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 		bvec_set_page(&aio_cmd->bvecs[i], sg_page(sg), sg->length,
285 			      sg->offset);
286 		len += sg->length;
287 	}
288 
289 	iov_iter_bvec(&iter, is_write, aio_cmd->bvecs, sgl_nents, len);
290 
291 	aio_cmd->cmd = cmd;
292 	aio_cmd->len = len;
293 	aio_cmd->iocb.ki_pos = cmd->t_task_lba * dev->dev_attrib.block_size;
294 	aio_cmd->iocb.ki_filp = file;
295 	aio_cmd->iocb.ki_complete = cmd_rw_aio_complete;
296 	aio_cmd->iocb.ki_flags = IOCB_DIRECT;
297 
298 	if (is_write && (cmd->se_cmd_flags & SCF_FUA))
299 		aio_cmd->iocb.ki_flags |= IOCB_DSYNC;
300 
301 	if (is_write)
302 		ret = call_write_iter(file, &aio_cmd->iocb, &iter);
303 	else
304 		ret = call_read_iter(file, &aio_cmd->iocb, &iter);
305 
306 	if (ret != -EIOCBQUEUED)
307 		cmd_rw_aio_complete(&aio_cmd->iocb, ret);
308 
309 	return 0;
310 }
311 
312 static int fd_do_rw(struct se_cmd *cmd, struct file *fd,
313 		    u32 block_size, struct scatterlist *sgl,
314 		    u32 sgl_nents, u32 data_length, int is_write)
315 {
316 	struct scatterlist *sg;
317 	struct iov_iter iter;
318 	struct bio_vec *bvec;
319 	ssize_t len = 0;
320 	loff_t pos = (cmd->t_task_lba * block_size);
321 	int ret = 0, i;
322 
323 	bvec = kcalloc(sgl_nents, sizeof(struct bio_vec), GFP_KERNEL);
324 	if (!bvec) {
325 		pr_err("Unable to allocate fd_do_readv iov[]\n");
326 		return -ENOMEM;
327 	}
328 
329 	for_each_sg(sgl, sg, sgl_nents, i) {
330 		bvec_set_page(&bvec[i], sg_page(sg), sg->length, sg->offset);
331 		len += sg->length;
332 	}
333 
334 	iov_iter_bvec(&iter, is_write, bvec, sgl_nents, len);
335 	if (is_write)
336 		ret = vfs_iter_write(fd, &iter, &pos, 0);
337 	else
338 		ret = vfs_iter_read(fd, &iter, &pos, 0);
339 
340 	if (is_write) {
341 		if (ret < 0 || ret != data_length) {
342 			pr_err("%s() write returned %d\n", __func__, ret);
343 			if (ret >= 0)
344 				ret = -EINVAL;
345 		}
346 	} else {
347 		/*
348 		 * Return zeros and GOOD status even if the READ did not return
349 		 * the expected virt_size for struct file w/o a backing struct
350 		 * block_device.
351 		 */
352 		if (S_ISBLK(file_inode(fd)->i_mode)) {
353 			if (ret < 0 || ret != data_length) {
354 				pr_err("%s() returned %d, expecting %u for "
355 						"S_ISBLK\n", __func__, ret,
356 						data_length);
357 				if (ret >= 0)
358 					ret = -EINVAL;
359 			}
360 		} else {
361 			if (ret < 0) {
362 				pr_err("%s() returned %d for non S_ISBLK\n",
363 						__func__, ret);
364 			} else if (ret != data_length) {
365 				/*
366 				 * Short read case:
367 				 * Probably some one truncate file under us.
368 				 * We must explicitly zero sg-pages to prevent
369 				 * expose uninizialized pages to userspace.
370 				 */
371 				if (ret < data_length)
372 					ret += iov_iter_zero(data_length - ret, &iter);
373 				else
374 					ret = -EINVAL;
375 			}
376 		}
377 	}
378 	kfree(bvec);
379 	return ret;
380 }
381 
382 static sense_reason_t
383 fd_execute_sync_cache(struct se_cmd *cmd)
384 {
385 	struct se_device *dev = cmd->se_dev;
386 	struct fd_dev *fd_dev = FD_DEV(dev);
387 	int immed = (cmd->t_task_cdb[1] & 0x2);
388 	loff_t start, end;
389 	int ret;
390 
391 	/*
392 	 * If the Immediate bit is set, queue up the GOOD response
393 	 * for this SYNCHRONIZE_CACHE op
394 	 */
395 	if (immed)
396 		target_complete_cmd(cmd, SAM_STAT_GOOD);
397 
398 	/*
399 	 * Determine if we will be flushing the entire device.
400 	 */
401 	if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
402 		start = 0;
403 		end = LLONG_MAX;
404 	} else {
405 		start = cmd->t_task_lba * dev->dev_attrib.block_size;
406 		if (cmd->data_length)
407 			end = start + cmd->data_length - 1;
408 		else
409 			end = LLONG_MAX;
410 	}
411 
412 	ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
413 	if (ret != 0)
414 		pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
415 
416 	if (immed)
417 		return 0;
418 
419 	if (ret)
420 		target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
421 	else
422 		target_complete_cmd(cmd, SAM_STAT_GOOD);
423 
424 	return 0;
425 }
426 
427 static sense_reason_t
428 fd_execute_write_same(struct se_cmd *cmd)
429 {
430 	struct se_device *se_dev = cmd->se_dev;
431 	struct fd_dev *fd_dev = FD_DEV(se_dev);
432 	loff_t pos = cmd->t_task_lba * se_dev->dev_attrib.block_size;
433 	sector_t nolb = sbc_get_write_same_sectors(cmd);
434 	struct iov_iter iter;
435 	struct bio_vec *bvec;
436 	unsigned int len = 0, i;
437 	ssize_t ret;
438 
439 	if (cmd->prot_op) {
440 		pr_err("WRITE_SAME: Protection information with FILEIO"
441 		       " backends not supported\n");
442 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
443 	}
444 
445 	if (!cmd->t_data_nents)
446 		return TCM_INVALID_CDB_FIELD;
447 
448 	if (cmd->t_data_nents > 1 ||
449 	    cmd->t_data_sg[0].length != cmd->se_dev->dev_attrib.block_size) {
450 		pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
451 			" block_size: %u\n",
452 			cmd->t_data_nents,
453 			cmd->t_data_sg[0].length,
454 			cmd->se_dev->dev_attrib.block_size);
455 		return TCM_INVALID_CDB_FIELD;
456 	}
457 
458 	bvec = kcalloc(nolb, sizeof(struct bio_vec), GFP_KERNEL);
459 	if (!bvec)
460 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
461 
462 	for (i = 0; i < nolb; i++) {
463 		bvec_set_page(&bvec[i], sg_page(&cmd->t_data_sg[0]),
464 			      cmd->t_data_sg[0].length,
465 			      cmd->t_data_sg[0].offset);
466 		len += se_dev->dev_attrib.block_size;
467 	}
468 
469 	iov_iter_bvec(&iter, ITER_SOURCE, bvec, nolb, len);
470 	ret = vfs_iter_write(fd_dev->fd_file, &iter, &pos, 0);
471 
472 	kfree(bvec);
473 	if (ret < 0 || ret != len) {
474 		pr_err("vfs_iter_write() returned %zd for write same\n", ret);
475 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
476 	}
477 
478 	target_complete_cmd(cmd, SAM_STAT_GOOD);
479 	return 0;
480 }
481 
482 static int
483 fd_do_prot_fill(struct se_device *se_dev, sector_t lba, sector_t nolb,
484 		void *buf, size_t bufsize)
485 {
486 	struct fd_dev *fd_dev = FD_DEV(se_dev);
487 	struct file *prot_fd = fd_dev->fd_prot_file;
488 	sector_t prot_length, prot;
489 	loff_t pos = lba * se_dev->prot_length;
490 
491 	if (!prot_fd) {
492 		pr_err("Unable to locate fd_dev->fd_prot_file\n");
493 		return -ENODEV;
494 	}
495 
496 	prot_length = nolb * se_dev->prot_length;
497 
498 	memset(buf, 0xff, bufsize);
499 	for (prot = 0; prot < prot_length;) {
500 		sector_t len = min_t(sector_t, bufsize, prot_length - prot);
501 		ssize_t ret = kernel_write(prot_fd, buf, len, &pos);
502 
503 		if (ret != len) {
504 			pr_err("vfs_write to prot file failed: %zd\n", ret);
505 			return ret < 0 ? ret : -ENODEV;
506 		}
507 		prot += ret;
508 	}
509 
510 	return 0;
511 }
512 
513 static int
514 fd_do_prot_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
515 {
516 	void *buf;
517 	int rc;
518 
519 	buf = (void *)__get_free_page(GFP_KERNEL);
520 	if (!buf) {
521 		pr_err("Unable to allocate FILEIO prot buf\n");
522 		return -ENOMEM;
523 	}
524 
525 	rc = fd_do_prot_fill(cmd->se_dev, lba, nolb, buf, PAGE_SIZE);
526 
527 	free_page((unsigned long)buf);
528 
529 	return rc;
530 }
531 
532 static sense_reason_t
533 fd_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
534 {
535 	struct file *file = FD_DEV(cmd->se_dev)->fd_file;
536 	struct inode *inode = file->f_mapping->host;
537 	int ret;
538 
539 	if (!nolb) {
540 		return 0;
541 	}
542 
543 	if (cmd->se_dev->dev_attrib.pi_prot_type) {
544 		ret = fd_do_prot_unmap(cmd, lba, nolb);
545 		if (ret)
546 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
547 	}
548 
549 	if (S_ISBLK(inode->i_mode)) {
550 		/* The backend is block device, use discard */
551 		struct block_device *bdev = I_BDEV(inode);
552 		struct se_device *dev = cmd->se_dev;
553 
554 		ret = blkdev_issue_discard(bdev,
555 					   target_to_linux_sector(dev, lba),
556 					   target_to_linux_sector(dev,  nolb),
557 					   GFP_KERNEL);
558 		if (ret < 0) {
559 			pr_warn("FILEIO: blkdev_issue_discard() failed: %d\n",
560 				ret);
561 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
562 		}
563 	} else {
564 		/* The backend is normal file, use fallocate */
565 		struct se_device *se_dev = cmd->se_dev;
566 		loff_t pos = lba * se_dev->dev_attrib.block_size;
567 		unsigned int len = nolb * se_dev->dev_attrib.block_size;
568 		int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
569 
570 		if (!file->f_op->fallocate)
571 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
572 
573 		ret = file->f_op->fallocate(file, mode, pos, len);
574 		if (ret < 0) {
575 			pr_warn("FILEIO: fallocate() failed: %d\n", ret);
576 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
577 		}
578 	}
579 
580 	return 0;
581 }
582 
583 static sense_reason_t
584 fd_execute_rw_buffered(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
585 	      enum dma_data_direction data_direction)
586 {
587 	struct se_device *dev = cmd->se_dev;
588 	struct fd_dev *fd_dev = FD_DEV(dev);
589 	struct file *file = fd_dev->fd_file;
590 	struct file *pfile = fd_dev->fd_prot_file;
591 	sense_reason_t rc;
592 	int ret = 0;
593 	/*
594 	 * Call vectorized fileio functions to map struct scatterlist
595 	 * physical memory addresses to struct iovec virtual memory.
596 	 */
597 	if (data_direction == DMA_FROM_DEVICE) {
598 		if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
599 			ret = fd_do_rw(cmd, pfile, dev->prot_length,
600 				       cmd->t_prot_sg, cmd->t_prot_nents,
601 				       cmd->prot_length, 0);
602 			if (ret < 0)
603 				return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
604 		}
605 
606 		ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
607 			       sgl, sgl_nents, cmd->data_length, 0);
608 
609 		if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type &&
610 		    dev->dev_attrib.pi_prot_verify) {
611 			u32 sectors = cmd->data_length >>
612 					ilog2(dev->dev_attrib.block_size);
613 
614 			rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
615 					    0, cmd->t_prot_sg, 0);
616 			if (rc)
617 				return rc;
618 		}
619 	} else {
620 		if (cmd->prot_type && dev->dev_attrib.pi_prot_type &&
621 		    dev->dev_attrib.pi_prot_verify) {
622 			u32 sectors = cmd->data_length >>
623 					ilog2(dev->dev_attrib.block_size);
624 
625 			rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
626 					    0, cmd->t_prot_sg, 0);
627 			if (rc)
628 				return rc;
629 		}
630 
631 		ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
632 			       sgl, sgl_nents, cmd->data_length, 1);
633 		/*
634 		 * Perform implicit vfs_fsync_range() for fd_do_writev() ops
635 		 * for SCSI WRITEs with Forced Unit Access (FUA) set.
636 		 * Allow this to happen independent of WCE=0 setting.
637 		 */
638 		if (ret > 0 && (cmd->se_cmd_flags & SCF_FUA)) {
639 			loff_t start = cmd->t_task_lba *
640 				dev->dev_attrib.block_size;
641 			loff_t end;
642 
643 			if (cmd->data_length)
644 				end = start + cmd->data_length - 1;
645 			else
646 				end = LLONG_MAX;
647 
648 			vfs_fsync_range(fd_dev->fd_file, start, end, 1);
649 		}
650 
651 		if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type) {
652 			ret = fd_do_rw(cmd, pfile, dev->prot_length,
653 				       cmd->t_prot_sg, cmd->t_prot_nents,
654 				       cmd->prot_length, 1);
655 			if (ret < 0)
656 				return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
657 		}
658 	}
659 
660 	if (ret < 0)
661 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
662 
663 	target_complete_cmd(cmd, SAM_STAT_GOOD);
664 	return 0;
665 }
666 
667 static sense_reason_t
668 fd_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
669 	      enum dma_data_direction data_direction)
670 {
671 	struct se_device *dev = cmd->se_dev;
672 	struct fd_dev *fd_dev = FD_DEV(dev);
673 
674 	/*
675 	 * We are currently limited by the number of iovecs (2048) per
676 	 * single vfs_[writev,readv] call.
677 	 */
678 	if (cmd->data_length > FD_MAX_BYTES) {
679 		pr_err("FILEIO: Not able to process I/O of %u bytes due to"
680 		       "FD_MAX_BYTES: %u iovec count limitation\n",
681 			cmd->data_length, FD_MAX_BYTES);
682 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
683 	}
684 
685 	if (fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO)
686 		return fd_execute_rw_aio(cmd, sgl, sgl_nents, data_direction);
687 	return fd_execute_rw_buffered(cmd, sgl, sgl_nents, data_direction);
688 }
689 
690 enum {
691 	Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io,
692 	Opt_fd_async_io, Opt_err
693 };
694 
695 static match_table_t tokens = {
696 	{Opt_fd_dev_name, "fd_dev_name=%s"},
697 	{Opt_fd_dev_size, "fd_dev_size=%s"},
698 	{Opt_fd_buffered_io, "fd_buffered_io=%d"},
699 	{Opt_fd_async_io, "fd_async_io=%d"},
700 	{Opt_err, NULL}
701 };
702 
703 static ssize_t fd_set_configfs_dev_params(struct se_device *dev,
704 		const char *page, ssize_t count)
705 {
706 	struct fd_dev *fd_dev = FD_DEV(dev);
707 	char *orig, *ptr, *arg_p, *opts;
708 	substring_t args[MAX_OPT_ARGS];
709 	int ret = 0, arg, token;
710 
711 	opts = kstrdup(page, GFP_KERNEL);
712 	if (!opts)
713 		return -ENOMEM;
714 
715 	orig = opts;
716 
717 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
718 		if (!*ptr)
719 			continue;
720 
721 		token = match_token(ptr, tokens, args);
722 		switch (token) {
723 		case Opt_fd_dev_name:
724 			if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
725 				FD_MAX_DEV_NAME) == 0) {
726 				ret = -EINVAL;
727 				break;
728 			}
729 			pr_debug("FILEIO: Referencing Path: %s\n",
730 					fd_dev->fd_dev_name);
731 			fd_dev->fbd_flags |= FBDF_HAS_PATH;
732 			break;
733 		case Opt_fd_dev_size:
734 			arg_p = match_strdup(&args[0]);
735 			if (!arg_p) {
736 				ret = -ENOMEM;
737 				break;
738 			}
739 			ret = kstrtoull(arg_p, 0, &fd_dev->fd_dev_size);
740 			kfree(arg_p);
741 			if (ret < 0) {
742 				pr_err("kstrtoull() failed for"
743 						" fd_dev_size=\n");
744 				goto out;
745 			}
746 			pr_debug("FILEIO: Referencing Size: %llu"
747 					" bytes\n", fd_dev->fd_dev_size);
748 			fd_dev->fbd_flags |= FBDF_HAS_SIZE;
749 			break;
750 		case Opt_fd_buffered_io:
751 			ret = match_int(args, &arg);
752 			if (ret)
753 				goto out;
754 			if (arg != 1) {
755 				pr_err("bogus fd_buffered_io=%d value\n", arg);
756 				ret = -EINVAL;
757 				goto out;
758 			}
759 
760 			pr_debug("FILEIO: Using buffered I/O"
761 				" operations for struct fd_dev\n");
762 
763 			fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
764 			break;
765 		case Opt_fd_async_io:
766 			ret = match_int(args, &arg);
767 			if (ret)
768 				goto out;
769 			if (arg != 1) {
770 				pr_err("bogus fd_async_io=%d value\n", arg);
771 				ret = -EINVAL;
772 				goto out;
773 			}
774 
775 			pr_debug("FILEIO: Using async I/O"
776 				" operations for struct fd_dev\n");
777 
778 			fd_dev->fbd_flags |= FDBD_HAS_ASYNC_IO;
779 			break;
780 		default:
781 			break;
782 		}
783 	}
784 
785 out:
786 	kfree(orig);
787 	return (!ret) ? count : ret;
788 }
789 
790 static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b)
791 {
792 	struct fd_dev *fd_dev = FD_DEV(dev);
793 	ssize_t bl = 0;
794 
795 	bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
796 	bl += sprintf(b + bl, "        File: %s  Size: %llu  Mode: %s Async: %d\n",
797 		fd_dev->fd_dev_name, fd_dev->fd_dev_size,
798 		(fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
799 		"Buffered-WCE" : "O_DSYNC",
800 		!!(fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO));
801 	return bl;
802 }
803 
804 static sector_t fd_get_blocks(struct se_device *dev)
805 {
806 	struct fd_dev *fd_dev = FD_DEV(dev);
807 	struct file *f = fd_dev->fd_file;
808 	struct inode *i = f->f_mapping->host;
809 	unsigned long long dev_size;
810 	/*
811 	 * When using a file that references an underlying struct block_device,
812 	 * ensure dev_size is always based on the current inode size in order
813 	 * to handle underlying block_device resize operations.
814 	 */
815 	if (S_ISBLK(i->i_mode))
816 		dev_size = i_size_read(i);
817 	else
818 		dev_size = fd_dev->fd_dev_size;
819 
820 	return div_u64(dev_size - dev->dev_attrib.block_size,
821 		       dev->dev_attrib.block_size);
822 }
823 
824 static int fd_init_prot(struct se_device *dev)
825 {
826 	struct fd_dev *fd_dev = FD_DEV(dev);
827 	struct file *prot_file, *file = fd_dev->fd_file;
828 	struct inode *inode;
829 	int ret, flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
830 	char buf[FD_MAX_DEV_PROT_NAME];
831 
832 	if (!file) {
833 		pr_err("Unable to locate fd_dev->fd_file\n");
834 		return -ENODEV;
835 	}
836 
837 	inode = file->f_mapping->host;
838 	if (S_ISBLK(inode->i_mode)) {
839 		pr_err("FILEIO Protection emulation only supported on"
840 		       " !S_ISBLK\n");
841 		return -ENOSYS;
842 	}
843 
844 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE)
845 		flags &= ~O_DSYNC;
846 
847 	snprintf(buf, FD_MAX_DEV_PROT_NAME, "%s.protection",
848 		 fd_dev->fd_dev_name);
849 
850 	prot_file = filp_open(buf, flags, 0600);
851 	if (IS_ERR(prot_file)) {
852 		pr_err("filp_open(%s) failed\n", buf);
853 		ret = PTR_ERR(prot_file);
854 		return ret;
855 	}
856 	fd_dev->fd_prot_file = prot_file;
857 
858 	return 0;
859 }
860 
861 static int fd_format_prot(struct se_device *dev)
862 {
863 	unsigned char *buf;
864 	int unit_size = FDBD_FORMAT_UNIT_SIZE * dev->dev_attrib.block_size;
865 	int ret;
866 
867 	if (!dev->dev_attrib.pi_prot_type) {
868 		pr_err("Unable to format_prot while pi_prot_type == 0\n");
869 		return -ENODEV;
870 	}
871 
872 	buf = vzalloc(unit_size);
873 	if (!buf) {
874 		pr_err("Unable to allocate FILEIO prot buf\n");
875 		return -ENOMEM;
876 	}
877 
878 	pr_debug("Using FILEIO prot_length: %llu\n",
879 		 (unsigned long long)(dev->transport->get_blocks(dev) + 1) *
880 					dev->prot_length);
881 
882 	ret = fd_do_prot_fill(dev, 0, dev->transport->get_blocks(dev) + 1,
883 			      buf, unit_size);
884 	vfree(buf);
885 	return ret;
886 }
887 
888 static void fd_free_prot(struct se_device *dev)
889 {
890 	struct fd_dev *fd_dev = FD_DEV(dev);
891 
892 	if (!fd_dev->fd_prot_file)
893 		return;
894 
895 	filp_close(fd_dev->fd_prot_file, NULL);
896 	fd_dev->fd_prot_file = NULL;
897 }
898 
899 static struct exec_cmd_ops fd_exec_cmd_ops = {
900 	.execute_rw		= fd_execute_rw,
901 	.execute_sync_cache	= fd_execute_sync_cache,
902 	.execute_write_same	= fd_execute_write_same,
903 	.execute_unmap		= fd_execute_unmap,
904 };
905 
906 static sense_reason_t
907 fd_parse_cdb(struct se_cmd *cmd)
908 {
909 	return sbc_parse_cdb(cmd, &fd_exec_cmd_ops);
910 }
911 
912 static const struct target_backend_ops fileio_ops = {
913 	.name			= "fileio",
914 	.inquiry_prod		= "FILEIO",
915 	.inquiry_rev		= FD_VERSION,
916 	.owner			= THIS_MODULE,
917 	.attach_hba		= fd_attach_hba,
918 	.detach_hba		= fd_detach_hba,
919 	.alloc_device		= fd_alloc_device,
920 	.configure_device	= fd_configure_device,
921 	.destroy_device		= fd_destroy_device,
922 	.free_device		= fd_free_device,
923 	.configure_unmap	= fd_configure_unmap,
924 	.parse_cdb		= fd_parse_cdb,
925 	.set_configfs_dev_params = fd_set_configfs_dev_params,
926 	.show_configfs_dev_params = fd_show_configfs_dev_params,
927 	.get_device_type	= sbc_get_device_type,
928 	.get_blocks		= fd_get_blocks,
929 	.init_prot		= fd_init_prot,
930 	.format_prot		= fd_format_prot,
931 	.free_prot		= fd_free_prot,
932 	.tb_dev_attrib_attrs	= sbc_attrib_attrs,
933 };
934 
935 static int __init fileio_module_init(void)
936 {
937 	return transport_backend_register(&fileio_ops);
938 }
939 
940 static void __exit fileio_module_exit(void)
941 {
942 	target_backend_unregister(&fileio_ops);
943 }
944 
945 MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
946 MODULE_AUTHOR("nab@Linux-iSCSI.org");
947 MODULE_LICENSE("GPL");
948 
949 module_init(fileio_module_init);
950 module_exit(fileio_module_exit);
951