool);
1569 		if (buf != NULL) {
1570 			return buf;
1571 		}
1572 	}
1573 
1574 	assert(false);
1575 	return NULL;
1576 }
1577 
1578 static struct cache_buffer *
1579 cache_insert_buffer(struct spdk_file *file, uint64_t offset)
1580 {
1581 	struct cache_buffer *buf;
1582 	int count = 0;
1583 
1584 	buf = calloc(1, sizeof(*buf));
1585 	if (buf == NULL) {
1586 		SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "calloc failed\n");
1587 		return NULL;
1588 	}
1589 
1590 	buf->buf = alloc_cache_memory_buffer(file);
1591 	if (buf->buf == NULL) {
1592 		while (buf->buf == NULL) {
1593 			/*
1594 			 * TODO: alloc_cache_memory_buffer() should eventually free
1595 			 *  some buffers.  Need a more sophisticated check here, instead
1596 			 *  of just bailing if 100 tries does not result in getting a
1597 			 *  free buffer.  This will involve using the sync channel's
1598 			 *  semaphore to block until a buffer becomes available.
1599 			 */
1600 			if (count++ == 100) {
1601 				SPDK_ERRLOG("could not allocate cache buffer\n");
1602 				assert(false);
1603 				free(buf);
1604 				return NULL;
1605 			}
1606 			buf->buf = alloc_cache_memory_buffer(file);
1607 		}
1608 	}
1609 
1610 	buf->buf_size = CACHE_BUFFER_SIZE;
1611 	buf->offset = offset;
1612 
1613 	pthread_spin_lock(&g_caches_lock);
1614 	if (file->tree->present_mask == 0) {
1615 		TAILQ_INSERT_TAIL(&g_caches, file, cache_tailq);
1616 	}
1617 	file->tree = spdk_tree_insert_buffer(file->tree, buf);
1618 	pthread_spin_unlock(&g_caches_lock);
1619 
1620 	return buf;
1621 }
1622 
1623 static struct cache_buffer *
1624 cache_append_buffer(struct spdk_file *file)
1625 {
1626 	struct cache_buffer *last;
1627 
1628 	assert(file->last == NULL || file->last->bytes_filled == file->last->buf_size);
1629 	assert((file->append_pos % CACHE_BUFFER_SIZE) == 0);
1630 
1631 	last = cache_insert_buffer(file, file->append_pos);
1632 	if (last == NULL) {
1633 		SPDK_TRACELOG(SPDK_TRACE_BLOBFS, "cache_insert_buffer failed\n");
1634 		return NULL;
1635 	}
1636 
1637 	if (file->last != NULL) {
1638 		file->last->next = last;
1639 	}
1640 	file->last = last;
1641 
1642 	return last;
1643 }
1644 
1645 static void
1646 __wake_caller(struct spdk_fs_cb_args *args)
1647 {
1648 	sem_post(args->sem);
1649 }
1650 
1651 static void __check_sync_reqs(struct spdk_file *file);
1652 
1653 static void
1654 __file_cache_finish_sync(struct spdk_file *file)
1655 {
1656 	struct spdk_fs_request *sync_req;
1657 	struct spdk_fs_cb_args *sync_args;
1658 
1659 	pthread_spin_lock(&file->lock);
1660 	sync_req = TAILQ_FIRST(&file->sync_requests);
1661 	sync_args = &sync_req->args;
1662 	assert(sync_args->op.sync.offset <= file->length_flushed);
1663 	BLOBFS_TRACE(file, "sync done offset=%jx\n", sync_args->op.sync.offset);
1664 	TAILQ_REMOVE(&file->sync_requests, sync_req, args.op.sync.tailq);
1665 	pthread_spin_unlock(&file->lock);
1666 
1667 	sync_args->fn.file_op(sync_args->arg, 0);
1668 	__check_sync_reqs(file);
1669 
1670 	pthread_spin_lock(&file->lock);
1671 	free_fs_request(sync_req);
1672 	pthread_spin_unlock(&file->lock);
1673 }
1674 
1675 static void
1676 __file_cache_finish_sync_bs_cb(void *ctx, int bserrno)
1677 {
1678 	struct spdk_file *file = ctx;
1679 
1680 	__file_cache_finish_sync(file);
1681 }
1682 
1683 static void
1684 __free_args(struct spdk_fs_cb_args *args)
1685 {
1686 	struct spdk_fs_request *req;
1687 
1688 	if (!args->from_request) {
1689 		free(args);
1690 	} else {
1691 		/* Depends on args being at the start of the spdk_fs_request structure. */
1692 		req = (struct spdk_fs_request *)args;
1693 		free_fs_request(req);
1694 	}
1695 }
1696 
1697 static void
1698 __check_sync_reqs(struct spdk_file *file)
1699 {
1700 	struct spdk_fs_request *sync_req;
1701 
1702 	pthread_spin_lock(&file->lock);
1703 
1704 	TAILQ_FOREACH(sync_req, &file->sync_requests, args.op.sync.tailq) {
1705 		if (sync_req->args.op.sync.offset <= file->length_flushed) {
1706 			break;
1707 		}
1708 	}
1709 
1710 	if (sync_req != NULL && !sync_req->args.op.sync.xattr_in_progress) {
1711 		BLOBFS_TRACE(file, "set xattr length 0x%jx\n", file->length_flushed);
1712 		sync_req->args.op.sync.xattr_in_progress = true;
1713 		spdk_blob_md_set_xattr(file->blob, "length", &file->length_flushed,
1714 				       sizeof(file->length_flushed));
1715 
1716 		pthread_spin_unlock(&file->lock);
1717 		spdk_bs_md_sync_blob(file->blob, __file_cache_finish_sync_bs_cb, file);
1718 	} else {
1719 		pthread_spin_unlock(&file->lock);
1720 	}
1721 }
1722 
1723 static void
1724 __file_flush_done(void *arg, int bserrno)
1725 {
1726 	struct spdk_fs_cb_args *args = arg;
1727 	struct spdk_file *file = args->file;
1728 	struct cache_buffer *next = args->op.flush.cache_buffer;
1729 
1730 	BLOBFS_TRACE(file, "length=%jx\n", args->op.flush.length);
1731 
1732 	pthread_spin_lock(&file->lock);
1733 	next->in_progress = false;
1734 	next->bytes_flushed += args->op.flush.length;
1735 	file->length_flushed += args->op.flush.length;
1736 	if (file->length_flushed > file->length) {
1737 		file->length = file->length_flushed;
1738 	}
1739 	if (next->bytes_flushed == next->buf_size) {
1740 		BLOBFS_TRACE(file, "write buffer fully flushed 0x%jx\n", file->length_flushed);
1741 		next = spdk_tree_find_buffer(file->tree, file->length_flushed);
1742 	}
1743 
1744 	/*
1745 	 * Assert that there is no cached data that extends past the end of the underlying
1746 	 *  blob.
1747 	 */
1748 	assert(next == NULL || next->offset < __file_get_blob_size(file) ||
1749 	       next->bytes_filled == 0);
1750 
1751 	pthread_spin_unlock(&file->lock);
1752 
1753 	__check_sync_reqs(file);
1754 
1755 	__file_flush(args);
1756 }
1757 
1758 static void
1759 __file_flush(void *_args)
1760 {
1761 	struct spdk_fs_cb_args *args = _args;
1762 	struct spdk_file *file = args->file;
1763 	struct cache_buffer *next;
1764 	uint64_t offset, length, start_page, num_pages;
1765 	uint32_t page_size;
1766 
1767 	pthread_spin_lock(&file->lock);
1768 	next = spdk_tree_find_buffer(file->tree, file->length_flushed);
1769 	if (next == NULL || next->in_progress) {
1770 		/*
1771 		 * There is either no data to flush, or a flush I/O is already in
1772 		 *  progress.  So return immediately - if a flush I/O is in
1773 		 *  progress we will flush more data after that is completed.
1774 		 */
1775 		__free_args(args);
1776 		pthread_spin_unlock(&file->lock);
1777 		return;
1778 	}
1779 
1780 	offset = next->offset + next->bytes_flushed;
1781 	length = next->bytes_filled - next->bytes_flushed;
1782 	if (length == 0) {
1783 		__free_args(args);
1784 		pthread_spin_unlock(&file->lock);
1785 		return;
1786 	}
1787 	args->op.flush.length = length;
1788 	args->op.flush.cache_buffer = next;
1789 
1790 	__get_page_parameters(file, offset, length, &start_page, &page_size, &num_pages);
1791 
1792 	next->in_progress = true;
1793 	BLOBFS_TRACE(file, "offset=%jx length=%jx page start=%jx num=%jx\n",
1794 		     offset, length, start_page, num_pages);
1795 	pthread_spin_unlock(&file->lock);
1796 	spdk_bs_io_write_blob(file->blob, file->fs->sync_target.sync_fs_channel->bs_channel,
1797 			      next->buf + (start_page * page_size) - next->offset,
1798 			      start_page, num_pages,
1799 			      __file_flush_done, args);
1800 }
1801 
1802 static void
1803 __file_extend_done(void *arg, int bserrno)
1804 {
1805 	struct spdk_fs_cb_args *args = arg;
1806 
1807 	__wake_caller(args);
1808 }
1809 
1810 static void
1811 __file_extend_blob(void *_args)
1812 {
1813 	struct spdk_fs_cb_args *args = _args;
1814 	struct spdk_file *file = args->file;
1815 
1816 	spdk_bs_md_resize_blob(file->blob, args->op.resize.num_clusters);
1817 
1818 	spdk_bs_md_sync_blob(file->blob, __file_extend_done, args);
1819 }
1820 
1821 static void
1822 __rw_from_file_done(void *arg, int bserrno)
1823 {
1824 	struct spdk_fs_cb_args *args = arg;
1825 
1826 	__wake_caller(args);
1827 	__free_args(args);
1828 }
1829 
1830 static void
1831 __rw_from_file(void *_args)
1832 {
1833 	struct spdk_fs_cb_args *args = _args;
1834 	struct spdk_file *file = args->file;
1835 
1836 	if (args->op.rw.is_read) {
1837 		spdk_file_read_async(file, file->fs->sync_target.sync_io_channel, args->op.rw.user_buf,
1838 				     args->op.rw.offset, args->op.rw.length,
1839 				     __rw_from_file_done, args);
1840 	} else {
1841 		spdk_file_write_async(file, file->fs->sync_target.sync_io_channel, args->op.rw.user_buf,
1842 				      args->op.rw.offset, args->op.rw.length,
1843 				      __rw_from_file_done, args);
1844 	}
1845 }
1846 
1847 static int
1848 __send_rw_from_file(struct spdk_file *file, sem_t *sem, void *payload,
1849 		    uint64_t offset, uint64_t length, bool is_read)
1850 {
1851 	struct spdk_fs_cb_args *args;
1852 
1853 	args = calloc(1, sizeof(*args));
1854 	if (args == NULL) {
1855 		sem_post(sem);
1856 		return -ENOMEM;
1857 	}
1858 
1859 	args->file = file;
1860 	args->sem = sem;
1861 	args->op.rw.user_buf = payload;
1862 	args->op.rw.offset = offset;
1863 	args->op.rw.length = length;
1864 	args->op.rw.is_read = is_read;
1865 	file->fs->send_request(__rw_from_file, args);
1866 	return 0;
1867 }
1868 
1869 int
1870 spdk_file_write(struct spdk_file *file, struct spdk_io_channel *_channel,
1871 		void *payload, uint64_t offset, uint64_t length)
1872 {
1873 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
1874 	struct spdk_fs_cb_args *args;
1875 	uint64_t rem_length, copy, blob_size, cluster_sz;
1876 	uint32_t cache_buffers_filled = 0;
1877 	uint8_t *cur_payload;
1878 	struct cache_buffer *last;
1879 
1880 	BLOBFS_TRACE_RW(file, "offset=%jx length=%jx\n", offset, length);
1881 
1882 	if (length == 0) {
1883 		return 0;
1884 	}
1885 
1886 	if (offset != file->append_pos) {
1887 		BLOBFS_TRACE(file, " error offset=%jx append_pos=%jx\n", offset, file->append_pos);
1888 		return -EINVAL;
1889 	}
1890 
1891 	pthread_spin_lock(&file->lock);
1892 	file->open_for_writing = true;
1893 
1894 	if (file->last == NULL) {
1895 		if (file->append_pos % CACHE_BUFFER_SIZE == 0) {
1896 			cache_append_buffer(file);
1897 		} else {
1898 			int rc;
1899 
1900 			file->append_pos += length;
1901 			pthread_spin_unlock(&file->lock);
1902 			rc = __send_rw_from_file(file, &channel->sem, payload,
1903 						 offset, length, false);
1904 			sem_wait(&channel->sem);
1905 			return rc;
1906 		}
1907 	}
1908 
1909 	blob_size = __file_get_blob_size(file);
1910 
1911 	if ((offset + length) > blob_size) {
1912 		struct spdk_fs_cb_args extend_args = {};
1913 
1914 		cluster_sz = file->fs->bs_opts.cluster_sz;
1915 		extend_args.sem = &channel->sem;
1916 		extend_args.op.resize.num_clusters = __bytes_to_clusters((offset + length), cluster_sz);
1917 		extend_args.file = file;
1918 		BLOBFS_TRACE(file, "start resize to %u clusters\n", extend_args.op.resize.num_clusters);
1919 		pthread_spin_unlock(&file->lock);
1920 		file->fs->send_request(__file_extend_blob, &extend_args);
1921 		sem_wait(&channel->sem);
1922 	}
1923 
1924 	last = file->last;
1925 	rem_length = length;
1926 	cur_payload = payload;
1927 	while (rem_length > 0) {
1928 		copy = last->buf_size - last->bytes_filled;
1929 		if (copy > rem_length) {
1930 			copy = rem_length;
1931 		}
1932 		BLOBFS_TRACE_RW(file, "  fill offset=%jx length=%jx\n", file->append_pos, copy);
1933 		memcpy(&last->buf[last->bytes_filled], cur_payload, copy);
1934 		file->append_pos += copy;
1935 		if (file->length < file->append_pos) {
1936 			file->length = file->append_pos;
1937 		}
1938 		cur_payload += copy;
1939 		last->bytes_filled += copy;
1940 		rem_length -= copy;
1941 		if (last->bytes_filled == last->buf_size) {
1942 			cache_buffers_filled++;
1943 			last = cache_append_buffer(file);
1944 			if (last == NULL) {
1945 				BLOBFS_TRACE(file, "nomem\n");
1946 				pthread_spin_unlock(&file->lock);
1947 				return -ENOMEM;
1948 			}
1949 		}
1950 	}
1951 
1952 	if (cache_buffers_filled == 0) {
1953 		pthread_spin_unlock(&file->lock);
1954 		return 0;
1955 	}
1956 
1957 	args = calloc(1, sizeof(*args));
1958 	if (args == NULL) {
1959 		pthread_spin_unlock(&file->lock);
1960 		return -ENOMEM;
1961 	}
1962 
1963 	args->file = file;
1964 	file->fs->send_request(__file_flush, args);
1965 	pthread_spin_unlock(&file->lock);
1966 	return 0;
1967 }
1968 
1969 static void
1970 __readahead_done(void *arg, int bserrno)
1971 {
1972 	struct spdk_fs_cb_args *args = arg;
1973 	struct cache_buffer *cache_buffer = args->op.readahead.cache_buffer;
1974 	struct spdk_file *file = args->file;
1975 
1976 	BLOBFS_TRACE(file, "offset=%jx\n", cache_buffer->offset);
1977 
1978 	pthread_spin_lock(&file->lock);
1979 	cache_buffer->bytes_filled = args->op.readahead.length;
1980 	cache_buffer->bytes_flushed = args->op.readahead.length;
1981 	cache_buffer->in_progress = false;
1982 	pthread_spin_unlock(&file->lock);
1983 
1984 	__free_args(args);
1985 }
1986 
1987 static void
1988 __readahead(void *_args)
1989 {
1990 	struct spdk_fs_cb_args *args = _args;
1991 	struct spdk_file *file = args->file;
1992 	uint64_t offset, length, start_page, num_pages;
1993 	uint32_t page_size;
1994 
1995 	offset = args->op.readahead.offset;
1996 	length = args->op.readahead.length;
1997 	assert(length > 0);
1998 
1999 	__get_page_parameters(file, offset, length, &start_page, &page_size, &num_pages);
2000 
2001 	BLOBFS_TRACE(file, "offset=%jx length=%jx page start=%jx num=%jx\n",
2002 		     offset, length, start_page, num_pages);
2003 	spdk_bs_io_read_blob(file->blob, file->fs->sync_target.sync_fs_channel->bs_channel,
2004 			     args->op.readahead.cache_buffer->buf,
2005 			     start_page, num_pages,
2006 			     __readahead_done, args);
2007 }
2008 
2009 static uint64_t
2010 __next_cache_buffer_offset(uint64_t offset)
2011 {
2012 	return (offset + CACHE_BUFFER_SIZE) & ~(CACHE_TREE_LEVEL_MASK(0));
2013 }
2014 
2015 static void
2016 check_readahead(struct spdk_file *file, uint64_t offset)
2017 {
2018 	struct spdk_fs_cb_args *args;
2019 
2020 	offset = __next_cache_buffer_offset(offset);
2021 	if (spdk_tree_find_buffer(file->tree, offset) != NULL || file->length <= offset) {
2022 		return;
2023 	}
2024 
2025 	args = calloc(1, sizeof(*args));
2026 	if (args == NULL) {
2027 		return;
2028 	}
2029 
2030 	BLOBFS_TRACE(file, "offset=%jx\n", offset);
2031 
2032 	args->file = file;
2033 	args->op.readahead.offset = offset;
2034 	args->op.readahead.cache_buffer = cache_insert_buffer(file, offset);
2035 	args->op.readahead.cache_buffer->in_progress = true;
2036 	if (file->length < (offset + CACHE_BUFFER_SIZE)) {
2037 		args->op.readahead.length = file->length & (CACHE_BUFFER_SIZE - 1);
2038 	} else {
2039 		args->op.readahead.length = CACHE_BUFFER_SIZE;
2040 	}
2041 	file->fs->send_request(__readahead, args);
2042 }
2043 
2044 static int
2045 __file_read(struct spdk_file *file, void *payload, uint64_t offset, uint64_t length, sem_t *sem)
2046 {
2047 	struct cache_buffer *buf;
2048 	int rc;
2049 
2050 	buf = spdk_tree_find_filled_buffer(file->tree, offset);
2051 	if (buf == NULL) {
2052 		pthread_spin_unlock(&file->lock);
2053 		rc = __send_rw_from_file(file, sem, payload, offset, length, true);
2054 		pthread_spin_lock(&file->lock);
2055 		return rc;
2056 	}
2057 
2058 	if ((offset + length) > (buf->offset + buf->bytes_filled)) {
2059 		length = buf->offset + buf->bytes_filled - offset;
2060 	}
2061 	BLOBFS_TRACE(file, "read %p offset=%ju length=%ju\n", payload, offset, length);
2062 	memcpy(payload, &buf->buf[offset - buf->offset], length);
2063 	if ((offset + length) % CACHE_BUFFER_SIZE == 0) {
2064 		pthread_spin_lock(&g_caches_lock);
2065 		spdk_tree_remove_buffer(file->tree, buf);
2066 		if (file->tree->present_mask == 0) {
2067 			TAILQ_REMOVE(&g_caches, file, cache_tailq);
2068 		}
2069 		pthread_spin_unlock(&g_caches_lock);
2070 	}
2071 
2072 	sem_post(sem);
2073 	return 0;
2074 }
2075 
2076 int64_t
2077 spdk_file_read(struct spdk_file *file, struct spdk_io_channel *_channel,
2078 	       void *payload, uint64_t offset, uint64_t length)
2079 {
2080 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
2081 	uint64_t final_offset, final_length;
2082 	uint32_t sub_reads = 0;
2083 	int rc = 0;
2084 
2085 	pthread_spin_lock(&file->lock);
2086 
2087 	BLOBFS_TRACE_RW(file, "offset=%ju length=%ju\n", offset, length);
2088 
2089 	file->open_for_writing = false;
2090 
2091 	if (length == 0 || offset >= file->append_pos) {
2092 		pthread_spin_unlock(&file->lock);
2093 		return 0;
2094 	}
2095 
2096 	if (offset + length > file->append_pos) {
2097 		length = file->append_pos - offset;
2098 	}
2099 
2100 	if (offset != file->next_seq_offset) {
2101 		file->seq_byte_count = 0;
2102 	}
2103 	file->seq_byte_count += length;
2104 	file->next_seq_offset = offset + length;
2105 	if (file->seq_byte_count >= CACHE_READAHEAD_THRESHOLD) {
2106 		check_readahead(file, offset);
2107 		check_readahead(file, offset + CACHE_BUFFER_SIZE);
2108 	}
2109 
2110 	final_length = 0;
2111 	final_offset = offset + length;
2112 	while (offset < final_offset) {
2113 		length = NEXT_CACHE_BUFFER_OFFSET(offset) - offset;
2114 		if (length > (final_offset - offset)) {
2115 			length = final_offset - offset;
2116 		}
2117 		rc = __file_read(file, payload, offset, length, &channel->sem);
2118 		if (rc == 0) {
2119 			final_length += length;
2120 		} else {
2121 			break;
2122 		}
2123 		payload += length;
2124 		offset += length;
2125 		sub_reads++;
2126 	}
2127 	pthread_spin_unlock(&file->lock);
2128 	while (sub_reads-- > 0) {
2129 		sem_wait(&channel->sem);
2130 	}
2131 	if (rc == 0) {
2132 		return final_length;
2133 	} else {
2134 		return rc;
2135 	}
2136 }
2137 
2138 static void
2139 _file_sync(struct spdk_file *file, struct spdk_fs_channel *channel,
2140 	   spdk_file_op_complete cb_fn, void *cb_arg)
2141 {
2142 	struct spdk_fs_request *sync_req;
2143 	struct spdk_fs_request *flush_req;
2144 	struct spdk_fs_cb_args *sync_args;
2145 	struct spdk_fs_cb_args *flush_args;
2146 
2147 	BLOBFS_TRACE(file, "offset=%jx\n", file->append_pos);
2148 
2149 	pthread_spin_lock(&file->lock);
2150 	if (file->append_pos <= file->length_flushed || file->last == NULL) {
2151 		BLOBFS_TRACE(file, "done - no data to flush\n");
2152 		pthread_spin_unlock(&file->lock);
2153 		cb_fn(cb_arg, 0);
2154 		return;
2155 	}
2156 
2157 	sync_req = alloc_fs_request(channel);
2158 	assert(sync_req != NULL);
2159 	sync_args = &sync_req->args;
2160 
2161 	flush_req = alloc_fs_request(channel);
2162 	assert(flush_req != NULL);
2163 	flush_args = &flush_req->args;
2164 
2165 	sync_args->file = file;
2166 	sync_args->fn.file_op = cb_fn;
2167 	sync_args->arg = cb_arg;
2168 	sync_args->op.sync.offset = file->append_pos;
2169 	sync_args->op.sync.xattr_in_progress = false;
2170 	TAILQ_INSERT_TAIL(&file->sync_requests, sync_req, args.op.sync.tailq);
2171 	pthread_spin_unlock(&file->lock);
2172 
2173 	flush_args->file = file;
2174 	channel->send_request(__file_flush, flush_args);
2175 }
2176 
2177 int
2178 spdk_file_sync(struct spdk_file *file, struct spdk_io_channel *_channel)
2179 {
2180 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
2181 
2182 	_file_sync(file, channel, __sem_post, &channel->sem);
2183 	sem_wait(&channel->sem);
2184 
2185 	return 0;
2186 }
2187 
2188 void
2189 spdk_file_sync_async(struct spdk_file *file, struct spdk_io_channel *_channel,
2190 		     spdk_file_op_complete cb_fn, void *cb_arg)
2191 {
2192 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
2193 
2194 	_file_sync(file, channel, cb_fn, cb_arg);
2195 }
2196 
2197 void
2198 spdk_file_set_priority(struct spdk_file *file, uint32_t priority)
2199 {
2200 	BLOBFS_TRACE(file, "priority=%u\n", priority);
2201 	file->priority = priority;
2202 
2203 }
2204 
2205 /*
2206  * Close routines
2207  */
2208 
2209 static void
2210 __file_close_async_done(void *ctx, int bserrno)
2211 {
2212 	struct spdk_fs_request *req = ctx;
2213 	struct spdk_fs_cb_args *args = &req->args;
2214 
2215 	args->fn.file_op(args->arg, bserrno);
2216 	free_fs_request(req);
2217 }
2218 
2219 static void
2220 __file_close_async(struct spdk_file *file, struct spdk_fs_request *req)
2221 {
2222 	pthread_spin_lock(&file->lock);
2223 	if (file->ref_count == 0) {
2224 		pthread_spin_unlock(&file->lock);
2225 		__file_close_async_done(req, -EBADF);
2226 		return;
2227 	}
2228 
2229 	file->ref_count--;
2230 	if (file->ref_count > 0) {
2231 		pthread_spin_unlock(&file->lock);
2232 		__file_close_async_done(req, 0);
2233 		return;
2234 	}
2235 
2236 	pthread_spin_unlock(&file->lock);
2237 
2238 	spdk_bs_md_close_blob(&file->blob, __file_close_async_done, req);
2239 }
2240 
2241 static void
2242 __file_close_async__sync_done(void *arg, int fserrno)
2243 {
2244 	struct spdk_fs_request *req = arg;
2245 	struct spdk_fs_cb_args *args = &req->args;
2246 
2247 	__file_close_async(args->file, req);
2248 }
2249 
2250 void
2251 spdk_file_close_async(struct spdk_file *file, spdk_file_op_complete cb_fn, void *cb_arg)
2252 {
2253 	struct spdk_fs_request *req;
2254 	struct spdk_fs_cb_args *args;
2255 
2256 	req = alloc_fs_request(file->fs->md_target.md_fs_channel);
2257 	if (req == NULL) {
2258 		cb_fn(cb_arg, -ENOMEM);
2259 		return;
2260 	}
2261 
2262 	args = &req->args;
2263 	args->file = file;
2264 	args->fn.file_op = cb_fn;
2265 	args->arg = cb_arg;
2266 
2267 	spdk_file_sync_async(file, file->fs->md_target.md_io_channel, __file_close_async__sync_done, req);
2268 }
2269 
2270 static void
2271 __file_close_done(void *arg, int fserrno)
2272 {
2273 	struct spdk_fs_cb_args *args = arg;
2274 
2275 	args->rc = fserrno;
2276 	sem_post(args->sem);
2277 }
2278 
2279 static void
2280 __file_close(void *arg)
2281 {
2282 	struct spdk_fs_request *req = arg;
2283 	struct spdk_fs_cb_args *args = &req->args;
2284 	struct spdk_file *file = args->file;
2285 
2286 	__file_close_async(file, req);
2287 }
2288 
2289 int
2290 spdk_file_close(struct spdk_file *file, struct spdk_io_channel *_channel)
2291 {
2292 	struct spdk_fs_channel *channel = spdk_io_channel_get_ctx(_channel);
2293 	struct spdk_fs_request *req;
2294 	struct spdk_fs_cb_args *args;
2295 
2296 	req = alloc_fs_request(channel);
2297 	assert(req != NULL);
2298 
2299 	args = &req->args;
2300 
2301 	spdk_file_sync(file, _channel);
2302 	BLOBFS_TRACE(file, "name=%s\n", file->name);
2303 	args->file = file;
2304 	args->sem = &channel->sem;
2305 	args->fn.file_op = __file_close_done;
2306 	args->arg = req;
2307 	channel->send_request(__file_close, req);
2308 	sem_wait(&channel->sem);
2309 
2310 	return args->rc;
2311 }
2312 
2313 static void
2314 cache_free_buffers(struct spdk_file *file)
2315 {
2316 	BLOBFS_TRACE(file, "free=%s\n", file->name);
2317 	pthread_spin_lock(&file->lock);
2318 	pthread_spin_lock(&g_caches_lock);
2319 	if (file->tree->present_mask == 0) {
2320 		pthread_spin_unlock(&g_caches_lock);
2321 		pthread_spin_unlock(&file->lock);
2322 		return;
2323 	}
2324 	spdk_tree_free_buffers(file->tree);
2325 	if (file->tree->present_mask == 0) {
2326 		TAILQ_REMOVE(&g_caches, file, cache_tailq);
2327 	}
2328 	file->last = NULL;
2329 	pthread_spin_unlock(&g_caches_lock);
2330 	pthread_spin_unlock(&file->lock);
2331 }
2332 
2333 SPDK_LOG_REGISTER_TRACE_FLAG("blobfs", SPDK_TRACE_BLOBFS);
2334 SPDK_LOG_REGISTER_TRACE_FLAG("blobfs_rw", SPDK_TRACE_BLOBFS_RW);
2335