1 /*******************************************************************************
2  * Filename:  target_core_file.c
3  *
4  * This file contains the Storage Engine <-> FILEIO transport specific functions
5  *
6  * (c) Copyright 2005-2013 Datera, Inc.
7  *
8  * Nicholas A. Bellinger <nab@kernel.org>
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
23  *
24  ******************************************************************************/
25 
26 #include <linux/string.h>
27 #include <linux/parser.h>
28 #include <linux/timer.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/spinlock.h>
32 #include <linux/module.h>
33 #include <linux/vmalloc.h>
34 #include <linux/falloc.h>
35 #include <linux/uio.h>
36 #include <scsi/scsi_proto.h>
37 #include <asm/unaligned.h>
38 
39 #include <target/target_core_base.h>
40 #include <target/target_core_backend.h>
41 
42 #include "target_core_file.h"
43 
44 static inline struct fd_dev *FD_DEV(struct se_device *dev)
45 {
46 	return container_of(dev, struct fd_dev, dev);
47 }
48 
49 static int fd_attach_hba(struct se_hba *hba, u32 host_id)
50 {
51 	struct fd_host *fd_host;
52 
53 	fd_host = kzalloc(sizeof(struct fd_host), GFP_KERNEL);
54 	if (!fd_host) {
55 		pr_err("Unable to allocate memory for struct fd_host\n");
56 		return -ENOMEM;
57 	}
58 
59 	fd_host->fd_host_id = host_id;
60 
61 	hba->hba_ptr = fd_host;
62 
63 	pr_debug("CORE_HBA[%d] - TCM FILEIO HBA Driver %s on Generic"
64 		" Target Core Stack %s\n", hba->hba_id, FD_VERSION,
65 		TARGET_CORE_VERSION);
66 	pr_debug("CORE_HBA[%d] - Attached FILEIO HBA: %u to Generic\n",
67 		hba->hba_id, fd_host->fd_host_id);
68 
69 	return 0;
70 }
71 
72 static void fd_detach_hba(struct se_hba *hba)
73 {
74 	struct fd_host *fd_host = hba->hba_ptr;
75 
76 	pr_debug("CORE_HBA[%d] - Detached FILEIO HBA: %u from Generic"
77 		" Target Core\n", hba->hba_id, fd_host->fd_host_id);
78 
79 	kfree(fd_host);
80 	hba->hba_ptr = NULL;
81 }
82 
83 static struct se_device *fd_alloc_device(struct se_hba *hba, const char *name)
84 {
85 	struct fd_dev *fd_dev;
86 	struct fd_host *fd_host = hba->hba_ptr;
87 
88 	fd_dev = kzalloc(sizeof(struct fd_dev), GFP_KERNEL);
89 	if (!fd_dev) {
90 		pr_err("Unable to allocate memory for struct fd_dev\n");
91 		return NULL;
92 	}
93 
94 	fd_dev->fd_host = fd_host;
95 
96 	pr_debug("FILEIO: Allocated fd_dev for %p\n", name);
97 
98 	return &fd_dev->dev;
99 }
100 
101 static int fd_configure_device(struct se_device *dev)
102 {
103 	struct fd_dev *fd_dev = FD_DEV(dev);
104 	struct fd_host *fd_host = dev->se_hba->hba_ptr;
105 	struct file *file;
106 	struct inode *inode = NULL;
107 	int flags, ret = -EINVAL;
108 
109 	if (!(fd_dev->fbd_flags & FBDF_HAS_PATH)) {
110 		pr_err("Missing fd_dev_name=\n");
111 		return -EINVAL;
112 	}
113 
114 	/*
115 	 * Use O_DSYNC by default instead of O_SYNC to forgo syncing
116 	 * of pure timestamp updates.
117 	 */
118 	flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
119 
120 	/*
121 	 * Optionally allow fd_buffered_io=1 to be enabled for people
122 	 * who want use the fs buffer cache as an WriteCache mechanism.
123 	 *
124 	 * This means that in event of a hard failure, there is a risk
125 	 * of silent data-loss if the SCSI client has *not* performed a
126 	 * forced unit access (FUA) write, or issued SYNCHRONIZE_CACHE
127 	 * to write-out the entire device cache.
128 	 */
129 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
130 		pr_debug("FILEIO: Disabling O_DSYNC, using buffered FILEIO\n");
131 		flags &= ~O_DSYNC;
132 	}
133 
134 	file = filp_open(fd_dev->fd_dev_name, flags, 0600);
135 	if (IS_ERR(file)) {
136 		pr_err("filp_open(%s) failed\n", fd_dev->fd_dev_name);
137 		ret = PTR_ERR(file);
138 		goto fail;
139 	}
140 	fd_dev->fd_file = file;
141 	/*
142 	 * If using a block backend with this struct file, we extract
143 	 * fd_dev->fd_[block,dev]_size from struct block_device.
144 	 *
145 	 * Otherwise, we use the passed fd_size= from configfs
146 	 */
147 	inode = file->f_mapping->host;
148 	if (S_ISBLK(inode->i_mode)) {
149 		struct request_queue *q = bdev_get_queue(inode->i_bdev);
150 		unsigned long long dev_size;
151 
152 		fd_dev->fd_block_size = bdev_logical_block_size(inode->i_bdev);
153 		/*
154 		 * Determine the number of bytes from i_size_read() minus
155 		 * one (1) logical sector from underlying struct block_device
156 		 */
157 		dev_size = (i_size_read(file->f_mapping->host) -
158 				       fd_dev->fd_block_size);
159 
160 		pr_debug("FILEIO: Using size: %llu bytes from struct"
161 			" block_device blocks: %llu logical_block_size: %d\n",
162 			dev_size, div_u64(dev_size, fd_dev->fd_block_size),
163 			fd_dev->fd_block_size);
164 
165 		if (target_configure_unmap_from_queue(&dev->dev_attrib, q))
166 			pr_debug("IFILE: BLOCK Discard support available,"
167 				 " disabled by default\n");
168 		/*
169 		 * Enable write same emulation for IBLOCK and use 0xFFFF as
170 		 * the smaller WRITE_SAME(10) only has a two-byte block count.
171 		 */
172 		dev->dev_attrib.max_write_same_len = 0xFFFF;
173 
174 		if (blk_queue_nonrot(q))
175 			dev->dev_attrib.is_nonrot = 1;
176 	} else {
177 		if (!(fd_dev->fbd_flags & FBDF_HAS_SIZE)) {
178 			pr_err("FILEIO: Missing fd_dev_size="
179 				" parameter, and no backing struct"
180 				" block_device\n");
181 			goto fail;
182 		}
183 
184 		fd_dev->fd_block_size = FD_BLOCKSIZE;
185 		/*
186 		 * Limit UNMAP emulation to 8k Number of LBAs (NoLB)
187 		 */
188 		dev->dev_attrib.max_unmap_lba_count = 0x2000;
189 		/*
190 		 * Currently hardcoded to 1 in Linux/SCSI code..
191 		 */
192 		dev->dev_attrib.max_unmap_block_desc_count = 1;
193 		dev->dev_attrib.unmap_granularity = 1;
194 		dev->dev_attrib.unmap_granularity_alignment = 0;
195 
196 		/*
197 		 * Limit WRITE_SAME w/ UNMAP=0 emulation to 8k Number of LBAs (NoLB)
198 		 * based upon struct iovec limit for vfs_writev()
199 		 */
200 		dev->dev_attrib.max_write_same_len = 0x1000;
201 	}
202 
203 	dev->dev_attrib.hw_block_size = fd_dev->fd_block_size;
204 	dev->dev_attrib.max_bytes_per_io = FD_MAX_BYTES;
205 	dev->dev_attrib.hw_max_sectors = FD_MAX_BYTES / fd_dev->fd_block_size;
206 	dev->dev_attrib.hw_queue_depth = FD_MAX_DEVICE_QUEUE_DEPTH;
207 
208 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
209 		pr_debug("FILEIO: Forcing setting of emulate_write_cache=1"
210 			" with FDBD_HAS_BUFFERED_IO_WCE\n");
211 		dev->dev_attrib.emulate_write_cache = 1;
212 	}
213 
214 	fd_dev->fd_dev_id = fd_host->fd_host_dev_id_count++;
215 	fd_dev->fd_queue_depth = dev->queue_depth;
216 
217 	pr_debug("CORE_FILE[%u] - Added TCM FILEIO Device ID: %u at %s,"
218 		" %llu total bytes\n", fd_host->fd_host_id, fd_dev->fd_dev_id,
219 			fd_dev->fd_dev_name, fd_dev->fd_dev_size);
220 
221 	return 0;
222 fail:
223 	if (fd_dev->fd_file) {
224 		filp_close(fd_dev->fd_file, NULL);
225 		fd_dev->fd_file = NULL;
226 	}
227 	return ret;
228 }
229 
230 static void fd_dev_call_rcu(struct rcu_head *p)
231 {
232 	struct se_device *dev = container_of(p, struct se_device, rcu_head);
233 	struct fd_dev *fd_dev = FD_DEV(dev);
234 
235 	kfree(fd_dev);
236 }
237 
238 static void fd_free_device(struct se_device *dev)
239 {
240 	call_rcu(&dev->rcu_head, fd_dev_call_rcu);
241 }
242 
243 static void fd_destroy_device(struct se_device *dev)
244 {
245 	struct fd_dev *fd_dev = FD_DEV(dev);
246 
247 	if (fd_dev->fd_file) {
248 		filp_close(fd_dev->fd_file, NULL);
249 		fd_dev->fd_file = NULL;
250 	}
251 }
252 
253 struct target_core_file_cmd {
254 	unsigned long	len;
255 	struct se_cmd	*cmd;
256 	struct kiocb	iocb;
257 };
258 
259 static void cmd_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
260 {
261 	struct target_core_file_cmd *cmd;
262 
263 	cmd = container_of(iocb, struct target_core_file_cmd, iocb);
264 
265 	if (ret != cmd->len)
266 		target_complete_cmd(cmd->cmd, SAM_STAT_CHECK_CONDITION);
267 	else
268 		target_complete_cmd(cmd->cmd, SAM_STAT_GOOD);
269 
270 	kfree(cmd);
271 }
272 
273 static sense_reason_t
274 fd_execute_rw_aio(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
275 	      enum dma_data_direction data_direction)
276 {
277 	int is_write = !(data_direction == DMA_FROM_DEVICE);
278 	struct se_device *dev = cmd->se_dev;
279 	struct fd_dev *fd_dev = FD_DEV(dev);
280 	struct file *file = fd_dev->fd_file;
281 	struct target_core_file_cmd *aio_cmd;
282 	struct iov_iter iter = {};
283 	struct scatterlist *sg;
284 	struct bio_vec *bvec;
285 	ssize_t len = 0;
286 	int ret = 0, i;
287 
288 	aio_cmd = kmalloc(sizeof(struct target_core_file_cmd), GFP_KERNEL);
289 	if (!aio_cmd)
290 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
291 
292 	bvec = kcalloc(sgl_nents, sizeof(struct bio_vec), GFP_KERNEL);
293 	if (!bvec) {
294 		kfree(aio_cmd);
295 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
296 	}
297 
298 	for_each_sg(sgl, sg, sgl_nents, i) {
299 		bvec[i].bv_page = sg_page(sg);
300 		bvec[i].bv_len = sg->length;
301 		bvec[i].bv_offset = sg->offset;
302 
303 		len += sg->length;
304 	}
305 
306 	iov_iter_bvec(&iter, ITER_BVEC | is_write, bvec, sgl_nents, len);
307 
308 	aio_cmd->cmd = cmd;
309 	aio_cmd->len = len;
310 	aio_cmd->iocb.ki_pos = cmd->t_task_lba * dev->dev_attrib.block_size;
311 	aio_cmd->iocb.ki_filp = file;
312 	aio_cmd->iocb.ki_complete = cmd_rw_aio_complete;
313 	aio_cmd->iocb.ki_flags = IOCB_DIRECT;
314 
315 	if (is_write && (cmd->se_cmd_flags & SCF_FUA))
316 		aio_cmd->iocb.ki_flags |= IOCB_DSYNC;
317 
318 	if (is_write)
319 		ret = call_write_iter(file, &aio_cmd->iocb, &iter);
320 	else
321 		ret = call_read_iter(file, &aio_cmd->iocb, &iter);
322 
323 	kfree(bvec);
324 
325 	if (ret != -EIOCBQUEUED)
326 		cmd_rw_aio_complete(&aio_cmd->iocb, ret, 0);
327 
328 	return 0;
329 }
330 
331 static int fd_do_rw(struct se_cmd *cmd, struct file *fd,
332 		    u32 block_size, struct scatterlist *sgl,
333 		    u32 sgl_nents, u32 data_length, int is_write)
334 {
335 	struct scatterlist *sg;
336 	struct iov_iter iter;
337 	struct bio_vec *bvec;
338 	ssize_t len = 0;
339 	loff_t pos = (cmd->t_task_lba * block_size);
340 	int ret = 0, i;
341 
342 	bvec = kcalloc(sgl_nents, sizeof(struct bio_vec), GFP_KERNEL);
343 	if (!bvec) {
344 		pr_err("Unable to allocate fd_do_readv iov[]\n");
345 		return -ENOMEM;
346 	}
347 
348 	for_each_sg(sgl, sg, sgl_nents, i) {
349 		bvec[i].bv_page = sg_page(sg);
350 		bvec[i].bv_len = sg->length;
351 		bvec[i].bv_offset = sg->offset;
352 
353 		len += sg->length;
354 	}
355 
356 	iov_iter_bvec(&iter, ITER_BVEC, bvec, sgl_nents, len);
357 	if (is_write)
358 		ret = vfs_iter_write(fd, &iter, &pos, 0);
359 	else
360 		ret = vfs_iter_read(fd, &iter, &pos, 0);
361 
362 	if (is_write) {
363 		if (ret < 0 || ret != data_length) {
364 			pr_err("%s() write returned %d\n", __func__, ret);
365 			if (ret >= 0)
366 				ret = -EINVAL;
367 		}
368 	} else {
369 		/*
370 		 * Return zeros and GOOD status even if the READ did not return
371 		 * the expected virt_size for struct file w/o a backing struct
372 		 * block_device.
373 		 */
374 		if (S_ISBLK(file_inode(fd)->i_mode)) {
375 			if (ret < 0 || ret != data_length) {
376 				pr_err("%s() returned %d, expecting %u for "
377 						"S_ISBLK\n", __func__, ret,
378 						data_length);
379 				if (ret >= 0)
380 					ret = -EINVAL;
381 			}
382 		} else {
383 			if (ret < 0) {
384 				pr_err("%s() returned %d for non S_ISBLK\n",
385 						__func__, ret);
386 			} else if (ret != data_length) {
387 				/*
388 				 * Short read case:
389 				 * Probably some one truncate file under us.
390 				 * We must explicitly zero sg-pages to prevent
391 				 * expose uninizialized pages to userspace.
392 				 */
393 				if (ret < data_length)
394 					ret += iov_iter_zero(data_length - ret, &iter);
395 				else
396 					ret = -EINVAL;
397 			}
398 		}
399 	}
400 	kfree(bvec);
401 	return ret;
402 }
403 
404 static sense_reason_t
405 fd_execute_sync_cache(struct se_cmd *cmd)
406 {
407 	struct se_device *dev = cmd->se_dev;
408 	struct fd_dev *fd_dev = FD_DEV(dev);
409 	int immed = (cmd->t_task_cdb[1] & 0x2);
410 	loff_t start, end;
411 	int ret;
412 
413 	/*
414 	 * If the Immediate bit is set, queue up the GOOD response
415 	 * for this SYNCHRONIZE_CACHE op
416 	 */
417 	if (immed)
418 		target_complete_cmd(cmd, SAM_STAT_GOOD);
419 
420 	/*
421 	 * Determine if we will be flushing the entire device.
422 	 */
423 	if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
424 		start = 0;
425 		end = LLONG_MAX;
426 	} else {
427 		start = cmd->t_task_lba * dev->dev_attrib.block_size;
428 		if (cmd->data_length)
429 			end = start + cmd->data_length - 1;
430 		else
431 			end = LLONG_MAX;
432 	}
433 
434 	ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
435 	if (ret != 0)
436 		pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
437 
438 	if (immed)
439 		return 0;
440 
441 	if (ret)
442 		target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
443 	else
444 		target_complete_cmd(cmd, SAM_STAT_GOOD);
445 
446 	return 0;
447 }
448 
449 static sense_reason_t
450 fd_execute_write_same(struct se_cmd *cmd)
451 {
452 	struct se_device *se_dev = cmd->se_dev;
453 	struct fd_dev *fd_dev = FD_DEV(se_dev);
454 	loff_t pos = cmd->t_task_lba * se_dev->dev_attrib.block_size;
455 	sector_t nolb = sbc_get_write_same_sectors(cmd);
456 	struct iov_iter iter;
457 	struct bio_vec *bvec;
458 	unsigned int len = 0, i;
459 	ssize_t ret;
460 
461 	if (!nolb) {
462 		target_complete_cmd(cmd, SAM_STAT_GOOD);
463 		return 0;
464 	}
465 	if (cmd->prot_op) {
466 		pr_err("WRITE_SAME: Protection information with FILEIO"
467 		       " backends not supported\n");
468 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
469 	}
470 
471 	if (cmd->t_data_nents > 1 ||
472 	    cmd->t_data_sg[0].length != cmd->se_dev->dev_attrib.block_size) {
473 		pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
474 			" block_size: %u\n",
475 			cmd->t_data_nents,
476 			cmd->t_data_sg[0].length,
477 			cmd->se_dev->dev_attrib.block_size);
478 		return TCM_INVALID_CDB_FIELD;
479 	}
480 
481 	bvec = kcalloc(nolb, sizeof(struct bio_vec), GFP_KERNEL);
482 	if (!bvec)
483 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
484 
485 	for (i = 0; i < nolb; i++) {
486 		bvec[i].bv_page = sg_page(&cmd->t_data_sg[0]);
487 		bvec[i].bv_len = cmd->t_data_sg[0].length;
488 		bvec[i].bv_offset = cmd->t_data_sg[0].offset;
489 
490 		len += se_dev->dev_attrib.block_size;
491 	}
492 
493 	iov_iter_bvec(&iter, ITER_BVEC, bvec, nolb, len);
494 	ret = vfs_iter_write(fd_dev->fd_file, &iter, &pos, 0);
495 
496 	kfree(bvec);
497 	if (ret < 0 || ret != len) {
498 		pr_err("vfs_iter_write() returned %zd for write same\n", ret);
499 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
500 	}
501 
502 	target_complete_cmd(cmd, SAM_STAT_GOOD);
503 	return 0;
504 }
505 
506 static int
507 fd_do_prot_fill(struct se_device *se_dev, sector_t lba, sector_t nolb,
508 		void *buf, size_t bufsize)
509 {
510 	struct fd_dev *fd_dev = FD_DEV(se_dev);
511 	struct file *prot_fd = fd_dev->fd_prot_file;
512 	sector_t prot_length, prot;
513 	loff_t pos = lba * se_dev->prot_length;
514 
515 	if (!prot_fd) {
516 		pr_err("Unable to locate fd_dev->fd_prot_file\n");
517 		return -ENODEV;
518 	}
519 
520 	prot_length = nolb * se_dev->prot_length;
521 
522 	for (prot = 0; prot < prot_length;) {
523 		sector_t len = min_t(sector_t, bufsize, prot_length - prot);
524 		ssize_t ret = kernel_write(prot_fd, buf, len, &pos);
525 
526 		if (ret != len) {
527 			pr_err("vfs_write to prot file failed: %zd\n", ret);
528 			return ret < 0 ? ret : -ENODEV;
529 		}
530 		prot += ret;
531 	}
532 
533 	return 0;
534 }
535 
536 static int
537 fd_do_prot_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
538 {
539 	void *buf;
540 	int rc;
541 
542 	buf = (void *)__get_free_page(GFP_KERNEL);
543 	if (!buf) {
544 		pr_err("Unable to allocate FILEIO prot buf\n");
545 		return -ENOMEM;
546 	}
547 	memset(buf, 0xff, PAGE_SIZE);
548 
549 	rc = fd_do_prot_fill(cmd->se_dev, lba, nolb, buf, PAGE_SIZE);
550 
551 	free_page((unsigned long)buf);
552 
553 	return rc;
554 }
555 
556 static sense_reason_t
557 fd_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
558 {
559 	struct file *file = FD_DEV(cmd->se_dev)->fd_file;
560 	struct inode *inode = file->f_mapping->host;
561 	int ret;
562 
563 	if (!nolb) {
564 		return 0;
565 	}
566 
567 	if (cmd->se_dev->dev_attrib.pi_prot_type) {
568 		ret = fd_do_prot_unmap(cmd, lba, nolb);
569 		if (ret)
570 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
571 	}
572 
573 	if (S_ISBLK(inode->i_mode)) {
574 		/* The backend is block device, use discard */
575 		struct block_device *bdev = inode->i_bdev;
576 		struct se_device *dev = cmd->se_dev;
577 
578 		ret = blkdev_issue_discard(bdev,
579 					   target_to_linux_sector(dev, lba),
580 					   target_to_linux_sector(dev,  nolb),
581 					   GFP_KERNEL, 0);
582 		if (ret < 0) {
583 			pr_warn("FILEIO: blkdev_issue_discard() failed: %d\n",
584 				ret);
585 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
586 		}
587 	} else {
588 		/* The backend is normal file, use fallocate */
589 		struct se_device *se_dev = cmd->se_dev;
590 		loff_t pos = lba * se_dev->dev_attrib.block_size;
591 		unsigned int len = nolb * se_dev->dev_attrib.block_size;
592 		int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
593 
594 		if (!file->f_op->fallocate)
595 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
596 
597 		ret = file->f_op->fallocate(file, mode, pos, len);
598 		if (ret < 0) {
599 			pr_warn("FILEIO: fallocate() failed: %d\n", ret);
600 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
601 		}
602 	}
603 
604 	return 0;
605 }
606 
607 static sense_reason_t
608 fd_execute_rw_buffered(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
609 	      enum dma_data_direction data_direction)
610 {
611 	struct se_device *dev = cmd->se_dev;
612 	struct fd_dev *fd_dev = FD_DEV(dev);
613 	struct file *file = fd_dev->fd_file;
614 	struct file *pfile = fd_dev->fd_prot_file;
615 	sense_reason_t rc;
616 	int ret = 0;
617 	/*
618 	 * Call vectorized fileio functions to map struct scatterlist
619 	 * physical memory addresses to struct iovec virtual memory.
620 	 */
621 	if (data_direction == DMA_FROM_DEVICE) {
622 		if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
623 			ret = fd_do_rw(cmd, pfile, dev->prot_length,
624 				       cmd->t_prot_sg, cmd->t_prot_nents,
625 				       cmd->prot_length, 0);
626 			if (ret < 0)
627 				return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
628 		}
629 
630 		ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
631 			       sgl, sgl_nents, cmd->data_length, 0);
632 
633 		if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type &&
634 		    dev->dev_attrib.pi_prot_verify) {
635 			u32 sectors = cmd->data_length >>
636 					ilog2(dev->dev_attrib.block_size);
637 
638 			rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
639 					    0, cmd->t_prot_sg, 0);
640 			if (rc)
641 				return rc;
642 		}
643 	} else {
644 		if (cmd->prot_type && dev->dev_attrib.pi_prot_type &&
645 		    dev->dev_attrib.pi_prot_verify) {
646 			u32 sectors = cmd->data_length >>
647 					ilog2(dev->dev_attrib.block_size);
648 
649 			rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
650 					    0, cmd->t_prot_sg, 0);
651 			if (rc)
652 				return rc;
653 		}
654 
655 		ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
656 			       sgl, sgl_nents, cmd->data_length, 1);
657 		/*
658 		 * Perform implicit vfs_fsync_range() for fd_do_writev() ops
659 		 * for SCSI WRITEs with Forced Unit Access (FUA) set.
660 		 * Allow this to happen independent of WCE=0 setting.
661 		 */
662 		if (ret > 0 && (cmd->se_cmd_flags & SCF_FUA)) {
663 			loff_t start = cmd->t_task_lba *
664 				dev->dev_attrib.block_size;
665 			loff_t end;
666 
667 			if (cmd->data_length)
668 				end = start + cmd->data_length - 1;
669 			else
670 				end = LLONG_MAX;
671 
672 			vfs_fsync_range(fd_dev->fd_file, start, end, 1);
673 		}
674 
675 		if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type) {
676 			ret = fd_do_rw(cmd, pfile, dev->prot_length,
677 				       cmd->t_prot_sg, cmd->t_prot_nents,
678 				       cmd->prot_length, 1);
679 			if (ret < 0)
680 				return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
681 		}
682 	}
683 
684 	if (ret < 0)
685 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
686 
687 	target_complete_cmd(cmd, SAM_STAT_GOOD);
688 	return 0;
689 }
690 
691 static sense_reason_t
692 fd_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
693 	      enum dma_data_direction data_direction)
694 {
695 	struct se_device *dev = cmd->se_dev;
696 	struct fd_dev *fd_dev = FD_DEV(dev);
697 
698 	/*
699 	 * We are currently limited by the number of iovecs (2048) per
700 	 * single vfs_[writev,readv] call.
701 	 */
702 	if (cmd->data_length > FD_MAX_BYTES) {
703 		pr_err("FILEIO: Not able to process I/O of %u bytes due to"
704 		       "FD_MAX_BYTES: %u iovec count limitation\n",
705 			cmd->data_length, FD_MAX_BYTES);
706 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
707 	}
708 
709 	if (fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO)
710 		return fd_execute_rw_aio(cmd, sgl, sgl_nents, data_direction);
711 	return fd_execute_rw_buffered(cmd, sgl, sgl_nents, data_direction);
712 }
713 
714 enum {
715 	Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io,
716 	Opt_fd_async_io, Opt_err
717 };
718 
719 static match_table_t tokens = {
720 	{Opt_fd_dev_name, "fd_dev_name=%s"},
721 	{Opt_fd_dev_size, "fd_dev_size=%s"},
722 	{Opt_fd_buffered_io, "fd_buffered_io=%d"},
723 	{Opt_fd_async_io, "fd_async_io=%d"},
724 	{Opt_err, NULL}
725 };
726 
727 static ssize_t fd_set_configfs_dev_params(struct se_device *dev,
728 		const char *page, ssize_t count)
729 {
730 	struct fd_dev *fd_dev = FD_DEV(dev);
731 	char *orig, *ptr, *arg_p, *opts;
732 	substring_t args[MAX_OPT_ARGS];
733 	int ret = 0, arg, token;
734 
735 	opts = kstrdup(page, GFP_KERNEL);
736 	if (!opts)
737 		return -ENOMEM;
738 
739 	orig = opts;
740 
741 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
742 		if (!*ptr)
743 			continue;
744 
745 		token = match_token(ptr, tokens, args);
746 		switch (token) {
747 		case Opt_fd_dev_name:
748 			if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
749 				FD_MAX_DEV_NAME) == 0) {
750 				ret = -EINVAL;
751 				break;
752 			}
753 			pr_debug("FILEIO: Referencing Path: %s\n",
754 					fd_dev->fd_dev_name);
755 			fd_dev->fbd_flags |= FBDF_HAS_PATH;
756 			break;
757 		case Opt_fd_dev_size:
758 			arg_p = match_strdup(&args[0]);
759 			if (!arg_p) {
760 				ret = -ENOMEM;
761 				break;
762 			}
763 			ret = kstrtoull(arg_p, 0, &fd_dev->fd_dev_size);
764 			kfree(arg_p);
765 			if (ret < 0) {
766 				pr_err("kstrtoull() failed for"
767 						" fd_dev_size=\n");
768 				goto out;
769 			}
770 			pr_debug("FILEIO: Referencing Size: %llu"
771 					" bytes\n", fd_dev->fd_dev_size);
772 			fd_dev->fbd_flags |= FBDF_HAS_SIZE;
773 			break;
774 		case Opt_fd_buffered_io:
775 			ret = match_int(args, &arg);
776 			if (ret)
777 				goto out;
778 			if (arg != 1) {
779 				pr_err("bogus fd_buffered_io=%d value\n", arg);
780 				ret = -EINVAL;
781 				goto out;
782 			}
783 
784 			pr_debug("FILEIO: Using buffered I/O"
785 				" operations for struct fd_dev\n");
786 
787 			fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
788 			break;
789 		case Opt_fd_async_io:
790 			ret = match_int(args, &arg);
791 			if (ret)
792 				goto out;
793 			if (arg != 1) {
794 				pr_err("bogus fd_async_io=%d value\n", arg);
795 				ret = -EINVAL;
796 				goto out;
797 			}
798 
799 			pr_debug("FILEIO: Using async I/O"
800 				" operations for struct fd_dev\n");
801 
802 			fd_dev->fbd_flags |= FDBD_HAS_ASYNC_IO;
803 			break;
804 		default:
805 			break;
806 		}
807 	}
808 
809 out:
810 	kfree(orig);
811 	return (!ret) ? count : ret;
812 }
813 
814 static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b)
815 {
816 	struct fd_dev *fd_dev = FD_DEV(dev);
817 	ssize_t bl = 0;
818 
819 	bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
820 	bl += sprintf(b + bl, "        File: %s  Size: %llu  Mode: %s Async: %d\n",
821 		fd_dev->fd_dev_name, fd_dev->fd_dev_size,
822 		(fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
823 		"Buffered-WCE" : "O_DSYNC",
824 		!!(fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO));
825 	return bl;
826 }
827 
828 static sector_t fd_get_blocks(struct se_device *dev)
829 {
830 	struct fd_dev *fd_dev = FD_DEV(dev);
831 	struct file *f = fd_dev->fd_file;
832 	struct inode *i = f->f_mapping->host;
833 	unsigned long long dev_size;
834 	/*
835 	 * When using a file that references an underlying struct block_device,
836 	 * ensure dev_size is always based on the current inode size in order
837 	 * to handle underlying block_device resize operations.
838 	 */
839 	if (S_ISBLK(i->i_mode))
840 		dev_size = i_size_read(i);
841 	else
842 		dev_size = fd_dev->fd_dev_size;
843 
844 	return div_u64(dev_size - dev->dev_attrib.block_size,
845 		       dev->dev_attrib.block_size);
846 }
847 
848 static int fd_init_prot(struct se_device *dev)
849 {
850 	struct fd_dev *fd_dev = FD_DEV(dev);
851 	struct file *prot_file, *file = fd_dev->fd_file;
852 	struct inode *inode;
853 	int ret, flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
854 	char buf[FD_MAX_DEV_PROT_NAME];
855 
856 	if (!file) {
857 		pr_err("Unable to locate fd_dev->fd_file\n");
858 		return -ENODEV;
859 	}
860 
861 	inode = file->f_mapping->host;
862 	if (S_ISBLK(inode->i_mode)) {
863 		pr_err("FILEIO Protection emulation only supported on"
864 		       " !S_ISBLK\n");
865 		return -ENOSYS;
866 	}
867 
868 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE)
869 		flags &= ~O_DSYNC;
870 
871 	snprintf(buf, FD_MAX_DEV_PROT_NAME, "%s.protection",
872 		 fd_dev->fd_dev_name);
873 
874 	prot_file = filp_open(buf, flags, 0600);
875 	if (IS_ERR(prot_file)) {
876 		pr_err("filp_open(%s) failed\n", buf);
877 		ret = PTR_ERR(prot_file);
878 		return ret;
879 	}
880 	fd_dev->fd_prot_file = prot_file;
881 
882 	return 0;
883 }
884 
885 static int fd_format_prot(struct se_device *dev)
886 {
887 	unsigned char *buf;
888 	int unit_size = FDBD_FORMAT_UNIT_SIZE * dev->dev_attrib.block_size;
889 	int ret;
890 
891 	if (!dev->dev_attrib.pi_prot_type) {
892 		pr_err("Unable to format_prot while pi_prot_type == 0\n");
893 		return -ENODEV;
894 	}
895 
896 	buf = vzalloc(unit_size);
897 	if (!buf) {
898 		pr_err("Unable to allocate FILEIO prot buf\n");
899 		return -ENOMEM;
900 	}
901 
902 	pr_debug("Using FILEIO prot_length: %llu\n",
903 		 (unsigned long long)(dev->transport->get_blocks(dev) + 1) *
904 					dev->prot_length);
905 
906 	memset(buf, 0xff, unit_size);
907 	ret = fd_do_prot_fill(dev, 0, dev->transport->get_blocks(dev) + 1,
908 			      buf, unit_size);
909 	vfree(buf);
910 	return ret;
911 }
912 
913 static void fd_free_prot(struct se_device *dev)
914 {
915 	struct fd_dev *fd_dev = FD_DEV(dev);
916 
917 	if (!fd_dev->fd_prot_file)
918 		return;
919 
920 	filp_close(fd_dev->fd_prot_file, NULL);
921 	fd_dev->fd_prot_file = NULL;
922 }
923 
924 static struct sbc_ops fd_sbc_ops = {
925 	.execute_rw		= fd_execute_rw,
926 	.execute_sync_cache	= fd_execute_sync_cache,
927 	.execute_write_same	= fd_execute_write_same,
928 	.execute_unmap		= fd_execute_unmap,
929 };
930 
931 static sense_reason_t
932 fd_parse_cdb(struct se_cmd *cmd)
933 {
934 	return sbc_parse_cdb(cmd, &fd_sbc_ops);
935 }
936 
937 static const struct target_backend_ops fileio_ops = {
938 	.name			= "fileio",
939 	.inquiry_prod		= "FILEIO",
940 	.inquiry_rev		= FD_VERSION,
941 	.owner			= THIS_MODULE,
942 	.attach_hba		= fd_attach_hba,
943 	.detach_hba		= fd_detach_hba,
944 	.alloc_device		= fd_alloc_device,
945 	.configure_device	= fd_configure_device,
946 	.destroy_device		= fd_destroy_device,
947 	.free_device		= fd_free_device,
948 	.parse_cdb		= fd_parse_cdb,
949 	.set_configfs_dev_params = fd_set_configfs_dev_params,
950 	.show_configfs_dev_params = fd_show_configfs_dev_params,
951 	.get_device_type	= sbc_get_device_type,
952 	.get_blocks		= fd_get_blocks,
953 	.init_prot		= fd_init_prot,
954 	.format_prot		= fd_format_prot,
955 	.free_prot		= fd_free_prot,
956 	.tb_dev_attrib_attrs	= sbc_attrib_attrs,
957 };
958 
959 static int __init fileio_module_init(void)
960 {
961 	return transport_backend_register(&fileio_ops);
962 }
963 
964 static void __exit fileio_module_exit(void)
965 {
966 	target_backend_unregister(&fileio_ops);
967 }
968 
969 MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
970 MODULE_AUTHOR("nab@Linux-iSCSI.org");
971 MODULE_LICENSE("GPL");
972 
973 module_init(fileio_module_init);
974 module_exit(fileio_module_exit);
975