1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*******************************************************************************
3  * Filename:  target_core_file.c
4  *
5  * This file contains the Storage Engine <-> FILEIO transport specific functions
6  *
7  * (c) Copyright 2005-2013 Datera, Inc.
8  *
9  * Nicholas A. Bellinger <nab@kernel.org>
10  *
11  ******************************************************************************/
12 
13 #include <linux/string.h>
14 #include <linux/parser.h>
15 #include <linux/timer.h>
16 #include <linux/blkdev.h>
17 #include <linux/slab.h>
18 #include <linux/spinlock.h>
19 #include <linux/module.h>
20 #include <linux/vmalloc.h>
21 #include <linux/falloc.h>
22 #include <linux/uio.h>
23 #include <linux/scatterlist.h>
24 #include <scsi/scsi_proto.h>
25 #include <asm/unaligned.h>
26 
27 #include <target/target_core_base.h>
28 #include <target/target_core_backend.h>
29 
30 #include "target_core_file.h"
31 
32 static inline struct fd_dev *FD_DEV(struct se_device *dev)
33 {
34 	return container_of(dev, struct fd_dev, dev);
35 }
36 
37 static int fd_attach_hba(struct se_hba *hba, u32 host_id)
38 {
39 	struct fd_host *fd_host;
40 
41 	fd_host = kzalloc(sizeof(struct fd_host), GFP_KERNEL);
42 	if (!fd_host) {
43 		pr_err("Unable to allocate memory for struct fd_host\n");
44 		return -ENOMEM;
45 	}
46 
47 	fd_host->fd_host_id = host_id;
48 
49 	hba->hba_ptr = fd_host;
50 
51 	pr_debug("CORE_HBA[%d] - TCM FILEIO HBA Driver %s on Generic"
52 		" Target Core Stack %s\n", hba->hba_id, FD_VERSION,
53 		TARGET_CORE_VERSION);
54 	pr_debug("CORE_HBA[%d] - Attached FILEIO HBA: %u to Generic\n",
55 		hba->hba_id, fd_host->fd_host_id);
56 
57 	return 0;
58 }
59 
60 static void fd_detach_hba(struct se_hba *hba)
61 {
62 	struct fd_host *fd_host = hba->hba_ptr;
63 
64 	pr_debug("CORE_HBA[%d] - Detached FILEIO HBA: %u from Generic"
65 		" Target Core\n", hba->hba_id, fd_host->fd_host_id);
66 
67 	kfree(fd_host);
68 	hba->hba_ptr = NULL;
69 }
70 
71 static struct se_device *fd_alloc_device(struct se_hba *hba, const char *name)
72 {
73 	struct fd_dev *fd_dev;
74 	struct fd_host *fd_host = hba->hba_ptr;
75 
76 	fd_dev = kzalloc(sizeof(struct fd_dev), GFP_KERNEL);
77 	if (!fd_dev) {
78 		pr_err("Unable to allocate memory for struct fd_dev\n");
79 		return NULL;
80 	}
81 
82 	fd_dev->fd_host = fd_host;
83 
84 	pr_debug("FILEIO: Allocated fd_dev for %p\n", name);
85 
86 	return &fd_dev->dev;
87 }
88 
89 static bool fd_configure_unmap(struct se_device *dev)
90 {
91 	struct file *file = FD_DEV(dev)->fd_file;
92 	struct inode *inode = file->f_mapping->host;
93 
94 	if (S_ISBLK(inode->i_mode))
95 		return target_configure_unmap_from_queue(&dev->dev_attrib,
96 							 I_BDEV(inode));
97 
98 	/* Limit UNMAP emulation to 8k Number of LBAs (NoLB) */
99 	dev->dev_attrib.max_unmap_lba_count = 0x2000;
100 	/* Currently hardcoded to 1 in Linux/SCSI code. */
101 	dev->dev_attrib.max_unmap_block_desc_count = 1;
102 	dev->dev_attrib.unmap_granularity = 1;
103 	dev->dev_attrib.unmap_granularity_alignment = 0;
104 	return true;
105 }
106 
107 static int fd_configure_device(struct se_device *dev)
108 {
109 	struct fd_dev *fd_dev = FD_DEV(dev);
110 	struct fd_host *fd_host = dev->se_hba->hba_ptr;
111 	struct file *file;
112 	struct inode *inode = NULL;
113 	int flags, ret = -EINVAL;
114 
115 	if (!(fd_dev->fbd_flags & FBDF_HAS_PATH)) {
116 		pr_err("Missing fd_dev_name=\n");
117 		return -EINVAL;
118 	}
119 
120 	/*
121 	 * Use O_DSYNC by default instead of O_SYNC to forgo syncing
122 	 * of pure timestamp updates.
123 	 */
124 	flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
125 
126 	/*
127 	 * Optionally allow fd_buffered_io=1 to be enabled for people
128 	 * who want use the fs buffer cache as an WriteCache mechanism.
129 	 *
130 	 * This means that in event of a hard failure, there is a risk
131 	 * of silent data-loss if the SCSI client has *not* performed a
132 	 * forced unit access (FUA) write, or issued SYNCHRONIZE_CACHE
133 	 * to write-out the entire device cache.
134 	 */
135 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
136 		pr_debug("FILEIO: Disabling O_DSYNC, using buffered FILEIO\n");
137 		flags &= ~O_DSYNC;
138 	}
139 
140 	file = filp_open(fd_dev->fd_dev_name, flags, 0600);
141 	if (IS_ERR(file)) {
142 		pr_err("filp_open(%s) failed\n", fd_dev->fd_dev_name);
143 		ret = PTR_ERR(file);
144 		goto fail;
145 	}
146 	fd_dev->fd_file = file;
147 	/*
148 	 * If using a block backend with this struct file, we extract
149 	 * fd_dev->fd_[block,dev]_size from struct block_device.
150 	 *
151 	 * Otherwise, we use the passed fd_size= from configfs
152 	 */
153 	inode = file->f_mapping->host;
154 	if (S_ISBLK(inode->i_mode)) {
155 		struct block_device *bdev = I_BDEV(inode);
156 		unsigned long long dev_size;
157 
158 		fd_dev->fd_block_size = bdev_logical_block_size(bdev);
159 		/*
160 		 * Determine the number of bytes from i_size_read() minus
161 		 * one (1) logical sector from underlying struct block_device
162 		 */
163 		dev_size = (i_size_read(file->f_mapping->host) -
164 				       fd_dev->fd_block_size);
165 
166 		pr_debug("FILEIO: Using size: %llu bytes from struct"
167 			" block_device blocks: %llu logical_block_size: %d\n",
168 			dev_size, div_u64(dev_size, fd_dev->fd_block_size),
169 			fd_dev->fd_block_size);
170 		/*
171 		 * Enable write same emulation for IBLOCK and use 0xFFFF as
172 		 * the smaller WRITE_SAME(10) only has a two-byte block count.
173 		 */
174 		dev->dev_attrib.max_write_same_len = 0xFFFF;
175 
176 		if (bdev_nonrot(bdev))
177 			dev->dev_attrib.is_nonrot = 1;
178 	} else {
179 		if (!(fd_dev->fbd_flags & FBDF_HAS_SIZE)) {
180 			pr_err("FILEIO: Missing fd_dev_size="
181 				" parameter, and no backing struct"
182 				" block_device\n");
183 			goto fail;
184 		}
185 
186 		fd_dev->fd_block_size = FD_BLOCKSIZE;
187 
188 		/*
189 		 * Limit WRITE_SAME w/ UNMAP=0 emulation to 8k Number of LBAs (NoLB)
190 		 * based upon struct iovec limit for vfs_writev()
191 		 */
192 		dev->dev_attrib.max_write_same_len = 0x1000;
193 	}
194 
195 	dev->dev_attrib.hw_block_size = fd_dev->fd_block_size;
196 	dev->dev_attrib.hw_max_sectors = FD_MAX_BYTES / fd_dev->fd_block_size;
197 	dev->dev_attrib.hw_queue_depth = FD_MAX_DEVICE_QUEUE_DEPTH;
198 
199 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
200 		pr_debug("FILEIO: Forcing setting of emulate_write_cache=1"
201 			" with FDBD_HAS_BUFFERED_IO_WCE\n");
202 		dev->dev_attrib.emulate_write_cache = 1;
203 	}
204 
205 	fd_dev->fd_dev_id = fd_host->fd_host_dev_id_count++;
206 	fd_dev->fd_queue_depth = dev->queue_depth;
207 
208 	pr_debug("CORE_FILE[%u] - Added TCM FILEIO Device ID: %u at %s,"
209 		" %llu total bytes\n", fd_host->fd_host_id, fd_dev->fd_dev_id,
210 			fd_dev->fd_dev_name, fd_dev->fd_dev_size);
211 
212 	return 0;
213 fail:
214 	if (fd_dev->fd_file) {
215 		filp_close(fd_dev->fd_file, NULL);
216 		fd_dev->fd_file = NULL;
217 	}
218 	return ret;
219 }
220 
221 static void fd_dev_call_rcu(struct rcu_head *p)
222 {
223 	struct se_device *dev = container_of(p, struct se_device, rcu_head);
224 	struct fd_dev *fd_dev = FD_DEV(dev);
225 
226 	kfree(fd_dev);
227 }
228 
229 static void fd_free_device(struct se_device *dev)
230 {
231 	call_rcu(&dev->rcu_head, fd_dev_call_rcu);
232 }
233 
234 static void fd_destroy_device(struct se_device *dev)
235 {
236 	struct fd_dev *fd_dev = FD_DEV(dev);
237 
238 	if (fd_dev->fd_file) {
239 		filp_close(fd_dev->fd_file, NULL);
240 		fd_dev->fd_file = NULL;
241 	}
242 }
243 
244 struct target_core_file_cmd {
245 	unsigned long	len;
246 	struct se_cmd	*cmd;
247 	struct kiocb	iocb;
248 	struct bio_vec	bvecs[];
249 };
250 
251 static void cmd_rw_aio_complete(struct kiocb *iocb, long ret)
252 {
253 	struct target_core_file_cmd *cmd;
254 
255 	cmd = container_of(iocb, struct target_core_file_cmd, iocb);
256 
257 	if (ret != cmd->len)
258 		target_complete_cmd(cmd->cmd, SAM_STAT_CHECK_CONDITION);
259 	else
260 		target_complete_cmd(cmd->cmd, SAM_STAT_GOOD);
261 
262 	kfree(cmd);
263 }
264 
265 static sense_reason_t
266 fd_execute_rw_aio(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
267 	      enum dma_data_direction data_direction)
268 {
269 	int is_write = !(data_direction == DMA_FROM_DEVICE);
270 	struct se_device *dev = cmd->se_dev;
271 	struct fd_dev *fd_dev = FD_DEV(dev);
272 	struct file *file = fd_dev->fd_file;
273 	struct target_core_file_cmd *aio_cmd;
274 	struct iov_iter iter;
275 	struct scatterlist *sg;
276 	ssize_t len = 0;
277 	int ret = 0, i;
278 
279 	aio_cmd = kmalloc(struct_size(aio_cmd, bvecs, sgl_nents), GFP_KERNEL);
280 	if (!aio_cmd)
281 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
282 
283 	for_each_sg(sgl, sg, sgl_nents, i) {
284 		aio_cmd->bvecs[i].bv_page = sg_page(sg);
285 		aio_cmd->bvecs[i].bv_len = sg->length;
286 		aio_cmd->bvecs[i].bv_offset = sg->offset;
287 
288 		len += sg->length;
289 	}
290 
291 	iov_iter_bvec(&iter, is_write, aio_cmd->bvecs, sgl_nents, len);
292 
293 	aio_cmd->cmd = cmd;
294 	aio_cmd->len = len;
295 	aio_cmd->iocb.ki_pos = cmd->t_task_lba * dev->dev_attrib.block_size;
296 	aio_cmd->iocb.ki_filp = file;
297 	aio_cmd->iocb.ki_complete = cmd_rw_aio_complete;
298 	aio_cmd->iocb.ki_flags = IOCB_DIRECT;
299 
300 	if (is_write && (cmd->se_cmd_flags & SCF_FUA))
301 		aio_cmd->iocb.ki_flags |= IOCB_DSYNC;
302 
303 	if (is_write)
304 		ret = call_write_iter(file, &aio_cmd->iocb, &iter);
305 	else
306 		ret = call_read_iter(file, &aio_cmd->iocb, &iter);
307 
308 	if (ret != -EIOCBQUEUED)
309 		cmd_rw_aio_complete(&aio_cmd->iocb, ret);
310 
311 	return 0;
312 }
313 
314 static int fd_do_rw(struct se_cmd *cmd, struct file *fd,
315 		    u32 block_size, struct scatterlist *sgl,
316 		    u32 sgl_nents, u32 data_length, int is_write)
317 {
318 	struct scatterlist *sg;
319 	struct iov_iter iter;
320 	struct bio_vec *bvec;
321 	ssize_t len = 0;
322 	loff_t pos = (cmd->t_task_lba * block_size);
323 	int ret = 0, i;
324 
325 	bvec = kcalloc(sgl_nents, sizeof(struct bio_vec), GFP_KERNEL);
326 	if (!bvec) {
327 		pr_err("Unable to allocate fd_do_readv iov[]\n");
328 		return -ENOMEM;
329 	}
330 
331 	for_each_sg(sgl, sg, sgl_nents, i) {
332 		bvec[i].bv_page = sg_page(sg);
333 		bvec[i].bv_len = sg->length;
334 		bvec[i].bv_offset = sg->offset;
335 
336 		len += sg->length;
337 	}
338 
339 	iov_iter_bvec(&iter, is_write, bvec, sgl_nents, len);
340 	if (is_write)
341 		ret = vfs_iter_write(fd, &iter, &pos, 0);
342 	else
343 		ret = vfs_iter_read(fd, &iter, &pos, 0);
344 
345 	if (is_write) {
346 		if (ret < 0 || ret != data_length) {
347 			pr_err("%s() write returned %d\n", __func__, ret);
348 			if (ret >= 0)
349 				ret = -EINVAL;
350 		}
351 	} else {
352 		/*
353 		 * Return zeros and GOOD status even if the READ did not return
354 		 * the expected virt_size for struct file w/o a backing struct
355 		 * block_device.
356 		 */
357 		if (S_ISBLK(file_inode(fd)->i_mode)) {
358 			if (ret < 0 || ret != data_length) {
359 				pr_err("%s() returned %d, expecting %u for "
360 						"S_ISBLK\n", __func__, ret,
361 						data_length);
362 				if (ret >= 0)
363 					ret = -EINVAL;
364 			}
365 		} else {
366 			if (ret < 0) {
367 				pr_err("%s() returned %d for non S_ISBLK\n",
368 						__func__, ret);
369 			} else if (ret != data_length) {
370 				/*
371 				 * Short read case:
372 				 * Probably some one truncate file under us.
373 				 * We must explicitly zero sg-pages to prevent
374 				 * expose uninizialized pages to userspace.
375 				 */
376 				if (ret < data_length)
377 					ret += iov_iter_zero(data_length - ret, &iter);
378 				else
379 					ret = -EINVAL;
380 			}
381 		}
382 	}
383 	kfree(bvec);
384 	return ret;
385 }
386 
387 static sense_reason_t
388 fd_execute_sync_cache(struct se_cmd *cmd)
389 {
390 	struct se_device *dev = cmd->se_dev;
391 	struct fd_dev *fd_dev = FD_DEV(dev);
392 	int immed = (cmd->t_task_cdb[1] & 0x2);
393 	loff_t start, end;
394 	int ret;
395 
396 	/*
397 	 * If the Immediate bit is set, queue up the GOOD response
398 	 * for this SYNCHRONIZE_CACHE op
399 	 */
400 	if (immed)
401 		target_complete_cmd(cmd, SAM_STAT_GOOD);
402 
403 	/*
404 	 * Determine if we will be flushing the entire device.
405 	 */
406 	if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
407 		start = 0;
408 		end = LLONG_MAX;
409 	} else {
410 		start = cmd->t_task_lba * dev->dev_attrib.block_size;
411 		if (cmd->data_length)
412 			end = start + cmd->data_length - 1;
413 		else
414 			end = LLONG_MAX;
415 	}
416 
417 	ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
418 	if (ret != 0)
419 		pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
420 
421 	if (immed)
422 		return 0;
423 
424 	if (ret)
425 		target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
426 	else
427 		target_complete_cmd(cmd, SAM_STAT_GOOD);
428 
429 	return 0;
430 }
431 
432 static sense_reason_t
433 fd_execute_write_same(struct se_cmd *cmd)
434 {
435 	struct se_device *se_dev = cmd->se_dev;
436 	struct fd_dev *fd_dev = FD_DEV(se_dev);
437 	loff_t pos = cmd->t_task_lba * se_dev->dev_attrib.block_size;
438 	sector_t nolb = sbc_get_write_same_sectors(cmd);
439 	struct iov_iter iter;
440 	struct bio_vec *bvec;
441 	unsigned int len = 0, i;
442 	ssize_t ret;
443 
444 	if (cmd->prot_op) {
445 		pr_err("WRITE_SAME: Protection information with FILEIO"
446 		       " backends not supported\n");
447 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
448 	}
449 
450 	if (!cmd->t_data_nents)
451 		return TCM_INVALID_CDB_FIELD;
452 
453 	if (cmd->t_data_nents > 1 ||
454 	    cmd->t_data_sg[0].length != cmd->se_dev->dev_attrib.block_size) {
455 		pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
456 			" block_size: %u\n",
457 			cmd->t_data_nents,
458 			cmd->t_data_sg[0].length,
459 			cmd->se_dev->dev_attrib.block_size);
460 		return TCM_INVALID_CDB_FIELD;
461 	}
462 
463 	bvec = kcalloc(nolb, sizeof(struct bio_vec), GFP_KERNEL);
464 	if (!bvec)
465 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
466 
467 	for (i = 0; i < nolb; i++) {
468 		bvec[i].bv_page = sg_page(&cmd->t_data_sg[0]);
469 		bvec[i].bv_len = cmd->t_data_sg[0].length;
470 		bvec[i].bv_offset = cmd->t_data_sg[0].offset;
471 
472 		len += se_dev->dev_attrib.block_size;
473 	}
474 
475 	iov_iter_bvec(&iter, ITER_SOURCE, bvec, nolb, len);
476 	ret = vfs_iter_write(fd_dev->fd_file, &iter, &pos, 0);
477 
478 	kfree(bvec);
479 	if (ret < 0 || ret != len) {
480 		pr_err("vfs_iter_write() returned %zd for write same\n", ret);
481 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
482 	}
483 
484 	target_complete_cmd(cmd, SAM_STAT_GOOD);
485 	return 0;
486 }
487 
488 static int
489 fd_do_prot_fill(struct se_device *se_dev, sector_t lba, sector_t nolb,
490 		void *buf, size_t bufsize)
491 {
492 	struct fd_dev *fd_dev = FD_DEV(se_dev);
493 	struct file *prot_fd = fd_dev->fd_prot_file;
494 	sector_t prot_length, prot;
495 	loff_t pos = lba * se_dev->prot_length;
496 
497 	if (!prot_fd) {
498 		pr_err("Unable to locate fd_dev->fd_prot_file\n");
499 		return -ENODEV;
500 	}
501 
502 	prot_length = nolb * se_dev->prot_length;
503 
504 	memset(buf, 0xff, bufsize);
505 	for (prot = 0; prot < prot_length;) {
506 		sector_t len = min_t(sector_t, bufsize, prot_length - prot);
507 		ssize_t ret = kernel_write(prot_fd, buf, len, &pos);
508 
509 		if (ret != len) {
510 			pr_err("vfs_write to prot file failed: %zd\n", ret);
511 			return ret < 0 ? ret : -ENODEV;
512 		}
513 		prot += ret;
514 	}
515 
516 	return 0;
517 }
518 
519 static int
520 fd_do_prot_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
521 {
522 	void *buf;
523 	int rc;
524 
525 	buf = (void *)__get_free_page(GFP_KERNEL);
526 	if (!buf) {
527 		pr_err("Unable to allocate FILEIO prot buf\n");
528 		return -ENOMEM;
529 	}
530 
531 	rc = fd_do_prot_fill(cmd->se_dev, lba, nolb, buf, PAGE_SIZE);
532 
533 	free_page((unsigned long)buf);
534 
535 	return rc;
536 }
537 
538 static sense_reason_t
539 fd_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
540 {
541 	struct file *file = FD_DEV(cmd->se_dev)->fd_file;
542 	struct inode *inode = file->f_mapping->host;
543 	int ret;
544 
545 	if (!nolb) {
546 		return 0;
547 	}
548 
549 	if (cmd->se_dev->dev_attrib.pi_prot_type) {
550 		ret = fd_do_prot_unmap(cmd, lba, nolb);
551 		if (ret)
552 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
553 	}
554 
555 	if (S_ISBLK(inode->i_mode)) {
556 		/* The backend is block device, use discard */
557 		struct block_device *bdev = I_BDEV(inode);
558 		struct se_device *dev = cmd->se_dev;
559 
560 		ret = blkdev_issue_discard(bdev,
561 					   target_to_linux_sector(dev, lba),
562 					   target_to_linux_sector(dev,  nolb),
563 					   GFP_KERNEL);
564 		if (ret < 0) {
565 			pr_warn("FILEIO: blkdev_issue_discard() failed: %d\n",
566 				ret);
567 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
568 		}
569 	} else {
570 		/* The backend is normal file, use fallocate */
571 		struct se_device *se_dev = cmd->se_dev;
572 		loff_t pos = lba * se_dev->dev_attrib.block_size;
573 		unsigned int len = nolb * se_dev->dev_attrib.block_size;
574 		int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
575 
576 		if (!file->f_op->fallocate)
577 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
578 
579 		ret = file->f_op->fallocate(file, mode, pos, len);
580 		if (ret < 0) {
581 			pr_warn("FILEIO: fallocate() failed: %d\n", ret);
582 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
583 		}
584 	}
585 
586 	return 0;
587 }
588 
589 static sense_reason_t
590 fd_execute_rw_buffered(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
591 	      enum dma_data_direction data_direction)
592 {
593 	struct se_device *dev = cmd->se_dev;
594 	struct fd_dev *fd_dev = FD_DEV(dev);
595 	struct file *file = fd_dev->fd_file;
596 	struct file *pfile = fd_dev->fd_prot_file;
597 	sense_reason_t rc;
598 	int ret = 0;
599 	/*
600 	 * Call vectorized fileio functions to map struct scatterlist
601 	 * physical memory addresses to struct iovec virtual memory.
602 	 */
603 	if (data_direction == DMA_FROM_DEVICE) {
604 		if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
605 			ret = fd_do_rw(cmd, pfile, dev->prot_length,
606 				       cmd->t_prot_sg, cmd->t_prot_nents,
607 				       cmd->prot_length, 0);
608 			if (ret < 0)
609 				return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
610 		}
611 
612 		ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
613 			       sgl, sgl_nents, cmd->data_length, 0);
614 
615 		if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type &&
616 		    dev->dev_attrib.pi_prot_verify) {
617 			u32 sectors = cmd->data_length >>
618 					ilog2(dev->dev_attrib.block_size);
619 
620 			rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
621 					    0, cmd->t_prot_sg, 0);
622 			if (rc)
623 				return rc;
624 		}
625 	} else {
626 		if (cmd->prot_type && dev->dev_attrib.pi_prot_type &&
627 		    dev->dev_attrib.pi_prot_verify) {
628 			u32 sectors = cmd->data_length >>
629 					ilog2(dev->dev_attrib.block_size);
630 
631 			rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
632 					    0, cmd->t_prot_sg, 0);
633 			if (rc)
634 				return rc;
635 		}
636 
637 		ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
638 			       sgl, sgl_nents, cmd->data_length, 1);
639 		/*
640 		 * Perform implicit vfs_fsync_range() for fd_do_writev() ops
641 		 * for SCSI WRITEs with Forced Unit Access (FUA) set.
642 		 * Allow this to happen independent of WCE=0 setting.
643 		 */
644 		if (ret > 0 && (cmd->se_cmd_flags & SCF_FUA)) {
645 			loff_t start = cmd->t_task_lba *
646 				dev->dev_attrib.block_size;
647 			loff_t end;
648 
649 			if (cmd->data_length)
650 				end = start + cmd->data_length - 1;
651 			else
652 				end = LLONG_MAX;
653 
654 			vfs_fsync_range(fd_dev->fd_file, start, end, 1);
655 		}
656 
657 		if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type) {
658 			ret = fd_do_rw(cmd, pfile, dev->prot_length,
659 				       cmd->t_prot_sg, cmd->t_prot_nents,
660 				       cmd->prot_length, 1);
661 			if (ret < 0)
662 				return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
663 		}
664 	}
665 
666 	if (ret < 0)
667 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
668 
669 	target_complete_cmd(cmd, SAM_STAT_GOOD);
670 	return 0;
671 }
672 
673 static sense_reason_t
674 fd_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
675 	      enum dma_data_direction data_direction)
676 {
677 	struct se_device *dev = cmd->se_dev;
678 	struct fd_dev *fd_dev = FD_DEV(dev);
679 
680 	/*
681 	 * We are currently limited by the number of iovecs (2048) per
682 	 * single vfs_[writev,readv] call.
683 	 */
684 	if (cmd->data_length > FD_MAX_BYTES) {
685 		pr_err("FILEIO: Not able to process I/O of %u bytes due to"
686 		       "FD_MAX_BYTES: %u iovec count limitation\n",
687 			cmd->data_length, FD_MAX_BYTES);
688 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
689 	}
690 
691 	if (fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO)
692 		return fd_execute_rw_aio(cmd, sgl, sgl_nents, data_direction);
693 	return fd_execute_rw_buffered(cmd, sgl, sgl_nents, data_direction);
694 }
695 
696 enum {
697 	Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io,
698 	Opt_fd_async_io, Opt_err
699 };
700 
701 static match_table_t tokens = {
702 	{Opt_fd_dev_name, "fd_dev_name=%s"},
703 	{Opt_fd_dev_size, "fd_dev_size=%s"},
704 	{Opt_fd_buffered_io, "fd_buffered_io=%d"},
705 	{Opt_fd_async_io, "fd_async_io=%d"},
706 	{Opt_err, NULL}
707 };
708 
709 static ssize_t fd_set_configfs_dev_params(struct se_device *dev,
710 		const char *page, ssize_t count)
711 {
712 	struct fd_dev *fd_dev = FD_DEV(dev);
713 	char *orig, *ptr, *arg_p, *opts;
714 	substring_t args[MAX_OPT_ARGS];
715 	int ret = 0, arg, token;
716 
717 	opts = kstrdup(page, GFP_KERNEL);
718 	if (!opts)
719 		return -ENOMEM;
720 
721 	orig = opts;
722 
723 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
724 		if (!*ptr)
725 			continue;
726 
727 		token = match_token(ptr, tokens, args);
728 		switch (token) {
729 		case Opt_fd_dev_name:
730 			if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
731 				FD_MAX_DEV_NAME) == 0) {
732 				ret = -EINVAL;
733 				break;
734 			}
735 			pr_debug("FILEIO: Referencing Path: %s\n",
736 					fd_dev->fd_dev_name);
737 			fd_dev->fbd_flags |= FBDF_HAS_PATH;
738 			break;
739 		case Opt_fd_dev_size:
740 			arg_p = match_strdup(&args[0]);
741 			if (!arg_p) {
742 				ret = -ENOMEM;
743 				break;
744 			}
745 			ret = kstrtoull(arg_p, 0, &fd_dev->fd_dev_size);
746 			kfree(arg_p);
747 			if (ret < 0) {
748 				pr_err("kstrtoull() failed for"
749 						" fd_dev_size=\n");
750 				goto out;
751 			}
752 			pr_debug("FILEIO: Referencing Size: %llu"
753 					" bytes\n", fd_dev->fd_dev_size);
754 			fd_dev->fbd_flags |= FBDF_HAS_SIZE;
755 			break;
756 		case Opt_fd_buffered_io:
757 			ret = match_int(args, &arg);
758 			if (ret)
759 				goto out;
760 			if (arg != 1) {
761 				pr_err("bogus fd_buffered_io=%d value\n", arg);
762 				ret = -EINVAL;
763 				goto out;
764 			}
765 
766 			pr_debug("FILEIO: Using buffered I/O"
767 				" operations for struct fd_dev\n");
768 
769 			fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
770 			break;
771 		case Opt_fd_async_io:
772 			ret = match_int(args, &arg);
773 			if (ret)
774 				goto out;
775 			if (arg != 1) {
776 				pr_err("bogus fd_async_io=%d value\n", arg);
777 				ret = -EINVAL;
778 				goto out;
779 			}
780 
781 			pr_debug("FILEIO: Using async I/O"
782 				" operations for struct fd_dev\n");
783 
784 			fd_dev->fbd_flags |= FDBD_HAS_ASYNC_IO;
785 			break;
786 		default:
787 			break;
788 		}
789 	}
790 
791 out:
792 	kfree(orig);
793 	return (!ret) ? count : ret;
794 }
795 
796 static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b)
797 {
798 	struct fd_dev *fd_dev = FD_DEV(dev);
799 	ssize_t bl = 0;
800 
801 	bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
802 	bl += sprintf(b + bl, "        File: %s  Size: %llu  Mode: %s Async: %d\n",
803 		fd_dev->fd_dev_name, fd_dev->fd_dev_size,
804 		(fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
805 		"Buffered-WCE" : "O_DSYNC",
806 		!!(fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO));
807 	return bl;
808 }
809 
810 static sector_t fd_get_blocks(struct se_device *dev)
811 {
812 	struct fd_dev *fd_dev = FD_DEV(dev);
813 	struct file *f = fd_dev->fd_file;
814 	struct inode *i = f->f_mapping->host;
815 	unsigned long long dev_size;
816 	/*
817 	 * When using a file that references an underlying struct block_device,
818 	 * ensure dev_size is always based on the current inode size in order
819 	 * to handle underlying block_device resize operations.
820 	 */
821 	if (S_ISBLK(i->i_mode))
822 		dev_size = i_size_read(i);
823 	else
824 		dev_size = fd_dev->fd_dev_size;
825 
826 	return div_u64(dev_size - dev->dev_attrib.block_size,
827 		       dev->dev_attrib.block_size);
828 }
829 
830 static int fd_init_prot(struct se_device *dev)
831 {
832 	struct fd_dev *fd_dev = FD_DEV(dev);
833 	struct file *prot_file, *file = fd_dev->fd_file;
834 	struct inode *inode;
835 	int ret, flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
836 	char buf[FD_MAX_DEV_PROT_NAME];
837 
838 	if (!file) {
839 		pr_err("Unable to locate fd_dev->fd_file\n");
840 		return -ENODEV;
841 	}
842 
843 	inode = file->f_mapping->host;
844 	if (S_ISBLK(inode->i_mode)) {
845 		pr_err("FILEIO Protection emulation only supported on"
846 		       " !S_ISBLK\n");
847 		return -ENOSYS;
848 	}
849 
850 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE)
851 		flags &= ~O_DSYNC;
852 
853 	snprintf(buf, FD_MAX_DEV_PROT_NAME, "%s.protection",
854 		 fd_dev->fd_dev_name);
855 
856 	prot_file = filp_open(buf, flags, 0600);
857 	if (IS_ERR(prot_file)) {
858 		pr_err("filp_open(%s) failed\n", buf);
859 		ret = PTR_ERR(prot_file);
860 		return ret;
861 	}
862 	fd_dev->fd_prot_file = prot_file;
863 
864 	return 0;
865 }
866 
867 static int fd_format_prot(struct se_device *dev)
868 {
869 	unsigned char *buf;
870 	int unit_size = FDBD_FORMAT_UNIT_SIZE * dev->dev_attrib.block_size;
871 	int ret;
872 
873 	if (!dev->dev_attrib.pi_prot_type) {
874 		pr_err("Unable to format_prot while pi_prot_type == 0\n");
875 		return -ENODEV;
876 	}
877 
878 	buf = vzalloc(unit_size);
879 	if (!buf) {
880 		pr_err("Unable to allocate FILEIO prot buf\n");
881 		return -ENOMEM;
882 	}
883 
884 	pr_debug("Using FILEIO prot_length: %llu\n",
885 		 (unsigned long long)(dev->transport->get_blocks(dev) + 1) *
886 					dev->prot_length);
887 
888 	ret = fd_do_prot_fill(dev, 0, dev->transport->get_blocks(dev) + 1,
889 			      buf, unit_size);
890 	vfree(buf);
891 	return ret;
892 }
893 
894 static void fd_free_prot(struct se_device *dev)
895 {
896 	struct fd_dev *fd_dev = FD_DEV(dev);
897 
898 	if (!fd_dev->fd_prot_file)
899 		return;
900 
901 	filp_close(fd_dev->fd_prot_file, NULL);
902 	fd_dev->fd_prot_file = NULL;
903 }
904 
905 static struct sbc_ops fd_sbc_ops = {
906 	.execute_rw		= fd_execute_rw,
907 	.execute_sync_cache	= fd_execute_sync_cache,
908 	.execute_write_same	= fd_execute_write_same,
909 	.execute_unmap		= fd_execute_unmap,
910 };
911 
912 static sense_reason_t
913 fd_parse_cdb(struct se_cmd *cmd)
914 {
915 	return sbc_parse_cdb(cmd, &fd_sbc_ops);
916 }
917 
918 static const struct target_backend_ops fileio_ops = {
919 	.name			= "fileio",
920 	.inquiry_prod		= "FILEIO",
921 	.inquiry_rev		= FD_VERSION,
922 	.owner			= THIS_MODULE,
923 	.attach_hba		= fd_attach_hba,
924 	.detach_hba		= fd_detach_hba,
925 	.alloc_device		= fd_alloc_device,
926 	.configure_device	= fd_configure_device,
927 	.destroy_device		= fd_destroy_device,
928 	.free_device		= fd_free_device,
929 	.configure_unmap	= fd_configure_unmap,
930 	.parse_cdb		= fd_parse_cdb,
931 	.set_configfs_dev_params = fd_set_configfs_dev_params,
932 	.show_configfs_dev_params = fd_show_configfs_dev_params,
933 	.get_device_type	= sbc_get_device_type,
934 	.get_blocks		= fd_get_blocks,
935 	.init_prot		= fd_init_prot,
936 	.format_prot		= fd_format_prot,
937 	.free_prot		= fd_free_prot,
938 	.tb_dev_attrib_attrs	= sbc_attrib_attrs,
939 };
940 
941 static int __init fileio_module_init(void)
942 {
943 	return transport_backend_register(&fileio_ops);
944 }
945 
946 static void __exit fileio_module_exit(void)
947 {
948 	target_backend_unregister(&fileio_ops);
949 }
950 
951 MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
952 MODULE_AUTHOR("nab@Linux-iSCSI.org");
953 MODULE_LICENSE("GPL");
954 
955 module_init(fileio_module_init);
956 module_exit(fileio_module_exit);
957