1 /*******************************************************************************
2  * Filename:  target_core_file.c
3  *
4  * This file contains the Storage Engine <-> FILEIO transport specific functions
5  *
6  * (c) Copyright 2005-2012 RisingTide Systems LLC.
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 <scsi/scsi.h>
34 #include <scsi/scsi_host.h>
35 
36 #include <target/target_core_base.h>
37 #include <target/target_core_backend.h>
38 
39 #include "target_core_file.h"
40 
41 static inline struct fd_dev *FD_DEV(struct se_device *dev)
42 {
43 	return container_of(dev, struct fd_dev, dev);
44 }
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 struct se_device *fd_alloc_device(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->dev;
101 }
102 
103 static int fd_configure_device(struct se_device *dev)
104 {
105 	struct fd_dev *fd_dev = FD_DEV(dev);
106 	struct fd_host *fd_host = dev->se_hba->hba_ptr;
107 	struct file *file;
108 	struct inode *inode = NULL;
109 	int flags, ret = -EINVAL;
110 
111 	if (!(fd_dev->fbd_flags & FBDF_HAS_PATH)) {
112 		pr_err("Missing fd_dev_name=\n");
113 		return -EINVAL;
114 	}
115 
116 	/*
117 	 * Use O_DSYNC by default instead of O_SYNC to forgo syncing
118 	 * of pure timestamp updates.
119 	 */
120 	flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
121 
122 	/*
123 	 * Optionally allow fd_buffered_io=1 to be enabled for people
124 	 * who want use the fs buffer cache as an WriteCache mechanism.
125 	 *
126 	 * This means that in event of a hard failure, there is a risk
127 	 * of silent data-loss if the SCSI client has *not* performed a
128 	 * forced unit access (FUA) write, or issued SYNCHRONIZE_CACHE
129 	 * to write-out the entire device cache.
130 	 */
131 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
132 		pr_debug("FILEIO: Disabling O_DSYNC, using buffered FILEIO\n");
133 		flags &= ~O_DSYNC;
134 	}
135 
136 	file = filp_open(fd_dev->fd_dev_name, flags, 0600);
137 	if (IS_ERR(file)) {
138 		pr_err("filp_open(%s) failed\n", fd_dev->fd_dev_name);
139 		ret = PTR_ERR(file);
140 		goto fail;
141 	}
142 	fd_dev->fd_file = file;
143 	/*
144 	 * If using a block backend with this struct file, we extract
145 	 * fd_dev->fd_[block,dev]_size from struct block_device.
146 	 *
147 	 * Otherwise, we use the passed fd_size= from configfs
148 	 */
149 	inode = file->f_mapping->host;
150 	if (S_ISBLK(inode->i_mode)) {
151 		struct request_queue *q = bdev_get_queue(inode->i_bdev);
152 		unsigned long long dev_size;
153 
154 		dev->dev_attrib.hw_block_size =
155 			bdev_logical_block_size(inode->i_bdev);
156 		dev->dev_attrib.hw_max_sectors = queue_max_hw_sectors(q);
157 
158 		/*
159 		 * Determine the number of bytes from i_size_read() minus
160 		 * one (1) logical sector from underlying struct block_device
161 		 */
162 		dev_size = (i_size_read(file->f_mapping->host) -
163 				       fd_dev->fd_block_size);
164 
165 		pr_debug("FILEIO: Using size: %llu bytes from struct"
166 			" block_device blocks: %llu logical_block_size: %d\n",
167 			dev_size, div_u64(dev_size, fd_dev->fd_block_size),
168 			fd_dev->fd_block_size);
169 	} else {
170 		if (!(fd_dev->fbd_flags & FBDF_HAS_SIZE)) {
171 			pr_err("FILEIO: Missing fd_dev_size="
172 				" parameter, and no backing struct"
173 				" block_device\n");
174 			goto fail;
175 		}
176 
177 		dev->dev_attrib.hw_block_size = FD_BLOCKSIZE;
178 		dev->dev_attrib.hw_max_sectors = FD_MAX_SECTORS;
179 	}
180 
181 	fd_dev->fd_block_size = dev->dev_attrib.hw_block_size;
182 
183 	dev->dev_attrib.hw_queue_depth = FD_MAX_DEVICE_QUEUE_DEPTH;
184 
185 	if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
186 		pr_debug("FILEIO: Forcing setting of emulate_write_cache=1"
187 			" with FDBD_HAS_BUFFERED_IO_WCE\n");
188 		dev->dev_attrib.emulate_write_cache = 1;
189 	}
190 
191 	fd_dev->fd_dev_id = fd_host->fd_host_dev_id_count++;
192 	fd_dev->fd_queue_depth = dev->queue_depth;
193 
194 	pr_debug("CORE_FILE[%u] - Added TCM FILEIO Device ID: %u at %s,"
195 		" %llu total bytes\n", fd_host->fd_host_id, fd_dev->fd_dev_id,
196 			fd_dev->fd_dev_name, fd_dev->fd_dev_size);
197 
198 	return 0;
199 fail:
200 	if (fd_dev->fd_file) {
201 		filp_close(fd_dev->fd_file, NULL);
202 		fd_dev->fd_file = NULL;
203 	}
204 	return ret;
205 }
206 
207 static void fd_free_device(struct se_device *dev)
208 {
209 	struct fd_dev *fd_dev = FD_DEV(dev);
210 
211 	if (fd_dev->fd_file) {
212 		filp_close(fd_dev->fd_file, NULL);
213 		fd_dev->fd_file = NULL;
214 	}
215 
216 	kfree(fd_dev);
217 }
218 
219 static int fd_do_rw(struct se_cmd *cmd, struct scatterlist *sgl,
220 		u32 sgl_nents, int is_write)
221 {
222 	struct se_device *se_dev = cmd->se_dev;
223 	struct fd_dev *dev = FD_DEV(se_dev);
224 	struct file *fd = dev->fd_file;
225 	struct scatterlist *sg;
226 	struct iovec *iov;
227 	mm_segment_t old_fs;
228 	loff_t pos = (cmd->t_task_lba * se_dev->dev_attrib.block_size);
229 	int ret = 0, i;
230 
231 	iov = kzalloc(sizeof(struct iovec) * sgl_nents, GFP_KERNEL);
232 	if (!iov) {
233 		pr_err("Unable to allocate fd_do_readv iov[]\n");
234 		return -ENOMEM;
235 	}
236 
237 	for_each_sg(sgl, sg, sgl_nents, i) {
238 		iov[i].iov_len = sg->length;
239 		iov[i].iov_base = kmap(sg_page(sg)) + sg->offset;
240 	}
241 
242 	old_fs = get_fs();
243 	set_fs(get_ds());
244 
245 	if (is_write)
246 		ret = vfs_writev(fd, &iov[0], sgl_nents, &pos);
247 	else
248 		ret = vfs_readv(fd, &iov[0], sgl_nents, &pos);
249 
250 	set_fs(old_fs);
251 
252 	for_each_sg(sgl, sg, sgl_nents, i)
253 		kunmap(sg_page(sg));
254 
255 	kfree(iov);
256 
257 	if (is_write) {
258 		if (ret < 0 || ret != cmd->data_length) {
259 			pr_err("%s() write returned %d\n", __func__, ret);
260 			return (ret < 0 ? ret : -EINVAL);
261 		}
262 	} else {
263 		/*
264 		 * Return zeros and GOOD status even if the READ did not return
265 		 * the expected virt_size for struct file w/o a backing struct
266 		 * block_device.
267 		 */
268 		if (S_ISBLK(fd->f_dentry->d_inode->i_mode)) {
269 			if (ret < 0 || ret != cmd->data_length) {
270 				pr_err("%s() returned %d, expecting %u for "
271 						"S_ISBLK\n", __func__, ret,
272 						cmd->data_length);
273 				return (ret < 0 ? ret : -EINVAL);
274 			}
275 		} else {
276 			if (ret < 0) {
277 				pr_err("%s() returned %d for non S_ISBLK\n",
278 						__func__, ret);
279 				return ret;
280 			}
281 		}
282 	}
283 	return 1;
284 }
285 
286 static sense_reason_t
287 fd_execute_sync_cache(struct se_cmd *cmd)
288 {
289 	struct se_device *dev = cmd->se_dev;
290 	struct fd_dev *fd_dev = FD_DEV(dev);
291 	int immed = (cmd->t_task_cdb[1] & 0x2);
292 	loff_t start, end;
293 	int ret;
294 
295 	/*
296 	 * If the Immediate bit is set, queue up the GOOD response
297 	 * for this SYNCHRONIZE_CACHE op
298 	 */
299 	if (immed)
300 		target_complete_cmd(cmd, SAM_STAT_GOOD);
301 
302 	/*
303 	 * Determine if we will be flushing the entire device.
304 	 */
305 	if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
306 		start = 0;
307 		end = LLONG_MAX;
308 	} else {
309 		start = cmd->t_task_lba * dev->dev_attrib.block_size;
310 		if (cmd->data_length)
311 			end = start + cmd->data_length;
312 		else
313 			end = LLONG_MAX;
314 	}
315 
316 	ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
317 	if (ret != 0)
318 		pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
319 
320 	if (immed)
321 		return 0;
322 
323 	if (ret)
324 		target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
325 	else
326 		target_complete_cmd(cmd, SAM_STAT_GOOD);
327 
328 	return 0;
329 }
330 
331 static sense_reason_t
332 fd_execute_rw(struct se_cmd *cmd)
333 {
334 	struct scatterlist *sgl = cmd->t_data_sg;
335 	u32 sgl_nents = cmd->t_data_nents;
336 	enum dma_data_direction data_direction = cmd->data_direction;
337 	struct se_device *dev = cmd->se_dev;
338 	int ret = 0;
339 
340 	/*
341 	 * Call vectorized fileio functions to map struct scatterlist
342 	 * physical memory addresses to struct iovec virtual memory.
343 	 */
344 	if (data_direction == DMA_FROM_DEVICE) {
345 		ret = fd_do_rw(cmd, sgl, sgl_nents, 0);
346 	} else {
347 		ret = fd_do_rw(cmd, sgl, sgl_nents, 1);
348 		/*
349 		 * Perform implict vfs_fsync_range() for fd_do_writev() ops
350 		 * for SCSI WRITEs with Forced Unit Access (FUA) set.
351 		 * Allow this to happen independent of WCE=0 setting.
352 		 */
353 		if (ret > 0 &&
354 		    dev->dev_attrib.emulate_fua_write > 0 &&
355 		    (cmd->se_cmd_flags & SCF_FUA)) {
356 			struct fd_dev *fd_dev = FD_DEV(dev);
357 			loff_t start = cmd->t_task_lba *
358 				dev->dev_attrib.block_size;
359 			loff_t end = start + cmd->data_length;
360 
361 			vfs_fsync_range(fd_dev->fd_file, start, end, 1);
362 		}
363 	}
364 
365 	if (ret < 0)
366 		return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
367 
368 	if (ret)
369 		target_complete_cmd(cmd, SAM_STAT_GOOD);
370 	return 0;
371 }
372 
373 enum {
374 	Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io, Opt_err
375 };
376 
377 static match_table_t tokens = {
378 	{Opt_fd_dev_name, "fd_dev_name=%s"},
379 	{Opt_fd_dev_size, "fd_dev_size=%s"},
380 	{Opt_fd_buffered_io, "fd_buffered_io=%d"},
381 	{Opt_err, NULL}
382 };
383 
384 static ssize_t fd_set_configfs_dev_params(struct se_device *dev,
385 		const char *page, ssize_t count)
386 {
387 	struct fd_dev *fd_dev = FD_DEV(dev);
388 	char *orig, *ptr, *arg_p, *opts;
389 	substring_t args[MAX_OPT_ARGS];
390 	int ret = 0, arg, token;
391 
392 	opts = kstrdup(page, GFP_KERNEL);
393 	if (!opts)
394 		return -ENOMEM;
395 
396 	orig = opts;
397 
398 	while ((ptr = strsep(&opts, ",\n")) != NULL) {
399 		if (!*ptr)
400 			continue;
401 
402 		token = match_token(ptr, tokens, args);
403 		switch (token) {
404 		case Opt_fd_dev_name:
405 			if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
406 				FD_MAX_DEV_NAME) == 0) {
407 				ret = -EINVAL;
408 				break;
409 			}
410 			pr_debug("FILEIO: Referencing Path: %s\n",
411 					fd_dev->fd_dev_name);
412 			fd_dev->fbd_flags |= FBDF_HAS_PATH;
413 			break;
414 		case Opt_fd_dev_size:
415 			arg_p = match_strdup(&args[0]);
416 			if (!arg_p) {
417 				ret = -ENOMEM;
418 				break;
419 			}
420 			ret = strict_strtoull(arg_p, 0, &fd_dev->fd_dev_size);
421 			kfree(arg_p);
422 			if (ret < 0) {
423 				pr_err("strict_strtoull() failed for"
424 						" fd_dev_size=\n");
425 				goto out;
426 			}
427 			pr_debug("FILEIO: Referencing Size: %llu"
428 					" bytes\n", fd_dev->fd_dev_size);
429 			fd_dev->fbd_flags |= FBDF_HAS_SIZE;
430 			break;
431 		case Opt_fd_buffered_io:
432 			match_int(args, &arg);
433 			if (arg != 1) {
434 				pr_err("bogus fd_buffered_io=%d value\n", arg);
435 				ret = -EINVAL;
436 				goto out;
437 			}
438 
439 			pr_debug("FILEIO: Using buffered I/O"
440 				" operations for struct fd_dev\n");
441 
442 			fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
443 			break;
444 		default:
445 			break;
446 		}
447 	}
448 
449 out:
450 	kfree(orig);
451 	return (!ret) ? count : ret;
452 }
453 
454 static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b)
455 {
456 	struct fd_dev *fd_dev = FD_DEV(dev);
457 	ssize_t bl = 0;
458 
459 	bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
460 	bl += sprintf(b + bl, "        File: %s  Size: %llu  Mode: %s\n",
461 		fd_dev->fd_dev_name, fd_dev->fd_dev_size,
462 		(fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
463 		"Buffered-WCE" : "O_DSYNC");
464 	return bl;
465 }
466 
467 static sector_t fd_get_blocks(struct se_device *dev)
468 {
469 	struct fd_dev *fd_dev = FD_DEV(dev);
470 	struct file *f = fd_dev->fd_file;
471 	struct inode *i = f->f_mapping->host;
472 	unsigned long long dev_size;
473 	/*
474 	 * When using a file that references an underlying struct block_device,
475 	 * ensure dev_size is always based on the current inode size in order
476 	 * to handle underlying block_device resize operations.
477 	 */
478 	if (S_ISBLK(i->i_mode))
479 		dev_size = (i_size_read(i) - fd_dev->fd_block_size);
480 	else
481 		dev_size = fd_dev->fd_dev_size;
482 
483 	return div_u64(dev_size, dev->dev_attrib.block_size);
484 }
485 
486 static struct sbc_ops fd_sbc_ops = {
487 	.execute_rw		= fd_execute_rw,
488 	.execute_sync_cache	= fd_execute_sync_cache,
489 };
490 
491 static sense_reason_t
492 fd_parse_cdb(struct se_cmd *cmd)
493 {
494 	return sbc_parse_cdb(cmd, &fd_sbc_ops);
495 }
496 
497 static struct se_subsystem_api fileio_template = {
498 	.name			= "fileio",
499 	.inquiry_prod		= "FILEIO",
500 	.inquiry_rev		= FD_VERSION,
501 	.owner			= THIS_MODULE,
502 	.transport_type		= TRANSPORT_PLUGIN_VHBA_PDEV,
503 	.attach_hba		= fd_attach_hba,
504 	.detach_hba		= fd_detach_hba,
505 	.alloc_device		= fd_alloc_device,
506 	.configure_device	= fd_configure_device,
507 	.free_device		= fd_free_device,
508 	.parse_cdb		= fd_parse_cdb,
509 	.set_configfs_dev_params = fd_set_configfs_dev_params,
510 	.show_configfs_dev_params = fd_show_configfs_dev_params,
511 	.get_device_type	= sbc_get_device_type,
512 	.get_blocks		= fd_get_blocks,
513 };
514 
515 static int __init fileio_module_init(void)
516 {
517 	return transport_subsystem_register(&fileio_template);
518 }
519 
520 static void fileio_module_exit(void)
521 {
522 	transport_subsystem_release(&fileio_template);
523 }
524 
525 MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
526 MODULE_AUTHOR("nab@Linux-iSCSI.org");
527 MODULE_LICENSE("GPL");
528 
529 module_init(fileio_module_init);
530 module_exit(fileio_module_exit);
531