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/falloc.h>
34 #include <scsi/scsi.h>
35 #include <scsi/scsi_host.h>
36 #include <asm/unaligned.h>
37 
38 #include <target/target_core_base.h>
39 #include <target/target_core_backend.h>
40 #include <target/target_core_backend_configfs.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 /*	fd_attach_hba(): (Part of se_subsystem_api_t template)
50  *
51  *
52  */
53 static int fd_attach_hba(struct se_hba *hba, u32 host_id)
54 {
55 	struct fd_host *fd_host;
56 
57 	fd_host = kzalloc(sizeof(struct fd_host), GFP_KERNEL);
58 	if (!fd_host) {
59 		pr_err("Unable to allocate memory for struct fd_host\n");
60 		return -ENOMEM;
61 	}
62 
63 	fd_host->fd_host_id = host_id;
64 
65 	hba->hba_ptr = fd_host;
66 
67 	pr_debug("CORE_HBA[%d] - TCM FILEIO HBA Driver %s on Generic"
68 		" Target Core Stack %s\n", hba->hba_id, FD_VERSION,
69 		TARGET_CORE_MOD_VERSION);
70 	pr_debug("CORE_HBA[%d] - Attached FILEIO HBA: %u to Generic\n",
71 		hba->hba_id, fd_host->fd_host_id);
72 
73 	return 0;
74 }
75 
76 static void fd_detach_hba(struct se_hba *hba)
77 {
78 	struct fd_host *fd_host = hba->hba_ptr;
79 
80 	pr_debug("CORE_HBA[%d] - Detached FILEIO HBA: %u from Generic"
81 		" Target Core\n", hba->hba_id, fd_host->fd_host_id);
82 
83 	kfree(fd_host);
84 	hba->hba_ptr = NULL;
85 }
86 
87 static struct se_device *fd_alloc_device(struct se_hba *hba, const char *name)
88 {
89 	struct fd_dev *fd_dev;
90 	struct fd_host *fd_host = hba->hba_ptr;
91 
92 	fd_dev = kzalloc(sizeof(struct fd_dev), GFP_KERNEL);
93 	if (!fd_dev) {
94 		pr_err("Unable to allocate memory for struct fd_dev\n");
95 		return NULL;
96 	}
97 
98 	fd_dev->fd_host = fd_host;
99 
100 	pr_debug("FILEIO: Allocated fd_dev for %p\n", name);
101 
102 	return &fd_dev->dev;
103 }
104 
105 static int fd_configure_device(struct se_device *dev)
106 {
107 	struct fd_dev *fd_dev = FD_DEV(dev);
108 	struct fd_host *fd_host = dev->se_hba->hba_ptr;
109 	struct file *file;
110 	struct inode *inode = NULL;
111 	int flags, ret = -EINVAL;
112 
113 	if (!(fd_dev->fbd_flags & FBDF_HAS_PATH)) {
114 		pr_err("Missing fd_dev_name=\n");
115 		return -EINVAL;
116 	}
117 
118 	/*
119 	 * Use O_DSYNC by default instead of O_SYNC to forgo syncing
120 	 * of pure timestamp updates.
121 	 */
122 	flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
123 
124 	/*
125 	 * Optionally allow fd_buffered_io=1 to be enabled for people
126 	 * who want use the fs buffer cache as an WriteCache mechanism.
127 	 *
128 	 * This means that in event of a hard failure, there is a risk
129 	 * of silent data-loss if the SCSI client has *not* performed a
130 	 * forced unit access (FUA) write, or issued SYNCHRONIZE_CACHE
131 	 * to write-out the entire device cache.
132 	 */
133 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
134 		pr_debug("FILEIO: Disabling O_DSYNC, using buffered FILEIO\n");
135 		flags &= ~O_DSYNC;
136 	}
137 
138 	file = filp_open(fd_dev->fd_dev_name, flags, 0600);
139 	if (IS_ERR(file)) {
140 		pr_err("filp_open(%s) failed\n", fd_dev->fd_dev_name);
141 		ret = PTR_ERR(file);
142 		goto fail;
143 	}
144 	fd_dev->fd_file = file;
145 	/*
146 	 * If using a block backend with this struct file, we extract
147 	 * fd_dev->fd_[block,dev]_size from struct block_device.
148 	 *
149 	 * Otherwise, we use the passed fd_size= from configfs
150 	 */
151 	inode = file->f_mapping->host;
152 	if (S_ISBLK(inode->i_mode)) {
153 		struct request_queue *q = bdev_get_queue(inode->i_bdev);
154 		unsigned long long dev_size;
155 
156 		fd_dev->fd_block_size = bdev_logical_block_size(inode->i_bdev);
157 		/*
158 		 * Determine the number of bytes from i_size_read() minus
159 		 * one (1) logical sector from underlying struct block_device
160 		 */
161 		dev_size = (i_size_read(file->f_mapping->host) -
162 				       fd_dev->fd_block_size);
163 
164 		pr_debug("FILEIO: Using size: %llu bytes from struct"
165 			" block_device blocks: %llu logical_block_size: %d\n",
166 			dev_size, div_u64(dev_size, fd_dev->fd_block_size),
167 			fd_dev->fd_block_size);
168 		/*
169 		 * Check if the underlying struct block_device request_queue supports
170 		 * the QUEUE_FLAG_DISCARD bit for UNMAP/WRITE_SAME in SCSI + TRIM
171 		 * in ATA and we need to set TPE=1
172 		 */
173 		if (blk_queue_discard(q)) {
174 			dev->dev_attrib.max_unmap_lba_count =
175 				q->limits.max_discard_sectors;
176 			/*
177 			 * Currently hardcoded to 1 in Linux/SCSI code..
178 			 */
179 			dev->dev_attrib.max_unmap_block_desc_count = 1;
180 			dev->dev_attrib.unmap_granularity =
181 				q->limits.discard_granularity >> 9;
182 			dev->dev_attrib.unmap_granularity_alignment =
183 				q->limits.discard_alignment;
184 			pr_debug("IFILE: BLOCK Discard support available,"
185 					" disabled by default\n");
186 		}
187 		/*
188 		 * Enable write same emulation for IBLOCK and use 0xFFFF as
189 		 * the smaller WRITE_SAME(10) only has a two-byte block count.
190 		 */
191 		dev->dev_attrib.max_write_same_len = 0xFFFF;
192 
193 		if (blk_queue_nonrot(q))
194 			dev->dev_attrib.is_nonrot = 1;
195 	} else {
196 		if (!(fd_dev->fbd_flags & FBDF_HAS_SIZE)) {
197 			pr_err("FILEIO: Missing fd_dev_size="
198 				" parameter, and no backing struct"
199 				" block_device\n");
200 			goto fail;
201 		}
202 
203 		fd_dev->fd_block_size = FD_BLOCKSIZE;
204 		/*
205 		 * Limit UNMAP emulation to 8k Number of LBAs (NoLB)
206 		 */
207 		dev->dev_attrib.max_unmap_lba_count = 0x2000;
208 		/*
209 		 * Currently hardcoded to 1 in Linux/SCSI code..
210 		 */
211 		dev->dev_attrib.max_unmap_block_desc_count = 1;
212 		dev->dev_attrib.unmap_granularity = 1;
213 		dev->dev_attrib.unmap_granularity_alignment = 0;
214 
215 		/*
216 		 * Limit WRITE_SAME w/ UNMAP=0 emulation to 8k Number of LBAs (NoLB)
217 		 * based upon struct iovec limit for vfs_writev()
218 		 */
219 		dev->dev_attrib.max_write_same_len = 0x1000;
220 	}
221 
222 	dev->dev_attrib.hw_block_size = fd_dev->fd_block_size;
223 	dev->dev_attrib.max_bytes_per_io = FD_MAX_BYTES;
224 	dev->dev_attrib.hw_max_sectors = FD_MAX_BYTES / fd_dev->fd_block_size;
225 	dev->dev_attrib.hw_queue_depth = FD_MAX_DEVICE_QUEUE_DEPTH;
226 
227 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
228 		pr_debug("FILEIO: Forcing setting of emulate_write_cache=1"
229 			" with FDBD_HAS_BUFFERED_IO_WCE\n");
230 		dev->dev_attrib.emulate_write_cache = 1;
231 	}
232 
233 	fd_dev->fd_dev_id = fd_host->fd_host_dev_id_count++;
234 	fd_dev->fd_queue_depth = dev->queue_depth;
235 
236 	pr_debug("CORE_FILE[%u] - Added TCM FILEIO Device ID: %u at %s,"
237 		" %llu total bytes\n", fd_host->fd_host_id, fd_dev->fd_dev_id,
238 			fd_dev->fd_dev_name, fd_dev->fd_dev_size);
239 
240 	return 0;
241 fail:
242 	if (fd_dev->fd_file) {
243 		filp_close(fd_dev->fd_file, NULL);
244 		fd_dev->fd_file = NULL;
245 	}
246 	return ret;
247 }
248 
249 static void fd_free_device(struct se_device *dev)
250 {
251 	struct fd_dev *fd_dev = FD_DEV(dev);
252 
253 	if (fd_dev->fd_file) {
254 		filp_close(fd_dev->fd_file, NULL);
255 		fd_dev->fd_file = NULL;
256 	}
257 
258 	kfree(fd_dev);
259 }
260 
261 static int fd_do_prot_rw(struct se_cmd *cmd, struct fd_prot *fd_prot,
262 			 int is_write)
263 {
264 	struct se_device *se_dev = cmd->se_dev;
265 	struct fd_dev *dev = FD_DEV(se_dev);
266 	struct file *prot_fd = dev->fd_prot_file;
267 	loff_t pos = (cmd->t_task_lba * se_dev->prot_length);
268 	unsigned char *buf;
269 	u32 prot_size;
270 	int rc, ret = 1;
271 
272 	prot_size = (cmd->data_length / se_dev->dev_attrib.block_size) *
273 		     se_dev->prot_length;
274 
275 	if (!is_write) {
276 		fd_prot->prot_buf = kzalloc(prot_size, GFP_KERNEL);
277 		if (!fd_prot->prot_buf) {
278 			pr_err("Unable to allocate fd_prot->prot_buf\n");
279 			return -ENOMEM;
280 		}
281 		buf = fd_prot->prot_buf;
282 
283 		fd_prot->prot_sg_nents = 1;
284 		fd_prot->prot_sg = kzalloc(sizeof(struct scatterlist),
285 					   GFP_KERNEL);
286 		if (!fd_prot->prot_sg) {
287 			pr_err("Unable to allocate fd_prot->prot_sg\n");
288 			kfree(fd_prot->prot_buf);
289 			return -ENOMEM;
290 		}
291 		sg_init_table(fd_prot->prot_sg, fd_prot->prot_sg_nents);
292 		sg_set_buf(fd_prot->prot_sg, buf, prot_size);
293 	}
294 
295 	if (is_write) {
296 		rc = kernel_write(prot_fd, fd_prot->prot_buf, prot_size, pos);
297 		if (rc < 0 || prot_size != rc) {
298 			pr_err("kernel_write() for fd_do_prot_rw failed:"
299 			       " %d\n", rc);
300 			ret = -EINVAL;
301 		}
302 	} else {
303 		rc = kernel_read(prot_fd, pos, fd_prot->prot_buf, prot_size);
304 		if (rc < 0) {
305 			pr_err("kernel_read() for fd_do_prot_rw failed:"
306 			       " %d\n", rc);
307 			ret = -EINVAL;
308 		}
309 	}
310 
311 	if (is_write || ret < 0) {
312 		kfree(fd_prot->prot_sg);
313 		kfree(fd_prot->prot_buf);
314 	}
315 
316 	return ret;
317 }
318 
319 static int fd_do_rw(struct se_cmd *cmd, struct scatterlist *sgl,
320 		u32 sgl_nents, int is_write)
321 {
322 	struct se_device *se_dev = cmd->se_dev;
323 	struct fd_dev *dev = FD_DEV(se_dev);
324 	struct file *fd = dev->fd_file;
325 	struct scatterlist *sg;
326 	struct iov_iter iter;
327 	struct bio_vec *bvec;
328 	ssize_t len = 0;
329 	loff_t pos = (cmd->t_task_lba * se_dev->dev_attrib.block_size);
330 	int ret = 0, i;
331 
332 	bvec = kcalloc(sgl_nents, sizeof(struct bio_vec), GFP_KERNEL);
333 	if (!bvec) {
334 		pr_err("Unable to allocate fd_do_readv iov[]\n");
335 		return -ENOMEM;
336 	}
337 
338 	for_each_sg(sgl, sg, sgl_nents, i) {
339 		bvec[i].bv_page = sg_page(sg);
340 		bvec[i].bv_len = sg->length;
341 		bvec[i].bv_offset = sg->offset;
342 
343 		len += sg->length;
344 	}
345 
346 	iov_iter_bvec(&iter, ITER_BVEC, bvec, sgl_nents, len);
347 	if (is_write)
348 		ret = vfs_iter_write(fd, &iter, &pos);
349 	else
350 		ret = vfs_iter_read(fd, &iter, &pos);
351 
352 	kfree(bvec);
353 
354 	if (is_write) {
355 		if (ret < 0 || ret != cmd->data_length) {
356 			pr_err("%s() write returned %d\n", __func__, ret);
357 			return (ret < 0 ? ret : -EINVAL);
358 		}
359 	} else {
360 		/*
361 		 * Return zeros and GOOD status even if the READ did not return
362 		 * the expected virt_size for struct file w/o a backing struct
363 		 * block_device.
364 		 */
365 		if (S_ISBLK(file_inode(fd)->i_mode)) {
366 			if (ret < 0 || ret != cmd->data_length) {
367 				pr_err("%s() returned %d, expecting %u for "
368 						"S_ISBLK\n", __func__, ret,
369 						cmd->data_length);
370 				return (ret < 0 ? ret : -EINVAL);
371 			}
372 		} else {
373 			if (ret < 0) {
374 				pr_err("%s() returned %d for non S_ISBLK\n",
375 						__func__, ret);
376 				return ret;
377 			}
378 		}
379 	}
380 	return 1;
381 }
382 
383 static sense_reason_t
384 fd_execute_sync_cache(struct se_cmd *cmd)
385 {
386 	struct se_device *dev = cmd->se_dev;
387 	struct fd_dev *fd_dev = FD_DEV(dev);
388 	int immed = (cmd->t_task_cdb[1] & 0x2);
389 	loff_t start, end;
390 	int ret;
391 
392 	/*
393 	 * If the Immediate bit is set, queue up the GOOD response
394 	 * for this SYNCHRONIZE_CACHE op
395 	 */
396 	if (immed)
397 		target_complete_cmd(cmd, SAM_STAT_GOOD);
398 
399 	/*
400 	 * Determine if we will be flushing the entire device.
401 	 */
402 	if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
403 		start = 0;
404 		end = LLONG_MAX;
405 	} else {
406 		start = cmd->t_task_lba * dev->dev_attrib.block_size;
407 		if (cmd->data_length)
408 			end = start + cmd->data_length - 1;
409 		else
410 			end = LLONG_MAX;
411 	}
412 
413 	ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
414 	if (ret != 0)
415 		pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
416 
417 	if (immed)
418 		return 0;
419 
420 	if (ret)
421 		target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
422 	else
423 		target_complete_cmd(cmd, SAM_STAT_GOOD);
424 
425 	return 0;
426 }
427 
428 static sense_reason_t
429 fd_execute_write_same(struct se_cmd *cmd)
430 {
431 	struct se_device *se_dev = cmd->se_dev;
432 	struct fd_dev *fd_dev = FD_DEV(se_dev);
433 	loff_t pos = cmd->t_task_lba * se_dev->dev_attrib.block_size;
434 	sector_t nolb = sbc_get_write_same_sectors(cmd);
435 	struct iov_iter iter;
436 	struct bio_vec *bvec;
437 	unsigned int len = 0, i;
438 	ssize_t ret;
439 
440 	if (!nolb) {
441 		target_complete_cmd(cmd, SAM_STAT_GOOD);
442 		return 0;
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 > 1 ||
451 	    cmd->t_data_sg[0].length != cmd->se_dev->dev_attrib.block_size) {
452 		pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
453 			" block_size: %u\n",
454 			cmd->t_data_nents,
455 			cmd->t_data_sg[0].length,
456 			cmd->se_dev->dev_attrib.block_size);
457 		return TCM_INVALID_CDB_FIELD;
458 	}
459 
460 	bvec = kcalloc(nolb, sizeof(struct bio_vec), GFP_KERNEL);
461 	if (!bvec)
462 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
463 
464 	for (i = 0; i < nolb; i++) {
465 		bvec[i].bv_page = sg_page(&cmd->t_data_sg[0]);
466 		bvec[i].bv_len = cmd->t_data_sg[0].length;
467 		bvec[i].bv_offset = cmd->t_data_sg[0].offset;
468 
469 		len += se_dev->dev_attrib.block_size;
470 	}
471 
472 	iov_iter_bvec(&iter, ITER_BVEC, bvec, nolb, len);
473 	ret = vfs_iter_write(fd_dev->fd_file, &iter, &pos);
474 
475 	kfree(bvec);
476 	if (ret < 0 || ret != len) {
477 		pr_err("vfs_iter_write() returned %zd for write same\n", ret);
478 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
479 	}
480 
481 	target_complete_cmd(cmd, SAM_STAT_GOOD);
482 	return 0;
483 }
484 
485 static int
486 fd_do_prot_fill(struct se_device *se_dev, sector_t lba, sector_t nolb,
487 		void *buf, size_t bufsize)
488 {
489 	struct fd_dev *fd_dev = FD_DEV(se_dev);
490 	struct file *prot_fd = fd_dev->fd_prot_file;
491 	sector_t prot_length, prot;
492 	loff_t pos = lba * se_dev->prot_length;
493 
494 	if (!prot_fd) {
495 		pr_err("Unable to locate fd_dev->fd_prot_file\n");
496 		return -ENODEV;
497 	}
498 
499 	prot_length = nolb * se_dev->prot_length;
500 
501 	for (prot = 0; prot < prot_length;) {
502 		sector_t len = min_t(sector_t, bufsize, prot_length - prot);
503 		ssize_t ret = kernel_write(prot_fd, buf, len, pos + prot);
504 
505 		if (ret != len) {
506 			pr_err("vfs_write to prot file failed: %zd\n", ret);
507 			return ret < 0 ? ret : -ENODEV;
508 		}
509 		prot += ret;
510 	}
511 
512 	return 0;
513 }
514 
515 static int
516 fd_do_prot_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
517 {
518 	void *buf;
519 	int rc;
520 
521 	buf = (void *)__get_free_page(GFP_KERNEL);
522 	if (!buf) {
523 		pr_err("Unable to allocate FILEIO prot buf\n");
524 		return -ENOMEM;
525 	}
526 	memset(buf, 0xff, PAGE_SIZE);
527 
528 	rc = fd_do_prot_fill(cmd->se_dev, lba, nolb, buf, PAGE_SIZE);
529 
530 	free_page((unsigned long)buf);
531 
532 	return rc;
533 }
534 
535 static sense_reason_t
536 fd_do_unmap(struct se_cmd *cmd, void *priv, sector_t lba, sector_t nolb)
537 {
538 	struct file *file = priv;
539 	struct inode *inode = file->f_mapping->host;
540 	int ret;
541 
542 	if (cmd->se_dev->dev_attrib.pi_prot_type) {
543 		ret = fd_do_prot_unmap(cmd, lba, nolb);
544 		if (ret)
545 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
546 	}
547 
548 	if (S_ISBLK(inode->i_mode)) {
549 		/* The backend is block device, use discard */
550 		struct block_device *bdev = inode->i_bdev;
551 
552 		ret = blkdev_issue_discard(bdev, lba,
553 				nolb, GFP_KERNEL, 0);
554 		if (ret < 0) {
555 			pr_warn("FILEIO: blkdev_issue_discard() failed: %d\n",
556 				ret);
557 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
558 		}
559 	} else {
560 		/* The backend is normal file, use fallocate */
561 		struct se_device *se_dev = cmd->se_dev;
562 		loff_t pos = lba * se_dev->dev_attrib.block_size;
563 		unsigned int len = nolb * se_dev->dev_attrib.block_size;
564 		int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
565 
566 		if (!file->f_op->fallocate)
567 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
568 
569 		ret = file->f_op->fallocate(file, mode, pos, len);
570 		if (ret < 0) {
571 			pr_warn("FILEIO: fallocate() failed: %d\n", ret);
572 			return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
573 		}
574 	}
575 
576 	return 0;
577 }
578 
579 static sense_reason_t
580 fd_execute_write_same_unmap(struct se_cmd *cmd)
581 {
582 	struct se_device *se_dev = cmd->se_dev;
583 	struct fd_dev *fd_dev = FD_DEV(se_dev);
584 	struct file *file = fd_dev->fd_file;
585 	sector_t lba = cmd->t_task_lba;
586 	sector_t nolb = sbc_get_write_same_sectors(cmd);
587 	sense_reason_t ret;
588 
589 	if (!nolb) {
590 		target_complete_cmd(cmd, SAM_STAT_GOOD);
591 		return 0;
592 	}
593 
594 	ret = fd_do_unmap(cmd, file, lba, nolb);
595 	if (ret)
596 		return ret;
597 
598 	target_complete_cmd(cmd, GOOD);
599 	return 0;
600 }
601 
602 static sense_reason_t
603 fd_execute_unmap(struct se_cmd *cmd)
604 {
605 	struct file *file = FD_DEV(cmd->se_dev)->fd_file;
606 
607 	return sbc_execute_unmap(cmd, fd_do_unmap, file);
608 }
609 
610 static sense_reason_t
611 fd_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
612 	      enum dma_data_direction data_direction)
613 {
614 	struct se_device *dev = cmd->se_dev;
615 	struct fd_prot fd_prot;
616 	sense_reason_t rc;
617 	int ret = 0;
618 	/*
619 	 * We are currently limited by the number of iovecs (2048) per
620 	 * single vfs_[writev,readv] call.
621 	 */
622 	if (cmd->data_length > FD_MAX_BYTES) {
623 		pr_err("FILEIO: Not able to process I/O of %u bytes due to"
624 		       "FD_MAX_BYTES: %u iovec count limitiation\n",
625 			cmd->data_length, FD_MAX_BYTES);
626 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
627 	}
628 	/*
629 	 * Call vectorized fileio functions to map struct scatterlist
630 	 * physical memory addresses to struct iovec virtual memory.
631 	 */
632 	if (data_direction == DMA_FROM_DEVICE) {
633 		memset(&fd_prot, 0, sizeof(struct fd_prot));
634 
635 		if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
636 			ret = fd_do_prot_rw(cmd, &fd_prot, false);
637 			if (ret < 0)
638 				return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
639 		}
640 
641 		ret = fd_do_rw(cmd, sgl, sgl_nents, 0);
642 
643 		if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type) {
644 			u32 sectors = cmd->data_length / dev->dev_attrib.block_size;
645 
646 			rc = sbc_dif_verify_read(cmd, cmd->t_task_lba, sectors,
647 						 0, fd_prot.prot_sg, 0);
648 			if (rc) {
649 				kfree(fd_prot.prot_sg);
650 				kfree(fd_prot.prot_buf);
651 				return rc;
652 			}
653 			kfree(fd_prot.prot_sg);
654 			kfree(fd_prot.prot_buf);
655 		}
656 	} else {
657 		memset(&fd_prot, 0, sizeof(struct fd_prot));
658 
659 		if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
660 			u32 sectors = cmd->data_length / dev->dev_attrib.block_size;
661 
662 			ret = fd_do_prot_rw(cmd, &fd_prot, false);
663 			if (ret < 0)
664 				return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
665 
666 			rc = sbc_dif_verify_write(cmd, cmd->t_task_lba, sectors,
667 						  0, fd_prot.prot_sg, 0);
668 			if (rc) {
669 				kfree(fd_prot.prot_sg);
670 				kfree(fd_prot.prot_buf);
671 				return rc;
672 			}
673 		}
674 
675 		ret = fd_do_rw(cmd, sgl, sgl_nents, 1);
676 		/*
677 		 * Perform implicit vfs_fsync_range() for fd_do_writev() ops
678 		 * for SCSI WRITEs with Forced Unit Access (FUA) set.
679 		 * Allow this to happen independent of WCE=0 setting.
680 		 */
681 		if (ret > 0 &&
682 		    dev->dev_attrib.emulate_fua_write > 0 &&
683 		    (cmd->se_cmd_flags & SCF_FUA)) {
684 			struct fd_dev *fd_dev = FD_DEV(dev);
685 			loff_t start = cmd->t_task_lba *
686 				dev->dev_attrib.block_size;
687 			loff_t end;
688 
689 			if (cmd->data_length)
690 				end = start + cmd->data_length - 1;
691 			else
692 				end = LLONG_MAX;
693 
694 			vfs_fsync_range(fd_dev->fd_file, start, end, 1);
695 		}
696 
697 		if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type) {
698 			ret = fd_do_prot_rw(cmd, &fd_prot, true);
699 			if (ret < 0)
700 				return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
701 		}
702 	}
703 
704 	if (ret < 0) {
705 		kfree(fd_prot.prot_sg);
706 		kfree(fd_prot.prot_buf);
707 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
708 	}
709 
710 	if (ret)
711 		target_complete_cmd(cmd, SAM_STAT_GOOD);
712 	return 0;
713 }
714 
715 enum {
716 	Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_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_err, NULL}
724 };
725 
726 static ssize_t fd_set_configfs_dev_params(struct se_device *dev,
727 		const char *page, ssize_t count)
728 {
729 	struct fd_dev *fd_dev = FD_DEV(dev);
730 	char *orig, *ptr, *arg_p, *opts;
731 	substring_t args[MAX_OPT_ARGS];
732 	int ret = 0, arg, token;
733 
734 	opts = kstrdup(page, GFP_KERNEL);
735 	if (!opts)
736 		return -ENOMEM;
737 
738 	orig = opts;
739 
740 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
741 		if (!*ptr)
742 			continue;
743 
744 		token = match_token(ptr, tokens, args);
745 		switch (token) {
746 		case Opt_fd_dev_name:
747 			if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
748 				FD_MAX_DEV_NAME) == 0) {
749 				ret = -EINVAL;
750 				break;
751 			}
752 			pr_debug("FILEIO: Referencing Path: %s\n",
753 					fd_dev->fd_dev_name);
754 			fd_dev->fbd_flags |= FBDF_HAS_PATH;
755 			break;
756 		case Opt_fd_dev_size:
757 			arg_p = match_strdup(&args[0]);
758 			if (!arg_p) {
759 				ret = -ENOMEM;
760 				break;
761 			}
762 			ret = kstrtoull(arg_p, 0, &fd_dev->fd_dev_size);
763 			kfree(arg_p);
764 			if (ret < 0) {
765 				pr_err("kstrtoull() failed for"
766 						" fd_dev_size=\n");
767 				goto out;
768 			}
769 			pr_debug("FILEIO: Referencing Size: %llu"
770 					" bytes\n", fd_dev->fd_dev_size);
771 			fd_dev->fbd_flags |= FBDF_HAS_SIZE;
772 			break;
773 		case Opt_fd_buffered_io:
774 			ret = match_int(args, &arg);
775 			if (ret)
776 				goto out;
777 			if (arg != 1) {
778 				pr_err("bogus fd_buffered_io=%d value\n", arg);
779 				ret = -EINVAL;
780 				goto out;
781 			}
782 
783 			pr_debug("FILEIO: Using buffered I/O"
784 				" operations for struct fd_dev\n");
785 
786 			fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
787 			break;
788 		default:
789 			break;
790 		}
791 	}
792 
793 out:
794 	kfree(orig);
795 	return (!ret) ? count : ret;
796 }
797 
798 static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b)
799 {
800 	struct fd_dev *fd_dev = FD_DEV(dev);
801 	ssize_t bl = 0;
802 
803 	bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
804 	bl += sprintf(b + bl, "        File: %s  Size: %llu  Mode: %s\n",
805 		fd_dev->fd_dev_name, fd_dev->fd_dev_size,
806 		(fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
807 		"Buffered-WCE" : "O_DSYNC");
808 	return bl;
809 }
810 
811 static sector_t fd_get_blocks(struct se_device *dev)
812 {
813 	struct fd_dev *fd_dev = FD_DEV(dev);
814 	struct file *f = fd_dev->fd_file;
815 	struct inode *i = f->f_mapping->host;
816 	unsigned long long dev_size;
817 	/*
818 	 * When using a file that references an underlying struct block_device,
819 	 * ensure dev_size is always based on the current inode size in order
820 	 * to handle underlying block_device resize operations.
821 	 */
822 	if (S_ISBLK(i->i_mode))
823 		dev_size = i_size_read(i);
824 	else
825 		dev_size = fd_dev->fd_dev_size;
826 
827 	return div_u64(dev_size - dev->dev_attrib.block_size,
828 		       dev->dev_attrib.block_size);
829 }
830 
831 static int fd_init_prot(struct se_device *dev)
832 {
833 	struct fd_dev *fd_dev = FD_DEV(dev);
834 	struct file *prot_file, *file = fd_dev->fd_file;
835 	struct inode *inode;
836 	int ret, flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
837 	char buf[FD_MAX_DEV_PROT_NAME];
838 
839 	if (!file) {
840 		pr_err("Unable to locate fd_dev->fd_file\n");
841 		return -ENODEV;
842 	}
843 
844 	inode = file->f_mapping->host;
845 	if (S_ISBLK(inode->i_mode)) {
846 		pr_err("FILEIO Protection emulation only supported on"
847 		       " !S_ISBLK\n");
848 		return -ENOSYS;
849 	}
850 
851 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE)
852 		flags &= ~O_DSYNC;
853 
854 	snprintf(buf, FD_MAX_DEV_PROT_NAME, "%s.protection",
855 		 fd_dev->fd_dev_name);
856 
857 	prot_file = filp_open(buf, flags, 0600);
858 	if (IS_ERR(prot_file)) {
859 		pr_err("filp_open(%s) failed\n", buf);
860 		ret = PTR_ERR(prot_file);
861 		return ret;
862 	}
863 	fd_dev->fd_prot_file = prot_file;
864 
865 	return 0;
866 }
867 
868 static int fd_format_prot(struct se_device *dev)
869 {
870 	unsigned char *buf;
871 	int unit_size = FDBD_FORMAT_UNIT_SIZE * dev->dev_attrib.block_size;
872 	int ret;
873 
874 	if (!dev->dev_attrib.pi_prot_type) {
875 		pr_err("Unable to format_prot while pi_prot_type == 0\n");
876 		return -ENODEV;
877 	}
878 
879 	buf = vzalloc(unit_size);
880 	if (!buf) {
881 		pr_err("Unable to allocate FILEIO prot buf\n");
882 		return -ENOMEM;
883 	}
884 
885 	pr_debug("Using FILEIO prot_length: %llu\n",
886 		 (unsigned long long)(dev->transport->get_blocks(dev) + 1) *
887 					dev->prot_length);
888 
889 	memset(buf, 0xff, unit_size);
890 	ret = fd_do_prot_fill(dev, 0, dev->transport->get_blocks(dev) + 1,
891 			      buf, unit_size);
892 	vfree(buf);
893 	return ret;
894 }
895 
896 static void fd_free_prot(struct se_device *dev)
897 {
898 	struct fd_dev *fd_dev = FD_DEV(dev);
899 
900 	if (!fd_dev->fd_prot_file)
901 		return;
902 
903 	filp_close(fd_dev->fd_prot_file, NULL);
904 	fd_dev->fd_prot_file = NULL;
905 }
906 
907 static struct sbc_ops fd_sbc_ops = {
908 	.execute_rw		= fd_execute_rw,
909 	.execute_sync_cache	= fd_execute_sync_cache,
910 	.execute_write_same	= fd_execute_write_same,
911 	.execute_write_same_unmap = fd_execute_write_same_unmap,
912 	.execute_unmap		= fd_execute_unmap,
913 };
914 
915 static sense_reason_t
916 fd_parse_cdb(struct se_cmd *cmd)
917 {
918 	return sbc_parse_cdb(cmd, &fd_sbc_ops);
919 }
920 
921 DEF_TB_DEFAULT_ATTRIBS(fileio);
922 
923 static struct configfs_attribute *fileio_backend_dev_attrs[] = {
924 	&fileio_dev_attrib_emulate_model_alias.attr,
925 	&fileio_dev_attrib_emulate_dpo.attr,
926 	&fileio_dev_attrib_emulate_fua_write.attr,
927 	&fileio_dev_attrib_emulate_fua_read.attr,
928 	&fileio_dev_attrib_emulate_write_cache.attr,
929 	&fileio_dev_attrib_emulate_ua_intlck_ctrl.attr,
930 	&fileio_dev_attrib_emulate_tas.attr,
931 	&fileio_dev_attrib_emulate_tpu.attr,
932 	&fileio_dev_attrib_emulate_tpws.attr,
933 	&fileio_dev_attrib_emulate_caw.attr,
934 	&fileio_dev_attrib_emulate_3pc.attr,
935 	&fileio_dev_attrib_pi_prot_type.attr,
936 	&fileio_dev_attrib_hw_pi_prot_type.attr,
937 	&fileio_dev_attrib_pi_prot_format.attr,
938 	&fileio_dev_attrib_enforce_pr_isids.attr,
939 	&fileio_dev_attrib_is_nonrot.attr,
940 	&fileio_dev_attrib_emulate_rest_reord.attr,
941 	&fileio_dev_attrib_force_pr_aptpl.attr,
942 	&fileio_dev_attrib_hw_block_size.attr,
943 	&fileio_dev_attrib_block_size.attr,
944 	&fileio_dev_attrib_hw_max_sectors.attr,
945 	&fileio_dev_attrib_optimal_sectors.attr,
946 	&fileio_dev_attrib_hw_queue_depth.attr,
947 	&fileio_dev_attrib_queue_depth.attr,
948 	&fileio_dev_attrib_max_unmap_lba_count.attr,
949 	&fileio_dev_attrib_max_unmap_block_desc_count.attr,
950 	&fileio_dev_attrib_unmap_granularity.attr,
951 	&fileio_dev_attrib_unmap_granularity_alignment.attr,
952 	&fileio_dev_attrib_max_write_same_len.attr,
953 	NULL,
954 };
955 
956 static struct se_subsystem_api fileio_template = {
957 	.name			= "fileio",
958 	.inquiry_prod		= "FILEIO",
959 	.inquiry_rev		= FD_VERSION,
960 	.owner			= THIS_MODULE,
961 	.transport_type		= TRANSPORT_PLUGIN_VHBA_PDEV,
962 	.attach_hba		= fd_attach_hba,
963 	.detach_hba		= fd_detach_hba,
964 	.alloc_device		= fd_alloc_device,
965 	.configure_device	= fd_configure_device,
966 	.free_device		= fd_free_device,
967 	.parse_cdb		= fd_parse_cdb,
968 	.set_configfs_dev_params = fd_set_configfs_dev_params,
969 	.show_configfs_dev_params = fd_show_configfs_dev_params,
970 	.get_device_type	= sbc_get_device_type,
971 	.get_blocks		= fd_get_blocks,
972 	.init_prot		= fd_init_prot,
973 	.format_prot		= fd_format_prot,
974 	.free_prot		= fd_free_prot,
975 };
976 
977 static int __init fileio_module_init(void)
978 {
979 	struct target_backend_cits *tbc = &fileio_template.tb_cits;
980 
981 	target_core_setup_sub_cits(&fileio_template);
982 	tbc->tb_dev_attrib_cit.ct_attrs = fileio_backend_dev_attrs;
983 
984 	return transport_subsystem_register(&fileio_template);
985 }
986 
987 static void __exit fileio_module_exit(void)
988 {
989 	transport_subsystem_release(&fileio_template);
990 }
991 
992 MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
993 MODULE_AUTHOR("nab@Linux-iSCSI.org");
994 MODULE_LICENSE("GPL");
995 
996 module_init(fileio_module_init);
997 module_exit(fileio_module_exit);
998