1 /*******************************************************************************
2  * Filename:  target_core_file.c
3  *
4  * This file contains the Storage Engine <-> FILEIO transport specific functions
5  *
6  * Copyright (c) 2005 PyX Technologies, Inc.
7  * Copyright (c) 2005-2006 SBE, Inc.  All Rights Reserved.
8  * Copyright (c) 2007-2010 Rising Tide Systems
9  * Copyright (c) 2008-2010 Linux-iSCSI.org
10  *
11  * Nicholas A. Bellinger <nab@kernel.org>
12  *
13  * This program is free software; you can redistribute it and/or modify
14  * it under the terms of the GNU General Public License as published by
15  * the Free Software Foundation; either version 2 of the License, or
16  * (at your option) any later version.
17  *
18  * This program is distributed in the hope that it will be useful,
19  * but WITHOUT ANY WARRANTY; without even the implied warranty of
20  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21  * GNU General Public License for more details.
22  *
23  * You should have received a copy of the GNU General Public License
24  * along with this program; if not, write to the Free Software
25  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
26  *
27  ******************************************************************************/
28 
29 #include <linux/string.h>
30 #include <linux/parser.h>
31 #include <linux/timer.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/spinlock.h>
35 #include <linux/module.h>
36 #include <scsi/scsi.h>
37 #include <scsi/scsi_host.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 struct se_subsystem_api fileio_template;
45 
46 /*	fd_attach_hba(): (Part of se_subsystem_api_t template)
47  *
48  *
49  */
50 static int fd_attach_hba(struct se_hba *hba, u32 host_id)
51 {
52 	struct fd_host *fd_host;
53 
54 	fd_host = kzalloc(sizeof(struct fd_host), GFP_KERNEL);
55 	if (!fd_host) {
56 		pr_err("Unable to allocate memory for struct fd_host\n");
57 		return -ENOMEM;
58 	}
59 
60 	fd_host->fd_host_id = host_id;
61 
62 	hba->hba_ptr = fd_host;
63 
64 	pr_debug("CORE_HBA[%d] - TCM FILEIO HBA Driver %s on Generic"
65 		" Target Core Stack %s\n", hba->hba_id, FD_VERSION,
66 		TARGET_CORE_MOD_VERSION);
67 	pr_debug("CORE_HBA[%d] - Attached FILEIO HBA: %u to Generic"
68 		" MaxSectors: %u\n",
69 		hba->hba_id, fd_host->fd_host_id, FD_MAX_SECTORS);
70 
71 	return 0;
72 }
73 
74 static void fd_detach_hba(struct se_hba *hba)
75 {
76 	struct fd_host *fd_host = hba->hba_ptr;
77 
78 	pr_debug("CORE_HBA[%d] - Detached FILEIO HBA: %u from Generic"
79 		" Target Core\n", hba->hba_id, fd_host->fd_host_id);
80 
81 	kfree(fd_host);
82 	hba->hba_ptr = NULL;
83 }
84 
85 static void *fd_allocate_virtdevice(struct se_hba *hba, const char *name)
86 {
87 	struct fd_dev *fd_dev;
88 	struct fd_host *fd_host = hba->hba_ptr;
89 
90 	fd_dev = kzalloc(sizeof(struct fd_dev), GFP_KERNEL);
91 	if (!fd_dev) {
92 		pr_err("Unable to allocate memory for struct fd_dev\n");
93 		return NULL;
94 	}
95 
96 	fd_dev->fd_host = fd_host;
97 
98 	pr_debug("FILEIO: Allocated fd_dev for %p\n", name);
99 
100 	return fd_dev;
101 }
102 
103 /*	fd_create_virtdevice(): (Part of se_subsystem_api_t template)
104  *
105  *
106  */
107 static struct se_device *fd_create_virtdevice(
108 	struct se_hba *hba,
109 	struct se_subsystem_dev *se_dev,
110 	void *p)
111 {
112 	struct se_device *dev;
113 	struct se_dev_limits dev_limits;
114 	struct queue_limits *limits;
115 	struct fd_dev *fd_dev = p;
116 	struct fd_host *fd_host = hba->hba_ptr;
117 	struct file *file;
118 	struct inode *inode = NULL;
119 	int dev_flags = 0, flags, ret = -EINVAL;
120 
121 	memset(&dev_limits, 0, sizeof(struct se_dev_limits));
122 
123 	/*
124 	 * Use O_DSYNC by default instead of O_SYNC to forgo syncing
125 	 * of pure timestamp updates.
126 	 */
127 	flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
128 	/*
129 	 * Optionally allow fd_buffered_io=1 to be enabled for people
130 	 * who want use the fs buffer cache as an WriteCache mechanism.
131 	 *
132 	 * This means that in event of a hard failure, there is a risk
133 	 * of silent data-loss if the SCSI client has *not* performed a
134 	 * forced unit access (FUA) write, or issued SYNCHRONIZE_CACHE
135 	 * to write-out the entire device cache.
136 	 */
137 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
138 		pr_debug("FILEIO: Disabling O_DSYNC, using buffered FILEIO\n");
139 		flags &= ~O_DSYNC;
140 	}
141 
142 	file = filp_open(fd_dev->fd_dev_name, flags, 0600);
143 	if (IS_ERR(file)) {
144 		pr_err("filp_open(%s) failed\n", fd_dev->fd_dev_name);
145 		ret = PTR_ERR(file);
146 		goto fail;
147 	}
148 	fd_dev->fd_file = file;
149 	/*
150 	 * If using a block backend with this struct file, we extract
151 	 * fd_dev->fd_[block,dev]_size from struct block_device.
152 	 *
153 	 * Otherwise, we use the passed fd_size= from configfs
154 	 */
155 	inode = file->f_mapping->host;
156 	if (S_ISBLK(inode->i_mode)) {
157 		struct request_queue *q;
158 		unsigned long long dev_size;
159 		/*
160 		 * Setup the local scope queue_limits from struct request_queue->limits
161 		 * to pass into transport_add_device_to_core_hba() as struct se_dev_limits.
162 		 */
163 		q = bdev_get_queue(inode->i_bdev);
164 		limits = &dev_limits.limits;
165 		limits->logical_block_size = bdev_logical_block_size(inode->i_bdev);
166 		limits->max_hw_sectors = queue_max_hw_sectors(q);
167 		limits->max_sectors = queue_max_sectors(q);
168 		/*
169 		 * Determine the number of bytes from i_size_read() minus
170 		 * one (1) logical sector from underlying struct block_device
171 		 */
172 		fd_dev->fd_block_size = bdev_logical_block_size(inode->i_bdev);
173 		dev_size = (i_size_read(file->f_mapping->host) -
174 				       fd_dev->fd_block_size);
175 
176 		pr_debug("FILEIO: Using size: %llu bytes from struct"
177 			" block_device blocks: %llu logical_block_size: %d\n",
178 			dev_size, div_u64(dev_size, fd_dev->fd_block_size),
179 			fd_dev->fd_block_size);
180 	} else {
181 		if (!(fd_dev->fbd_flags & FBDF_HAS_SIZE)) {
182 			pr_err("FILEIO: Missing fd_dev_size="
183 				" parameter, and no backing struct"
184 				" block_device\n");
185 			goto fail;
186 		}
187 
188 		limits = &dev_limits.limits;
189 		limits->logical_block_size = FD_BLOCKSIZE;
190 		limits->max_hw_sectors = FD_MAX_SECTORS;
191 		limits->max_sectors = FD_MAX_SECTORS;
192 		fd_dev->fd_block_size = FD_BLOCKSIZE;
193 	}
194 
195 	dev_limits.hw_queue_depth = FD_MAX_DEVICE_QUEUE_DEPTH;
196 	dev_limits.queue_depth = FD_DEVICE_QUEUE_DEPTH;
197 
198 	dev = transport_add_device_to_core_hba(hba, &fileio_template,
199 				se_dev, dev_flags, fd_dev,
200 				&dev_limits, "FILEIO", FD_VERSION);
201 	if (!dev)
202 		goto fail;
203 
204 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
205 		pr_debug("FILEIO: Forcing setting of emulate_write_cache=1"
206 			" with FDBD_HAS_BUFFERED_IO_WCE\n");
207 		dev->se_sub_dev->se_dev_attrib.emulate_write_cache = 1;
208 	}
209 
210 	fd_dev->fd_dev_id = fd_host->fd_host_dev_id_count++;
211 	fd_dev->fd_queue_depth = dev->queue_depth;
212 
213 	pr_debug("CORE_FILE[%u] - Added TCM FILEIO Device ID: %u at %s,"
214 		" %llu total bytes\n", fd_host->fd_host_id, fd_dev->fd_dev_id,
215 			fd_dev->fd_dev_name, fd_dev->fd_dev_size);
216 
217 	return dev;
218 fail:
219 	if (fd_dev->fd_file) {
220 		filp_close(fd_dev->fd_file, NULL);
221 		fd_dev->fd_file = NULL;
222 	}
223 	return ERR_PTR(ret);
224 }
225 
226 /*	fd_free_device(): (Part of se_subsystem_api_t template)
227  *
228  *
229  */
230 static void fd_free_device(void *p)
231 {
232 	struct fd_dev *fd_dev = p;
233 
234 	if (fd_dev->fd_file) {
235 		filp_close(fd_dev->fd_file, NULL);
236 		fd_dev->fd_file = NULL;
237 	}
238 
239 	kfree(fd_dev);
240 }
241 
242 static int fd_do_readv(struct se_cmd *cmd, struct scatterlist *sgl,
243 		u32 sgl_nents)
244 {
245 	struct se_device *se_dev = cmd->se_dev;
246 	struct fd_dev *dev = se_dev->dev_ptr;
247 	struct file *fd = dev->fd_file;
248 	struct scatterlist *sg;
249 	struct iovec *iov;
250 	mm_segment_t old_fs;
251 	loff_t pos = (cmd->t_task_lba *
252 		      se_dev->se_sub_dev->se_dev_attrib.block_size);
253 	int ret = 0, i;
254 
255 	iov = kzalloc(sizeof(struct iovec) * sgl_nents, GFP_KERNEL);
256 	if (!iov) {
257 		pr_err("Unable to allocate fd_do_readv iov[]\n");
258 		return -ENOMEM;
259 	}
260 
261 	for_each_sg(sgl, sg, sgl_nents, i) {
262 		iov[i].iov_len = sg->length;
263 		iov[i].iov_base = sg_virt(sg);
264 	}
265 
266 	old_fs = get_fs();
267 	set_fs(get_ds());
268 	ret = vfs_readv(fd, &iov[0], sgl_nents, &pos);
269 	set_fs(old_fs);
270 
271 	kfree(iov);
272 	/*
273 	 * Return zeros and GOOD status even if the READ did not return
274 	 * the expected virt_size for struct file w/o a backing struct
275 	 * block_device.
276 	 */
277 	if (S_ISBLK(fd->f_dentry->d_inode->i_mode)) {
278 		if (ret < 0 || ret != cmd->data_length) {
279 			pr_err("vfs_readv() returned %d,"
280 				" expecting %d for S_ISBLK\n", ret,
281 				(int)cmd->data_length);
282 			return (ret < 0 ? ret : -EINVAL);
283 		}
284 	} else {
285 		if (ret < 0) {
286 			pr_err("vfs_readv() returned %d for non"
287 				" S_ISBLK\n", ret);
288 			return ret;
289 		}
290 	}
291 
292 	return 1;
293 }
294 
295 static int fd_do_writev(struct se_cmd *cmd, struct scatterlist *sgl,
296 		u32 sgl_nents)
297 {
298 	struct se_device *se_dev = cmd->se_dev;
299 	struct fd_dev *dev = se_dev->dev_ptr;
300 	struct file *fd = dev->fd_file;
301 	struct scatterlist *sg;
302 	struct iovec *iov;
303 	mm_segment_t old_fs;
304 	loff_t pos = (cmd->t_task_lba *
305 		      se_dev->se_sub_dev->se_dev_attrib.block_size);
306 	int ret, i = 0;
307 
308 	iov = kzalloc(sizeof(struct iovec) * sgl_nents, GFP_KERNEL);
309 	if (!iov) {
310 		pr_err("Unable to allocate fd_do_writev iov[]\n");
311 		return -ENOMEM;
312 	}
313 
314 	for_each_sg(sgl, sg, sgl_nents, i) {
315 		iov[i].iov_len = sg->length;
316 		iov[i].iov_base = sg_virt(sg);
317 	}
318 
319 	old_fs = get_fs();
320 	set_fs(get_ds());
321 	ret = vfs_writev(fd, &iov[0], sgl_nents, &pos);
322 	set_fs(old_fs);
323 
324 	kfree(iov);
325 
326 	if (ret < 0 || ret != cmd->data_length) {
327 		pr_err("vfs_writev() returned %d\n", ret);
328 		return (ret < 0 ? ret : -EINVAL);
329 	}
330 
331 	return 1;
332 }
333 
334 static int fd_execute_sync_cache(struct se_cmd *cmd)
335 {
336 	struct se_device *dev = cmd->se_dev;
337 	struct fd_dev *fd_dev = dev->dev_ptr;
338 	int immed = (cmd->t_task_cdb[1] & 0x2);
339 	loff_t start, end;
340 	int ret;
341 
342 	/*
343 	 * If the Immediate bit is set, queue up the GOOD response
344 	 * for this SYNCHRONIZE_CACHE op
345 	 */
346 	if (immed)
347 		target_complete_cmd(cmd, SAM_STAT_GOOD);
348 
349 	/*
350 	 * Determine if we will be flushing the entire device.
351 	 */
352 	if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
353 		start = 0;
354 		end = LLONG_MAX;
355 	} else {
356 		start = cmd->t_task_lba * dev->se_sub_dev->se_dev_attrib.block_size;
357 		if (cmd->data_length)
358 			end = start + cmd->data_length;
359 		else
360 			end = LLONG_MAX;
361 	}
362 
363 	ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
364 	if (ret != 0)
365 		pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
366 
367 	if (immed)
368 		return 0;
369 
370 	if (ret) {
371 		cmd->scsi_sense_reason = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
372 		target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
373 	} else {
374 		target_complete_cmd(cmd, SAM_STAT_GOOD);
375 	}
376 
377 	return 0;
378 }
379 
380 static int fd_execute_rw(struct se_cmd *cmd)
381 {
382 	struct scatterlist *sgl = cmd->t_data_sg;
383 	u32 sgl_nents = cmd->t_data_nents;
384 	enum dma_data_direction data_direction = cmd->data_direction;
385 	struct se_device *dev = cmd->se_dev;
386 	int ret = 0;
387 
388 	/*
389 	 * Call vectorized fileio functions to map struct scatterlist
390 	 * physical memory addresses to struct iovec virtual memory.
391 	 */
392 	if (data_direction == DMA_FROM_DEVICE) {
393 		ret = fd_do_readv(cmd, sgl, sgl_nents);
394 	} else {
395 		ret = fd_do_writev(cmd, sgl, sgl_nents);
396 		/*
397 		 * Perform implict vfs_fsync_range() for fd_do_writev() ops
398 		 * for SCSI WRITEs with Forced Unit Access (FUA) set.
399 		 * Allow this to happen independent of WCE=0 setting.
400 		 */
401 		if (ret > 0 &&
402 		    dev->se_sub_dev->se_dev_attrib.emulate_fua_write > 0 &&
403 		    (cmd->se_cmd_flags & SCF_FUA)) {
404 			struct fd_dev *fd_dev = dev->dev_ptr;
405 			loff_t start = cmd->t_task_lba *
406 				dev->se_sub_dev->se_dev_attrib.block_size;
407 			loff_t end = start + cmd->data_length;
408 
409 			vfs_fsync_range(fd_dev->fd_file, start, end, 1);
410 		}
411 	}
412 
413 	if (ret < 0) {
414 		cmd->scsi_sense_reason = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
415 		return ret;
416 	}
417 	if (ret)
418 		target_complete_cmd(cmd, SAM_STAT_GOOD);
419 	return 0;
420 }
421 
422 enum {
423 	Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io, Opt_err
424 };
425 
426 static match_table_t tokens = {
427 	{Opt_fd_dev_name, "fd_dev_name=%s"},
428 	{Opt_fd_dev_size, "fd_dev_size=%s"},
429 	{Opt_fd_buffered_io, "fd_buffered_io=%d"},
430 	{Opt_err, NULL}
431 };
432 
433 static ssize_t fd_set_configfs_dev_params(
434 	struct se_hba *hba,
435 	struct se_subsystem_dev *se_dev,
436 	const char *page, ssize_t count)
437 {
438 	struct fd_dev *fd_dev = se_dev->se_dev_su_ptr;
439 	char *orig, *ptr, *arg_p, *opts;
440 	substring_t args[MAX_OPT_ARGS];
441 	int ret = 0, arg, token;
442 
443 	opts = kstrdup(page, GFP_KERNEL);
444 	if (!opts)
445 		return -ENOMEM;
446 
447 	orig = opts;
448 
449 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
450 		if (!*ptr)
451 			continue;
452 
453 		token = match_token(ptr, tokens, args);
454 		switch (token) {
455 		case Opt_fd_dev_name:
456 			if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
457 				FD_MAX_DEV_NAME) == 0) {
458 				ret = -EINVAL;
459 				break;
460 			}
461 			pr_debug("FILEIO: Referencing Path: %s\n",
462 					fd_dev->fd_dev_name);
463 			fd_dev->fbd_flags |= FBDF_HAS_PATH;
464 			break;
465 		case Opt_fd_dev_size:
466 			arg_p = match_strdup(&args[0]);
467 			if (!arg_p) {
468 				ret = -ENOMEM;
469 				break;
470 			}
471 			ret = strict_strtoull(arg_p, 0, &fd_dev->fd_dev_size);
472 			kfree(arg_p);
473 			if (ret < 0) {
474 				pr_err("strict_strtoull() failed for"
475 						" fd_dev_size=\n");
476 				goto out;
477 			}
478 			pr_debug("FILEIO: Referencing Size: %llu"
479 					" bytes\n", fd_dev->fd_dev_size);
480 			fd_dev->fbd_flags |= FBDF_HAS_SIZE;
481 			break;
482 		case Opt_fd_buffered_io:
483 			match_int(args, &arg);
484 			if (arg != 1) {
485 				pr_err("bogus fd_buffered_io=%d value\n", arg);
486 				ret = -EINVAL;
487 				goto out;
488 			}
489 
490 			pr_debug("FILEIO: Using buffered I/O"
491 				" operations for struct fd_dev\n");
492 
493 			fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
494 			break;
495 		default:
496 			break;
497 		}
498 	}
499 
500 out:
501 	kfree(orig);
502 	return (!ret) ? count : ret;
503 }
504 
505 static ssize_t fd_check_configfs_dev_params(struct se_hba *hba, struct se_subsystem_dev *se_dev)
506 {
507 	struct fd_dev *fd_dev = se_dev->se_dev_su_ptr;
508 
509 	if (!(fd_dev->fbd_flags & FBDF_HAS_PATH)) {
510 		pr_err("Missing fd_dev_name=\n");
511 		return -EINVAL;
512 	}
513 
514 	return 0;
515 }
516 
517 static ssize_t fd_show_configfs_dev_params(
518 	struct se_hba *hba,
519 	struct se_subsystem_dev *se_dev,
520 	char *b)
521 {
522 	struct fd_dev *fd_dev = se_dev->se_dev_su_ptr;
523 	ssize_t bl = 0;
524 
525 	bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
526 	bl += sprintf(b + bl, "        File: %s  Size: %llu  Mode: %s\n",
527 		fd_dev->fd_dev_name, fd_dev->fd_dev_size,
528 		(fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
529 		"Buffered-WCE" : "O_DSYNC");
530 	return bl;
531 }
532 
533 /*	fd_get_device_rev(): (Part of se_subsystem_api_t template)
534  *
535  *
536  */
537 static u32 fd_get_device_rev(struct se_device *dev)
538 {
539 	return SCSI_SPC_2; /* Returns SPC-3 in Initiator Data */
540 }
541 
542 /*	fd_get_device_type(): (Part of se_subsystem_api_t template)
543  *
544  *
545  */
546 static u32 fd_get_device_type(struct se_device *dev)
547 {
548 	return TYPE_DISK;
549 }
550 
551 static sector_t fd_get_blocks(struct se_device *dev)
552 {
553 	struct fd_dev *fd_dev = dev->dev_ptr;
554 	struct file *f = fd_dev->fd_file;
555 	struct inode *i = f->f_mapping->host;
556 	unsigned long long dev_size;
557 	/*
558 	 * When using a file that references an underlying struct block_device,
559 	 * ensure dev_size is always based on the current inode size in order
560 	 * to handle underlying block_device resize operations.
561 	 */
562 	if (S_ISBLK(i->i_mode))
563 		dev_size = (i_size_read(i) - fd_dev->fd_block_size);
564 	else
565 		dev_size = fd_dev->fd_dev_size;
566 
567 	return div_u64(dev_size, dev->se_sub_dev->se_dev_attrib.block_size);
568 }
569 
570 static struct spc_ops fd_spc_ops = {
571 	.execute_rw		= fd_execute_rw,
572 	.execute_sync_cache	= fd_execute_sync_cache,
573 };
574 
575 static int fd_parse_cdb(struct se_cmd *cmd)
576 {
577 	return sbc_parse_cdb(cmd, &fd_spc_ops);
578 }
579 
580 static struct se_subsystem_api fileio_template = {
581 	.name			= "fileio",
582 	.owner			= THIS_MODULE,
583 	.transport_type		= TRANSPORT_PLUGIN_VHBA_PDEV,
584 	.attach_hba		= fd_attach_hba,
585 	.detach_hba		= fd_detach_hba,
586 	.allocate_virtdevice	= fd_allocate_virtdevice,
587 	.create_virtdevice	= fd_create_virtdevice,
588 	.free_device		= fd_free_device,
589 	.parse_cdb		= fd_parse_cdb,
590 	.check_configfs_dev_params = fd_check_configfs_dev_params,
591 	.set_configfs_dev_params = fd_set_configfs_dev_params,
592 	.show_configfs_dev_params = fd_show_configfs_dev_params,
593 	.get_device_rev		= fd_get_device_rev,
594 	.get_device_type	= fd_get_device_type,
595 	.get_blocks		= fd_get_blocks,
596 };
597 
598 static int __init fileio_module_init(void)
599 {
600 	return transport_subsystem_register(&fileio_template);
601 }
602 
603 static void fileio_module_exit(void)
604 {
605 	transport_subsystem_release(&fileio_template);
606 }
607 
608 MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
609 MODULE_AUTHOR("nab@Linux-iSCSI.org");
610 MODULE_LICENSE("GPL");
611 
612 module_init(fileio_module_init);
613 module_exit(fileio_module_exit);
614