Loading...
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
32static inline struct fd_dev *FD_DEV(struct se_device *dev)
33{
34 return container_of(dev, struct fd_dev, dev);
35}
36
37static 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
60static 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
71static 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
89static 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
107static 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;
213fail:
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
221static 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
229static void fd_free_device(struct se_device *dev)
230{
231 call_rcu(&dev->rcu_head, fd_dev_call_rcu);
232}
233
234static 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
244struct target_core_file_cmd {
245 unsigned long len;
246 struct se_cmd *cmd;
247 struct kiocb iocb;
248 struct bio_vec bvecs[];
249};
250
251static 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
265static sense_reason_t
266fd_execute_rw_aio(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
267 enum dma_data_direction data_direction)
268{
269 int is_write = !(data_direction == DMA_FROM_DEVICE);
270 struct se_device *dev = cmd->se_dev;
271 struct fd_dev *fd_dev = FD_DEV(dev);
272 struct file *file = fd_dev->fd_file;
273 struct target_core_file_cmd *aio_cmd;
274 struct iov_iter iter;
275 struct scatterlist *sg;
276 ssize_t len = 0;
277 int ret = 0, i;
278
279 aio_cmd = kmalloc(struct_size(aio_cmd, bvecs, sgl_nents), GFP_KERNEL);
280 if (!aio_cmd)
281 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
282
283 for_each_sg(sgl, sg, sgl_nents, i) {
284 aio_cmd->bvecs[i].bv_page = sg_page(sg);
285 aio_cmd->bvecs[i].bv_len = sg->length;
286 aio_cmd->bvecs[i].bv_offset = sg->offset;
287
288 len += sg->length;
289 }
290
291 iov_iter_bvec(&iter, is_write, aio_cmd->bvecs, sgl_nents, len);
292
293 aio_cmd->cmd = cmd;
294 aio_cmd->len = len;
295 aio_cmd->iocb.ki_pos = cmd->t_task_lba * dev->dev_attrib.block_size;
296 aio_cmd->iocb.ki_filp = file;
297 aio_cmd->iocb.ki_complete = cmd_rw_aio_complete;
298 aio_cmd->iocb.ki_flags = IOCB_DIRECT;
299
300 if (is_write && (cmd->se_cmd_flags & SCF_FUA))
301 aio_cmd->iocb.ki_flags |= IOCB_DSYNC;
302
303 if (is_write)
304 ret = call_write_iter(file, &aio_cmd->iocb, &iter);
305 else
306 ret = call_read_iter(file, &aio_cmd->iocb, &iter);
307
308 if (ret != -EIOCBQUEUED)
309 cmd_rw_aio_complete(&aio_cmd->iocb, ret);
310
311 return 0;
312}
313
314static int fd_do_rw(struct se_cmd *cmd, struct file *fd,
315 u32 block_size, struct scatterlist *sgl,
316 u32 sgl_nents, u32 data_length, int is_write)
317{
318 struct scatterlist *sg;
319 struct iov_iter iter;
320 struct bio_vec *bvec;
321 ssize_t len = 0;
322 loff_t pos = (cmd->t_task_lba * block_size);
323 int ret = 0, i;
324
325 bvec = kcalloc(sgl_nents, sizeof(struct bio_vec), GFP_KERNEL);
326 if (!bvec) {
327 pr_err("Unable to allocate fd_do_readv iov[]\n");
328 return -ENOMEM;
329 }
330
331 for_each_sg(sgl, sg, sgl_nents, i) {
332 bvec[i].bv_page = sg_page(sg);
333 bvec[i].bv_len = sg->length;
334 bvec[i].bv_offset = sg->offset;
335
336 len += sg->length;
337 }
338
339 iov_iter_bvec(&iter, is_write, bvec, sgl_nents, len);
340 if (is_write)
341 ret = vfs_iter_write(fd, &iter, &pos, 0);
342 else
343 ret = vfs_iter_read(fd, &iter, &pos, 0);
344
345 if (is_write) {
346 if (ret < 0 || ret != data_length) {
347 pr_err("%s() write returned %d\n", __func__, ret);
348 if (ret >= 0)
349 ret = -EINVAL;
350 }
351 } else {
352 /*
353 * Return zeros and GOOD status even if the READ did not return
354 * the expected virt_size for struct file w/o a backing struct
355 * block_device.
356 */
357 if (S_ISBLK(file_inode(fd)->i_mode)) {
358 if (ret < 0 || ret != data_length) {
359 pr_err("%s() returned %d, expecting %u for "
360 "S_ISBLK\n", __func__, ret,
361 data_length);
362 if (ret >= 0)
363 ret = -EINVAL;
364 }
365 } else {
366 if (ret < 0) {
367 pr_err("%s() returned %d for non S_ISBLK\n",
368 __func__, ret);
369 } else if (ret != data_length) {
370 /*
371 * Short read case:
372 * Probably some one truncate file under us.
373 * We must explicitly zero sg-pages to prevent
374 * expose uninizialized pages to userspace.
375 */
376 if (ret < data_length)
377 ret += iov_iter_zero(data_length - ret, &iter);
378 else
379 ret = -EINVAL;
380 }
381 }
382 }
383 kfree(bvec);
384 return ret;
385}
386
387static sense_reason_t
388fd_execute_sync_cache(struct se_cmd *cmd)
389{
390 struct se_device *dev = cmd->se_dev;
391 struct fd_dev *fd_dev = FD_DEV(dev);
392 int immed = (cmd->t_task_cdb[1] & 0x2);
393 loff_t start, end;
394 int ret;
395
396 /*
397 * If the Immediate bit is set, queue up the GOOD response
398 * for this SYNCHRONIZE_CACHE op
399 */
400 if (immed)
401 target_complete_cmd(cmd, SAM_STAT_GOOD);
402
403 /*
404 * Determine if we will be flushing the entire device.
405 */
406 if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
407 start = 0;
408 end = LLONG_MAX;
409 } else {
410 start = cmd->t_task_lba * dev->dev_attrib.block_size;
411 if (cmd->data_length)
412 end = start + cmd->data_length - 1;
413 else
414 end = LLONG_MAX;
415 }
416
417 ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
418 if (ret != 0)
419 pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
420
421 if (immed)
422 return 0;
423
424 if (ret)
425 target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
426 else
427 target_complete_cmd(cmd, SAM_STAT_GOOD);
428
429 return 0;
430}
431
432static sense_reason_t
433fd_execute_write_same(struct se_cmd *cmd)
434{
435 struct se_device *se_dev = cmd->se_dev;
436 struct fd_dev *fd_dev = FD_DEV(se_dev);
437 loff_t pos = cmd->t_task_lba * se_dev->dev_attrib.block_size;
438 sector_t nolb = sbc_get_write_same_sectors(cmd);
439 struct iov_iter iter;
440 struct bio_vec *bvec;
441 unsigned int len = 0, i;
442 ssize_t ret;
443
444 if (cmd->prot_op) {
445 pr_err("WRITE_SAME: Protection information with FILEIO"
446 " backends not supported\n");
447 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
448 }
449
450 if (!cmd->t_data_nents)
451 return TCM_INVALID_CDB_FIELD;
452
453 if (cmd->t_data_nents > 1 ||
454 cmd->t_data_sg[0].length != cmd->se_dev->dev_attrib.block_size) {
455 pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
456 " block_size: %u\n",
457 cmd->t_data_nents,
458 cmd->t_data_sg[0].length,
459 cmd->se_dev->dev_attrib.block_size);
460 return TCM_INVALID_CDB_FIELD;
461 }
462
463 bvec = kcalloc(nolb, sizeof(struct bio_vec), GFP_KERNEL);
464 if (!bvec)
465 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
466
467 for (i = 0; i < nolb; i++) {
468 bvec[i].bv_page = sg_page(&cmd->t_data_sg[0]);
469 bvec[i].bv_len = cmd->t_data_sg[0].length;
470 bvec[i].bv_offset = cmd->t_data_sg[0].offset;
471
472 len += se_dev->dev_attrib.block_size;
473 }
474
475 iov_iter_bvec(&iter, ITER_SOURCE, bvec, nolb, len);
476 ret = vfs_iter_write(fd_dev->fd_file, &iter, &pos, 0);
477
478 kfree(bvec);
479 if (ret < 0 || ret != len) {
480 pr_err("vfs_iter_write() returned %zd for write same\n", ret);
481 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
482 }
483
484 target_complete_cmd(cmd, SAM_STAT_GOOD);
485 return 0;
486}
487
488static int
489fd_do_prot_fill(struct se_device *se_dev, sector_t lba, sector_t nolb,
490 void *buf, size_t bufsize)
491{
492 struct fd_dev *fd_dev = FD_DEV(se_dev);
493 struct file *prot_fd = fd_dev->fd_prot_file;
494 sector_t prot_length, prot;
495 loff_t pos = lba * se_dev->prot_length;
496
497 if (!prot_fd) {
498 pr_err("Unable to locate fd_dev->fd_prot_file\n");
499 return -ENODEV;
500 }
501
502 prot_length = nolb * se_dev->prot_length;
503
504 memset(buf, 0xff, bufsize);
505 for (prot = 0; prot < prot_length;) {
506 sector_t len = min_t(sector_t, bufsize, prot_length - prot);
507 ssize_t ret = kernel_write(prot_fd, buf, len, &pos);
508
509 if (ret != len) {
510 pr_err("vfs_write to prot file failed: %zd\n", ret);
511 return ret < 0 ? ret : -ENODEV;
512 }
513 prot += ret;
514 }
515
516 return 0;
517}
518
519static int
520fd_do_prot_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
521{
522 void *buf;
523 int rc;
524
525 buf = (void *)__get_free_page(GFP_KERNEL);
526 if (!buf) {
527 pr_err("Unable to allocate FILEIO prot buf\n");
528 return -ENOMEM;
529 }
530
531 rc = fd_do_prot_fill(cmd->se_dev, lba, nolb, buf, PAGE_SIZE);
532
533 free_page((unsigned long)buf);
534
535 return rc;
536}
537
538static sense_reason_t
539fd_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
540{
541 struct file *file = FD_DEV(cmd->se_dev)->fd_file;
542 struct inode *inode = file->f_mapping->host;
543 int ret;
544
545 if (!nolb) {
546 return 0;
547 }
548
549 if (cmd->se_dev->dev_attrib.pi_prot_type) {
550 ret = fd_do_prot_unmap(cmd, lba, nolb);
551 if (ret)
552 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
553 }
554
555 if (S_ISBLK(inode->i_mode)) {
556 /* The backend is block device, use discard */
557 struct block_device *bdev = I_BDEV(inode);
558 struct se_device *dev = cmd->se_dev;
559
560 ret = blkdev_issue_discard(bdev,
561 target_to_linux_sector(dev, lba),
562 target_to_linux_sector(dev, nolb),
563 GFP_KERNEL);
564 if (ret < 0) {
565 pr_warn("FILEIO: blkdev_issue_discard() failed: %d\n",
566 ret);
567 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
568 }
569 } else {
570 /* The backend is normal file, use fallocate */
571 struct se_device *se_dev = cmd->se_dev;
572 loff_t pos = lba * se_dev->dev_attrib.block_size;
573 unsigned int len = nolb * se_dev->dev_attrib.block_size;
574 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
575
576 if (!file->f_op->fallocate)
577 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
578
579 ret = file->f_op->fallocate(file, mode, pos, len);
580 if (ret < 0) {
581 pr_warn("FILEIO: fallocate() failed: %d\n", ret);
582 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
583 }
584 }
585
586 return 0;
587}
588
589static sense_reason_t
590fd_execute_rw_buffered(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
591 enum dma_data_direction data_direction)
592{
593 struct se_device *dev = cmd->se_dev;
594 struct fd_dev *fd_dev = FD_DEV(dev);
595 struct file *file = fd_dev->fd_file;
596 struct file *pfile = fd_dev->fd_prot_file;
597 sense_reason_t rc;
598 int ret = 0;
599 /*
600 * Call vectorized fileio functions to map struct scatterlist
601 * physical memory addresses to struct iovec virtual memory.
602 */
603 if (data_direction == DMA_FROM_DEVICE) {
604 if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
605 ret = fd_do_rw(cmd, pfile, dev->prot_length,
606 cmd->t_prot_sg, cmd->t_prot_nents,
607 cmd->prot_length, 0);
608 if (ret < 0)
609 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
610 }
611
612 ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
613 sgl, sgl_nents, cmd->data_length, 0);
614
615 if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type &&
616 dev->dev_attrib.pi_prot_verify) {
617 u32 sectors = cmd->data_length >>
618 ilog2(dev->dev_attrib.block_size);
619
620 rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
621 0, cmd->t_prot_sg, 0);
622 if (rc)
623 return rc;
624 }
625 } else {
626 if (cmd->prot_type && dev->dev_attrib.pi_prot_type &&
627 dev->dev_attrib.pi_prot_verify) {
628 u32 sectors = cmd->data_length >>
629 ilog2(dev->dev_attrib.block_size);
630
631 rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
632 0, cmd->t_prot_sg, 0);
633 if (rc)
634 return rc;
635 }
636
637 ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
638 sgl, sgl_nents, cmd->data_length, 1);
639 /*
640 * Perform implicit vfs_fsync_range() for fd_do_writev() ops
641 * for SCSI WRITEs with Forced Unit Access (FUA) set.
642 * Allow this to happen independent of WCE=0 setting.
643 */
644 if (ret > 0 && (cmd->se_cmd_flags & SCF_FUA)) {
645 loff_t start = cmd->t_task_lba *
646 dev->dev_attrib.block_size;
647 loff_t end;
648
649 if (cmd->data_length)
650 end = start + cmd->data_length - 1;
651 else
652 end = LLONG_MAX;
653
654 vfs_fsync_range(fd_dev->fd_file, start, end, 1);
655 }
656
657 if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type) {
658 ret = fd_do_rw(cmd, pfile, dev->prot_length,
659 cmd->t_prot_sg, cmd->t_prot_nents,
660 cmd->prot_length, 1);
661 if (ret < 0)
662 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
663 }
664 }
665
666 if (ret < 0)
667 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
668
669 target_complete_cmd(cmd, SAM_STAT_GOOD);
670 return 0;
671}
672
673static sense_reason_t
674fd_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
675 enum dma_data_direction data_direction)
676{
677 struct se_device *dev = cmd->se_dev;
678 struct fd_dev *fd_dev = FD_DEV(dev);
679
680 /*
681 * We are currently limited by the number of iovecs (2048) per
682 * single vfs_[writev,readv] call.
683 */
684 if (cmd->data_length > FD_MAX_BYTES) {
685 pr_err("FILEIO: Not able to process I/O of %u bytes due to"
686 "FD_MAX_BYTES: %u iovec count limitation\n",
687 cmd->data_length, FD_MAX_BYTES);
688 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
689 }
690
691 if (fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO)
692 return fd_execute_rw_aio(cmd, sgl, sgl_nents, data_direction);
693 return fd_execute_rw_buffered(cmd, sgl, sgl_nents, data_direction);
694}
695
696enum {
697 Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io,
698 Opt_fd_async_io, Opt_err
699};
700
701static match_table_t tokens = {
702 {Opt_fd_dev_name, "fd_dev_name=%s"},
703 {Opt_fd_dev_size, "fd_dev_size=%s"},
704 {Opt_fd_buffered_io, "fd_buffered_io=%d"},
705 {Opt_fd_async_io, "fd_async_io=%d"},
706 {Opt_err, NULL}
707};
708
709static ssize_t fd_set_configfs_dev_params(struct se_device *dev,
710 const char *page, ssize_t count)
711{
712 struct fd_dev *fd_dev = FD_DEV(dev);
713 char *orig, *ptr, *arg_p, *opts;
714 substring_t args[MAX_OPT_ARGS];
715 int ret = 0, arg, token;
716
717 opts = kstrdup(page, GFP_KERNEL);
718 if (!opts)
719 return -ENOMEM;
720
721 orig = opts;
722
723 while ((ptr = strsep(&opts, ",\n")) != NULL) {
724 if (!*ptr)
725 continue;
726
727 token = match_token(ptr, tokens, args);
728 switch (token) {
729 case Opt_fd_dev_name:
730 if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
731 FD_MAX_DEV_NAME) == 0) {
732 ret = -EINVAL;
733 break;
734 }
735 pr_debug("FILEIO: Referencing Path: %s\n",
736 fd_dev->fd_dev_name);
737 fd_dev->fbd_flags |= FBDF_HAS_PATH;
738 break;
739 case Opt_fd_dev_size:
740 arg_p = match_strdup(&args[0]);
741 if (!arg_p) {
742 ret = -ENOMEM;
743 break;
744 }
745 ret = kstrtoull(arg_p, 0, &fd_dev->fd_dev_size);
746 kfree(arg_p);
747 if (ret < 0) {
748 pr_err("kstrtoull() failed for"
749 " fd_dev_size=\n");
750 goto out;
751 }
752 pr_debug("FILEIO: Referencing Size: %llu"
753 " bytes\n", fd_dev->fd_dev_size);
754 fd_dev->fbd_flags |= FBDF_HAS_SIZE;
755 break;
756 case Opt_fd_buffered_io:
757 ret = match_int(args, &arg);
758 if (ret)
759 goto out;
760 if (arg != 1) {
761 pr_err("bogus fd_buffered_io=%d value\n", arg);
762 ret = -EINVAL;
763 goto out;
764 }
765
766 pr_debug("FILEIO: Using buffered I/O"
767 " operations for struct fd_dev\n");
768
769 fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
770 break;
771 case Opt_fd_async_io:
772 ret = match_int(args, &arg);
773 if (ret)
774 goto out;
775 if (arg != 1) {
776 pr_err("bogus fd_async_io=%d value\n", arg);
777 ret = -EINVAL;
778 goto out;
779 }
780
781 pr_debug("FILEIO: Using async I/O"
782 " operations for struct fd_dev\n");
783
784 fd_dev->fbd_flags |= FDBD_HAS_ASYNC_IO;
785 break;
786 default:
787 break;
788 }
789 }
790
791out:
792 kfree(orig);
793 return (!ret) ? count : ret;
794}
795
796static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b)
797{
798 struct fd_dev *fd_dev = FD_DEV(dev);
799 ssize_t bl = 0;
800
801 bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
802 bl += sprintf(b + bl, " File: %s Size: %llu Mode: %s Async: %d\n",
803 fd_dev->fd_dev_name, fd_dev->fd_dev_size,
804 (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
805 "Buffered-WCE" : "O_DSYNC",
806 !!(fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO));
807 return bl;
808}
809
810static sector_t fd_get_blocks(struct se_device *dev)
811{
812 struct fd_dev *fd_dev = FD_DEV(dev);
813 struct file *f = fd_dev->fd_file;
814 struct inode *i = f->f_mapping->host;
815 unsigned long long dev_size;
816 /*
817 * When using a file that references an underlying struct block_device,
818 * ensure dev_size is always based on the current inode size in order
819 * to handle underlying block_device resize operations.
820 */
821 if (S_ISBLK(i->i_mode))
822 dev_size = i_size_read(i);
823 else
824 dev_size = fd_dev->fd_dev_size;
825
826 return div_u64(dev_size - dev->dev_attrib.block_size,
827 dev->dev_attrib.block_size);
828}
829
830static int fd_init_prot(struct se_device *dev)
831{
832 struct fd_dev *fd_dev = FD_DEV(dev);
833 struct file *prot_file, *file = fd_dev->fd_file;
834 struct inode *inode;
835 int ret, flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
836 char buf[FD_MAX_DEV_PROT_NAME];
837
838 if (!file) {
839 pr_err("Unable to locate fd_dev->fd_file\n");
840 return -ENODEV;
841 }
842
843 inode = file->f_mapping->host;
844 if (S_ISBLK(inode->i_mode)) {
845 pr_err("FILEIO Protection emulation only supported on"
846 " !S_ISBLK\n");
847 return -ENOSYS;
848 }
849
850 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE)
851 flags &= ~O_DSYNC;
852
853 snprintf(buf, FD_MAX_DEV_PROT_NAME, "%s.protection",
854 fd_dev->fd_dev_name);
855
856 prot_file = filp_open(buf, flags, 0600);
857 if (IS_ERR(prot_file)) {
858 pr_err("filp_open(%s) failed\n", buf);
859 ret = PTR_ERR(prot_file);
860 return ret;
861 }
862 fd_dev->fd_prot_file = prot_file;
863
864 return 0;
865}
866
867static int fd_format_prot(struct se_device *dev)
868{
869 unsigned char *buf;
870 int unit_size = FDBD_FORMAT_UNIT_SIZE * dev->dev_attrib.block_size;
871 int ret;
872
873 if (!dev->dev_attrib.pi_prot_type) {
874 pr_err("Unable to format_prot while pi_prot_type == 0\n");
875 return -ENODEV;
876 }
877
878 buf = vzalloc(unit_size);
879 if (!buf) {
880 pr_err("Unable to allocate FILEIO prot buf\n");
881 return -ENOMEM;
882 }
883
884 pr_debug("Using FILEIO prot_length: %llu\n",
885 (unsigned long long)(dev->transport->get_blocks(dev) + 1) *
886 dev->prot_length);
887
888 ret = fd_do_prot_fill(dev, 0, dev->transport->get_blocks(dev) + 1,
889 buf, unit_size);
890 vfree(buf);
891 return ret;
892}
893
894static void fd_free_prot(struct se_device *dev)
895{
896 struct fd_dev *fd_dev = FD_DEV(dev);
897
898 if (!fd_dev->fd_prot_file)
899 return;
900
901 filp_close(fd_dev->fd_prot_file, NULL);
902 fd_dev->fd_prot_file = NULL;
903}
904
905static struct sbc_ops fd_sbc_ops = {
906 .execute_rw = fd_execute_rw,
907 .execute_sync_cache = fd_execute_sync_cache,
908 .execute_write_same = fd_execute_write_same,
909 .execute_unmap = fd_execute_unmap,
910};
911
912static sense_reason_t
913fd_parse_cdb(struct se_cmd *cmd)
914{
915 return sbc_parse_cdb(cmd, &fd_sbc_ops);
916}
917
918static const struct target_backend_ops fileio_ops = {
919 .name = "fileio",
920 .inquiry_prod = "FILEIO",
921 .inquiry_rev = FD_VERSION,
922 .owner = THIS_MODULE,
923 .attach_hba = fd_attach_hba,
924 .detach_hba = fd_detach_hba,
925 .alloc_device = fd_alloc_device,
926 .configure_device = fd_configure_device,
927 .destroy_device = fd_destroy_device,
928 .free_device = fd_free_device,
929 .configure_unmap = fd_configure_unmap,
930 .parse_cdb = fd_parse_cdb,
931 .set_configfs_dev_params = fd_set_configfs_dev_params,
932 .show_configfs_dev_params = fd_show_configfs_dev_params,
933 .get_device_type = sbc_get_device_type,
934 .get_blocks = fd_get_blocks,
935 .init_prot = fd_init_prot,
936 .format_prot = fd_format_prot,
937 .free_prot = fd_free_prot,
938 .tb_dev_attrib_attrs = sbc_attrib_attrs,
939};
940
941static int __init fileio_module_init(void)
942{
943 return transport_backend_register(&fileio_ops);
944}
945
946static void __exit fileio_module_exit(void)
947{
948 target_backend_unregister(&fileio_ops);
949}
950
951MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
952MODULE_AUTHOR("nab@Linux-iSCSI.org");
953MODULE_LICENSE("GPL");
954
955module_init(fileio_module_init);
956module_exit(fileio_module_exit);
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
41#include "target_core_file.h"
42
43static inline struct fd_dev *FD_DEV(struct se_device *dev)
44{
45 return container_of(dev, struct fd_dev, dev);
46}
47
48/* fd_attach_hba(): (Part of se_subsystem_api_t template)
49 *
50 *
51 */
52static int fd_attach_hba(struct se_hba *hba, u32 host_id)
53{
54 struct fd_host *fd_host;
55
56 fd_host = kzalloc(sizeof(struct fd_host), GFP_KERNEL);
57 if (!fd_host) {
58 pr_err("Unable to allocate memory for struct fd_host\n");
59 return -ENOMEM;
60 }
61
62 fd_host->fd_host_id = host_id;
63
64 hba->hba_ptr = fd_host;
65
66 pr_debug("CORE_HBA[%d] - TCM FILEIO HBA Driver %s on Generic"
67 " Target Core Stack %s\n", hba->hba_id, FD_VERSION,
68 TARGET_CORE_MOD_VERSION);
69 pr_debug("CORE_HBA[%d] - Attached FILEIO HBA: %u to Generic\n",
70 hba->hba_id, fd_host->fd_host_id);
71
72 return 0;
73}
74
75static void fd_detach_hba(struct se_hba *hba)
76{
77 struct fd_host *fd_host = hba->hba_ptr;
78
79 pr_debug("CORE_HBA[%d] - Detached FILEIO HBA: %u from Generic"
80 " Target Core\n", hba->hba_id, fd_host->fd_host_id);
81
82 kfree(fd_host);
83 hba->hba_ptr = NULL;
84}
85
86static struct se_device *fd_alloc_device(struct se_hba *hba, const char *name)
87{
88 struct fd_dev *fd_dev;
89 struct fd_host *fd_host = hba->hba_ptr;
90
91 fd_dev = kzalloc(sizeof(struct fd_dev), GFP_KERNEL);
92 if (!fd_dev) {
93 pr_err("Unable to allocate memory for struct fd_dev\n");
94 return NULL;
95 }
96
97 fd_dev->fd_host = fd_host;
98
99 pr_debug("FILEIO: Allocated fd_dev for %p\n", name);
100
101 return &fd_dev->dev;
102}
103
104static int fd_configure_device(struct se_device *dev)
105{
106 struct fd_dev *fd_dev = FD_DEV(dev);
107 struct fd_host *fd_host = dev->se_hba->hba_ptr;
108 struct file *file;
109 struct inode *inode = NULL;
110 int flags, ret = -EINVAL;
111
112 if (!(fd_dev->fbd_flags & FBDF_HAS_PATH)) {
113 pr_err("Missing fd_dev_name=\n");
114 return -EINVAL;
115 }
116
117 /*
118 * Use O_DSYNC by default instead of O_SYNC to forgo syncing
119 * of pure timestamp updates.
120 */
121 flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
122
123 /*
124 * Optionally allow fd_buffered_io=1 to be enabled for people
125 * who want use the fs buffer cache as an WriteCache mechanism.
126 *
127 * This means that in event of a hard failure, there is a risk
128 * of silent data-loss if the SCSI client has *not* performed a
129 * forced unit access (FUA) write, or issued SYNCHRONIZE_CACHE
130 * to write-out the entire device cache.
131 */
132 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
133 pr_debug("FILEIO: Disabling O_DSYNC, using buffered FILEIO\n");
134 flags &= ~O_DSYNC;
135 }
136
137 file = filp_open(fd_dev->fd_dev_name, flags, 0600);
138 if (IS_ERR(file)) {
139 pr_err("filp_open(%s) failed\n", fd_dev->fd_dev_name);
140 ret = PTR_ERR(file);
141 goto fail;
142 }
143 fd_dev->fd_file = file;
144 /*
145 * If using a block backend with this struct file, we extract
146 * fd_dev->fd_[block,dev]_size from struct block_device.
147 *
148 * Otherwise, we use the passed fd_size= from configfs
149 */
150 inode = file->f_mapping->host;
151 if (S_ISBLK(inode->i_mode)) {
152 struct request_queue *q = bdev_get_queue(inode->i_bdev);
153 unsigned long long dev_size;
154
155 fd_dev->fd_block_size = bdev_logical_block_size(inode->i_bdev);
156 /*
157 * Determine the number of bytes from i_size_read() minus
158 * one (1) logical sector from underlying struct block_device
159 */
160 dev_size = (i_size_read(file->f_mapping->host) -
161 fd_dev->fd_block_size);
162
163 pr_debug("FILEIO: Using size: %llu bytes from struct"
164 " block_device blocks: %llu logical_block_size: %d\n",
165 dev_size, div_u64(dev_size, fd_dev->fd_block_size),
166 fd_dev->fd_block_size);
167 /*
168 * Check if the underlying struct block_device request_queue supports
169 * the QUEUE_FLAG_DISCARD bit for UNMAP/WRITE_SAME in SCSI + TRIM
170 * in ATA and we need to set TPE=1
171 */
172 if (blk_queue_discard(q)) {
173 dev->dev_attrib.max_unmap_lba_count =
174 q->limits.max_discard_sectors;
175 /*
176 * Currently hardcoded to 1 in Linux/SCSI code..
177 */
178 dev->dev_attrib.max_unmap_block_desc_count = 1;
179 dev->dev_attrib.unmap_granularity =
180 q->limits.discard_granularity >> 9;
181 dev->dev_attrib.unmap_granularity_alignment =
182 q->limits.discard_alignment;
183 pr_debug("IFILE: BLOCK Discard support available,"
184 " disabled by default\n");
185 }
186 /*
187 * Enable write same emulation for IBLOCK and use 0xFFFF as
188 * the smaller WRITE_SAME(10) only has a two-byte block count.
189 */
190 dev->dev_attrib.max_write_same_len = 0xFFFF;
191
192 if (blk_queue_nonrot(q))
193 dev->dev_attrib.is_nonrot = 1;
194 } else {
195 if (!(fd_dev->fbd_flags & FBDF_HAS_SIZE)) {
196 pr_err("FILEIO: Missing fd_dev_size="
197 " parameter, and no backing struct"
198 " block_device\n");
199 goto fail;
200 }
201
202 fd_dev->fd_block_size = FD_BLOCKSIZE;
203 /*
204 * Limit UNMAP emulation to 8k Number of LBAs (NoLB)
205 */
206 dev->dev_attrib.max_unmap_lba_count = 0x2000;
207 /*
208 * Currently hardcoded to 1 in Linux/SCSI code..
209 */
210 dev->dev_attrib.max_unmap_block_desc_count = 1;
211 dev->dev_attrib.unmap_granularity = 1;
212 dev->dev_attrib.unmap_granularity_alignment = 0;
213
214 /*
215 * Limit WRITE_SAME w/ UNMAP=0 emulation to 8k Number of LBAs (NoLB)
216 * based upon struct iovec limit for vfs_writev()
217 */
218 dev->dev_attrib.max_write_same_len = 0x1000;
219 }
220
221 dev->dev_attrib.hw_block_size = fd_dev->fd_block_size;
222 dev->dev_attrib.max_bytes_per_io = FD_MAX_BYTES;
223 dev->dev_attrib.hw_max_sectors = FD_MAX_BYTES / fd_dev->fd_block_size;
224 dev->dev_attrib.hw_queue_depth = FD_MAX_DEVICE_QUEUE_DEPTH;
225
226 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
227 pr_debug("FILEIO: Forcing setting of emulate_write_cache=1"
228 " with FDBD_HAS_BUFFERED_IO_WCE\n");
229 dev->dev_attrib.emulate_write_cache = 1;
230 }
231
232 fd_dev->fd_dev_id = fd_host->fd_host_dev_id_count++;
233 fd_dev->fd_queue_depth = dev->queue_depth;
234
235 pr_debug("CORE_FILE[%u] - Added TCM FILEIO Device ID: %u at %s,"
236 " %llu total bytes\n", fd_host->fd_host_id, fd_dev->fd_dev_id,
237 fd_dev->fd_dev_name, fd_dev->fd_dev_size);
238
239 return 0;
240fail:
241 if (fd_dev->fd_file) {
242 filp_close(fd_dev->fd_file, NULL);
243 fd_dev->fd_file = NULL;
244 }
245 return ret;
246}
247
248static void fd_free_device(struct se_device *dev)
249{
250 struct fd_dev *fd_dev = FD_DEV(dev);
251
252 if (fd_dev->fd_file) {
253 filp_close(fd_dev->fd_file, NULL);
254 fd_dev->fd_file = NULL;
255 }
256
257 kfree(fd_dev);
258}
259
260static int fd_do_prot_rw(struct se_cmd *cmd, struct fd_prot *fd_prot,
261 int is_write)
262{
263 struct se_device *se_dev = cmd->se_dev;
264 struct fd_dev *dev = FD_DEV(se_dev);
265 struct file *prot_fd = dev->fd_prot_file;
266 struct scatterlist *sg;
267 loff_t pos = (cmd->t_task_lba * se_dev->prot_length);
268 unsigned char *buf;
269 u32 prot_size, len, size;
270 int rc, ret = 1, i;
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 = vzalloc(prot_size);
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 = cmd->t_prot_nents;
284 fd_prot->prot_sg = kzalloc(sizeof(struct scatterlist) *
285 fd_prot->prot_sg_nents, GFP_KERNEL);
286 if (!fd_prot->prot_sg) {
287 pr_err("Unable to allocate fd_prot->prot_sg\n");
288 vfree(fd_prot->prot_buf);
289 return -ENOMEM;
290 }
291 size = prot_size;
292
293 for_each_sg(fd_prot->prot_sg, sg, fd_prot->prot_sg_nents, i) {
294
295 len = min_t(u32, PAGE_SIZE, size);
296 sg_set_buf(sg, buf, len);
297 size -= len;
298 buf += len;
299 }
300 }
301
302 if (is_write) {
303 rc = kernel_write(prot_fd, fd_prot->prot_buf, prot_size, pos);
304 if (rc < 0 || prot_size != rc) {
305 pr_err("kernel_write() for fd_do_prot_rw failed:"
306 " %d\n", rc);
307 ret = -EINVAL;
308 }
309 } else {
310 rc = kernel_read(prot_fd, pos, fd_prot->prot_buf, prot_size);
311 if (rc < 0) {
312 pr_err("kernel_read() for fd_do_prot_rw failed:"
313 " %d\n", rc);
314 ret = -EINVAL;
315 }
316 }
317
318 if (is_write || ret < 0) {
319 kfree(fd_prot->prot_sg);
320 vfree(fd_prot->prot_buf);
321 }
322
323 return ret;
324}
325
326static int fd_do_rw(struct se_cmd *cmd, struct scatterlist *sgl,
327 u32 sgl_nents, int is_write)
328{
329 struct se_device *se_dev = cmd->se_dev;
330 struct fd_dev *dev = FD_DEV(se_dev);
331 struct file *fd = dev->fd_file;
332 struct scatterlist *sg;
333 struct iovec *iov;
334 mm_segment_t old_fs;
335 loff_t pos = (cmd->t_task_lba * se_dev->dev_attrib.block_size);
336 int ret = 0, i;
337
338 iov = kzalloc(sizeof(struct iovec) * sgl_nents, GFP_KERNEL);
339 if (!iov) {
340 pr_err("Unable to allocate fd_do_readv iov[]\n");
341 return -ENOMEM;
342 }
343
344 for_each_sg(sgl, sg, sgl_nents, i) {
345 iov[i].iov_len = sg->length;
346 iov[i].iov_base = kmap(sg_page(sg)) + sg->offset;
347 }
348
349 old_fs = get_fs();
350 set_fs(get_ds());
351
352 if (is_write)
353 ret = vfs_writev(fd, &iov[0], sgl_nents, &pos);
354 else
355 ret = vfs_readv(fd, &iov[0], sgl_nents, &pos);
356
357 set_fs(old_fs);
358
359 for_each_sg(sgl, sg, sgl_nents, i)
360 kunmap(sg_page(sg));
361
362 kfree(iov);
363
364 if (is_write) {
365 if (ret < 0 || ret != cmd->data_length) {
366 pr_err("%s() write returned %d\n", __func__, ret);
367 return (ret < 0 ? ret : -EINVAL);
368 }
369 } else {
370 /*
371 * Return zeros and GOOD status even if the READ did not return
372 * the expected virt_size for struct file w/o a backing struct
373 * block_device.
374 */
375 if (S_ISBLK(file_inode(fd)->i_mode)) {
376 if (ret < 0 || ret != cmd->data_length) {
377 pr_err("%s() returned %d, expecting %u for "
378 "S_ISBLK\n", __func__, ret,
379 cmd->data_length);
380 return (ret < 0 ? ret : -EINVAL);
381 }
382 } else {
383 if (ret < 0) {
384 pr_err("%s() returned %d for non S_ISBLK\n",
385 __func__, ret);
386 return ret;
387 }
388 }
389 }
390 return 1;
391}
392
393static sense_reason_t
394fd_execute_sync_cache(struct se_cmd *cmd)
395{
396 struct se_device *dev = cmd->se_dev;
397 struct fd_dev *fd_dev = FD_DEV(dev);
398 int immed = (cmd->t_task_cdb[1] & 0x2);
399 loff_t start, end;
400 int ret;
401
402 /*
403 * If the Immediate bit is set, queue up the GOOD response
404 * for this SYNCHRONIZE_CACHE op
405 */
406 if (immed)
407 target_complete_cmd(cmd, SAM_STAT_GOOD);
408
409 /*
410 * Determine if we will be flushing the entire device.
411 */
412 if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
413 start = 0;
414 end = LLONG_MAX;
415 } else {
416 start = cmd->t_task_lba * dev->dev_attrib.block_size;
417 if (cmd->data_length)
418 end = start + cmd->data_length;
419 else
420 end = LLONG_MAX;
421 }
422
423 ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
424 if (ret != 0)
425 pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
426
427 if (immed)
428 return 0;
429
430 if (ret)
431 target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
432 else
433 target_complete_cmd(cmd, SAM_STAT_GOOD);
434
435 return 0;
436}
437
438static unsigned char *
439fd_setup_write_same_buf(struct se_cmd *cmd, struct scatterlist *sg,
440 unsigned int len)
441{
442 struct se_device *se_dev = cmd->se_dev;
443 unsigned int block_size = se_dev->dev_attrib.block_size;
444 unsigned int i = 0, end;
445 unsigned char *buf, *p, *kmap_buf;
446
447 buf = kzalloc(min_t(unsigned int, len, PAGE_SIZE), GFP_KERNEL);
448 if (!buf) {
449 pr_err("Unable to allocate fd_execute_write_same buf\n");
450 return NULL;
451 }
452
453 kmap_buf = kmap(sg_page(sg)) + sg->offset;
454 if (!kmap_buf) {
455 pr_err("kmap() failed in fd_setup_write_same\n");
456 kfree(buf);
457 return NULL;
458 }
459 /*
460 * Fill local *buf to contain multiple WRITE_SAME blocks up to
461 * min(len, PAGE_SIZE)
462 */
463 p = buf;
464 end = min_t(unsigned int, len, PAGE_SIZE);
465
466 while (i < end) {
467 memcpy(p, kmap_buf, block_size);
468
469 i += block_size;
470 p += block_size;
471 }
472 kunmap(sg_page(sg));
473
474 return buf;
475}
476
477static sense_reason_t
478fd_execute_write_same(struct se_cmd *cmd)
479{
480 struct se_device *se_dev = cmd->se_dev;
481 struct fd_dev *fd_dev = FD_DEV(se_dev);
482 struct file *f = fd_dev->fd_file;
483 struct scatterlist *sg;
484 struct iovec *iov;
485 mm_segment_t old_fs;
486 sector_t nolb = sbc_get_write_same_sectors(cmd);
487 loff_t pos = cmd->t_task_lba * se_dev->dev_attrib.block_size;
488 unsigned int len, len_tmp, iov_num;
489 int i, rc;
490 unsigned char *buf;
491
492 if (!nolb) {
493 target_complete_cmd(cmd, SAM_STAT_GOOD);
494 return 0;
495 }
496 sg = &cmd->t_data_sg[0];
497
498 if (cmd->t_data_nents > 1 ||
499 sg->length != cmd->se_dev->dev_attrib.block_size) {
500 pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
501 " block_size: %u\n", cmd->t_data_nents, sg->length,
502 cmd->se_dev->dev_attrib.block_size);
503 return TCM_INVALID_CDB_FIELD;
504 }
505
506 len = len_tmp = nolb * se_dev->dev_attrib.block_size;
507 iov_num = DIV_ROUND_UP(len, PAGE_SIZE);
508
509 buf = fd_setup_write_same_buf(cmd, sg, len);
510 if (!buf)
511 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
512
513 iov = vzalloc(sizeof(struct iovec) * iov_num);
514 if (!iov) {
515 pr_err("Unable to allocate fd_execute_write_same iovecs\n");
516 kfree(buf);
517 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
518 }
519 /*
520 * Map the single fabric received scatterlist block now populated
521 * in *buf into each iovec for I/O submission.
522 */
523 for (i = 0; i < iov_num; i++) {
524 iov[i].iov_base = buf;
525 iov[i].iov_len = min_t(unsigned int, len_tmp, PAGE_SIZE);
526 len_tmp -= iov[i].iov_len;
527 }
528
529 old_fs = get_fs();
530 set_fs(get_ds());
531 rc = vfs_writev(f, &iov[0], iov_num, &pos);
532 set_fs(old_fs);
533
534 vfree(iov);
535 kfree(buf);
536
537 if (rc < 0 || rc != len) {
538 pr_err("vfs_writev() returned %d for write same\n", rc);
539 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
540 }
541
542 target_complete_cmd(cmd, SAM_STAT_GOOD);
543 return 0;
544}
545
546static sense_reason_t
547fd_do_unmap(struct se_cmd *cmd, void *priv, sector_t lba, sector_t nolb)
548{
549 struct file *file = priv;
550 struct inode *inode = file->f_mapping->host;
551 int ret;
552
553 if (S_ISBLK(inode->i_mode)) {
554 /* The backend is block device, use discard */
555 struct block_device *bdev = inode->i_bdev;
556
557 ret = blkdev_issue_discard(bdev, lba,
558 nolb, GFP_KERNEL, 0);
559 if (ret < 0) {
560 pr_warn("FILEIO: blkdev_issue_discard() failed: %d\n",
561 ret);
562 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
563 }
564 } else {
565 /* The backend is normal file, use fallocate */
566 struct se_device *se_dev = cmd->se_dev;
567 loff_t pos = lba * se_dev->dev_attrib.block_size;
568 unsigned int len = nolb * se_dev->dev_attrib.block_size;
569 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
570
571 if (!file->f_op->fallocate)
572 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
573
574 ret = file->f_op->fallocate(file, mode, pos, len);
575 if (ret < 0) {
576 pr_warn("FILEIO: fallocate() failed: %d\n", ret);
577 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
578 }
579 }
580
581 return 0;
582}
583
584static sense_reason_t
585fd_execute_write_same_unmap(struct se_cmd *cmd)
586{
587 struct se_device *se_dev = cmd->se_dev;
588 struct fd_dev *fd_dev = FD_DEV(se_dev);
589 struct file *file = fd_dev->fd_file;
590 sector_t lba = cmd->t_task_lba;
591 sector_t nolb = sbc_get_write_same_sectors(cmd);
592 int ret;
593
594 if (!nolb) {
595 target_complete_cmd(cmd, SAM_STAT_GOOD);
596 return 0;
597 }
598
599 ret = fd_do_unmap(cmd, file, lba, nolb);
600 if (ret)
601 return ret;
602
603 target_complete_cmd(cmd, GOOD);
604 return 0;
605}
606
607static sense_reason_t
608fd_execute_unmap(struct se_cmd *cmd)
609{
610 struct file *file = FD_DEV(cmd->se_dev)->fd_file;
611
612 return sbc_execute_unmap(cmd, fd_do_unmap, file);
613}
614
615static sense_reason_t
616fd_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
617 enum dma_data_direction data_direction)
618{
619 struct se_device *dev = cmd->se_dev;
620 struct fd_prot fd_prot;
621 sense_reason_t rc;
622 int ret = 0;
623
624 /*
625 * Call vectorized fileio functions to map struct scatterlist
626 * physical memory addresses to struct iovec virtual memory.
627 */
628 if (data_direction == DMA_FROM_DEVICE) {
629 memset(&fd_prot, 0, sizeof(struct fd_prot));
630
631 if (cmd->prot_type) {
632 ret = fd_do_prot_rw(cmd, &fd_prot, false);
633 if (ret < 0)
634 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
635 }
636
637 ret = fd_do_rw(cmd, sgl, sgl_nents, 0);
638
639 if (ret > 0 && cmd->prot_type) {
640 u32 sectors = cmd->data_length / dev->dev_attrib.block_size;
641
642 rc = sbc_dif_verify_read(cmd, cmd->t_task_lba, sectors,
643 0, fd_prot.prot_sg, 0);
644 if (rc) {
645 kfree(fd_prot.prot_sg);
646 vfree(fd_prot.prot_buf);
647 return rc;
648 }
649 kfree(fd_prot.prot_sg);
650 vfree(fd_prot.prot_buf);
651 }
652 } else {
653 memset(&fd_prot, 0, sizeof(struct fd_prot));
654
655 if (cmd->prot_type) {
656 u32 sectors = cmd->data_length / dev->dev_attrib.block_size;
657
658 ret = fd_do_prot_rw(cmd, &fd_prot, false);
659 if (ret < 0)
660 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
661
662 rc = sbc_dif_verify_write(cmd, cmd->t_task_lba, sectors,
663 0, fd_prot.prot_sg, 0);
664 if (rc) {
665 kfree(fd_prot.prot_sg);
666 vfree(fd_prot.prot_buf);
667 return rc;
668 }
669 }
670
671 ret = fd_do_rw(cmd, sgl, sgl_nents, 1);
672 /*
673 * Perform implicit vfs_fsync_range() for fd_do_writev() ops
674 * for SCSI WRITEs with Forced Unit Access (FUA) set.
675 * Allow this to happen independent of WCE=0 setting.
676 */
677 if (ret > 0 &&
678 dev->dev_attrib.emulate_fua_write > 0 &&
679 (cmd->se_cmd_flags & SCF_FUA)) {
680 struct fd_dev *fd_dev = FD_DEV(dev);
681 loff_t start = cmd->t_task_lba *
682 dev->dev_attrib.block_size;
683 loff_t end = start + cmd->data_length;
684
685 vfs_fsync_range(fd_dev->fd_file, start, end, 1);
686 }
687
688 if (ret > 0 && cmd->prot_type) {
689 ret = fd_do_prot_rw(cmd, &fd_prot, true);
690 if (ret < 0)
691 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
692 }
693 }
694
695 if (ret < 0) {
696 kfree(fd_prot.prot_sg);
697 vfree(fd_prot.prot_buf);
698 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
699 }
700
701 if (ret)
702 target_complete_cmd(cmd, SAM_STAT_GOOD);
703 return 0;
704}
705
706enum {
707 Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io, Opt_err
708};
709
710static match_table_t tokens = {
711 {Opt_fd_dev_name, "fd_dev_name=%s"},
712 {Opt_fd_dev_size, "fd_dev_size=%s"},
713 {Opt_fd_buffered_io, "fd_buffered_io=%d"},
714 {Opt_err, NULL}
715};
716
717static ssize_t fd_set_configfs_dev_params(struct se_device *dev,
718 const char *page, ssize_t count)
719{
720 struct fd_dev *fd_dev = FD_DEV(dev);
721 char *orig, *ptr, *arg_p, *opts;
722 substring_t args[MAX_OPT_ARGS];
723 int ret = 0, arg, token;
724
725 opts = kstrdup(page, GFP_KERNEL);
726 if (!opts)
727 return -ENOMEM;
728
729 orig = opts;
730
731 while ((ptr = strsep(&opts, ",\n")) != NULL) {
732 if (!*ptr)
733 continue;
734
735 token = match_token(ptr, tokens, args);
736 switch (token) {
737 case Opt_fd_dev_name:
738 if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
739 FD_MAX_DEV_NAME) == 0) {
740 ret = -EINVAL;
741 break;
742 }
743 pr_debug("FILEIO: Referencing Path: %s\n",
744 fd_dev->fd_dev_name);
745 fd_dev->fbd_flags |= FBDF_HAS_PATH;
746 break;
747 case Opt_fd_dev_size:
748 arg_p = match_strdup(&args[0]);
749 if (!arg_p) {
750 ret = -ENOMEM;
751 break;
752 }
753 ret = kstrtoull(arg_p, 0, &fd_dev->fd_dev_size);
754 kfree(arg_p);
755 if (ret < 0) {
756 pr_err("kstrtoull() failed for"
757 " fd_dev_size=\n");
758 goto out;
759 }
760 pr_debug("FILEIO: Referencing Size: %llu"
761 " bytes\n", fd_dev->fd_dev_size);
762 fd_dev->fbd_flags |= FBDF_HAS_SIZE;
763 break;
764 case Opt_fd_buffered_io:
765 match_int(args, &arg);
766 if (arg != 1) {
767 pr_err("bogus fd_buffered_io=%d value\n", arg);
768 ret = -EINVAL;
769 goto out;
770 }
771
772 pr_debug("FILEIO: Using buffered I/O"
773 " operations for struct fd_dev\n");
774
775 fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
776 break;
777 default:
778 break;
779 }
780 }
781
782out:
783 kfree(orig);
784 return (!ret) ? count : ret;
785}
786
787static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b)
788{
789 struct fd_dev *fd_dev = FD_DEV(dev);
790 ssize_t bl = 0;
791
792 bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
793 bl += sprintf(b + bl, " File: %s Size: %llu Mode: %s\n",
794 fd_dev->fd_dev_name, fd_dev->fd_dev_size,
795 (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
796 "Buffered-WCE" : "O_DSYNC");
797 return bl;
798}
799
800static sector_t fd_get_blocks(struct se_device *dev)
801{
802 struct fd_dev *fd_dev = FD_DEV(dev);
803 struct file *f = fd_dev->fd_file;
804 struct inode *i = f->f_mapping->host;
805 unsigned long long dev_size;
806 /*
807 * When using a file that references an underlying struct block_device,
808 * ensure dev_size is always based on the current inode size in order
809 * to handle underlying block_device resize operations.
810 */
811 if (S_ISBLK(i->i_mode))
812 dev_size = i_size_read(i);
813 else
814 dev_size = fd_dev->fd_dev_size;
815
816 return div_u64(dev_size - dev->dev_attrib.block_size,
817 dev->dev_attrib.block_size);
818}
819
820static int fd_init_prot(struct se_device *dev)
821{
822 struct fd_dev *fd_dev = FD_DEV(dev);
823 struct file *prot_file, *file = fd_dev->fd_file;
824 struct inode *inode;
825 int ret, flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
826 char buf[FD_MAX_DEV_PROT_NAME];
827
828 if (!file) {
829 pr_err("Unable to locate fd_dev->fd_file\n");
830 return -ENODEV;
831 }
832
833 inode = file->f_mapping->host;
834 if (S_ISBLK(inode->i_mode)) {
835 pr_err("FILEIO Protection emulation only supported on"
836 " !S_ISBLK\n");
837 return -ENOSYS;
838 }
839
840 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE)
841 flags &= ~O_DSYNC;
842
843 snprintf(buf, FD_MAX_DEV_PROT_NAME, "%s.protection",
844 fd_dev->fd_dev_name);
845
846 prot_file = filp_open(buf, flags, 0600);
847 if (IS_ERR(prot_file)) {
848 pr_err("filp_open(%s) failed\n", buf);
849 ret = PTR_ERR(prot_file);
850 return ret;
851 }
852 fd_dev->fd_prot_file = prot_file;
853
854 return 0;
855}
856
857static int fd_format_prot(struct se_device *dev)
858{
859 struct fd_dev *fd_dev = FD_DEV(dev);
860 struct file *prot_fd = fd_dev->fd_prot_file;
861 sector_t prot_length, prot;
862 unsigned char *buf;
863 loff_t pos = 0;
864 int unit_size = FDBD_FORMAT_UNIT_SIZE * dev->dev_attrib.block_size;
865 int rc, ret = 0, size, len;
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 if (!prot_fd) {
872 pr_err("Unable to locate fd_dev->fd_prot_file\n");
873 return -ENODEV;
874 }
875
876 buf = vzalloc(unit_size);
877 if (!buf) {
878 pr_err("Unable to allocate FILEIO prot buf\n");
879 return -ENOMEM;
880 }
881 prot_length = (dev->transport->get_blocks(dev) + 1) * dev->prot_length;
882 size = prot_length;
883
884 pr_debug("Using FILEIO prot_length: %llu\n",
885 (unsigned long long)prot_length);
886
887 memset(buf, 0xff, unit_size);
888 for (prot = 0; prot < prot_length; prot += unit_size) {
889 len = min(unit_size, size);
890 rc = kernel_write(prot_fd, buf, len, pos);
891 if (rc != len) {
892 pr_err("vfs_write to prot file failed: %d\n", rc);
893 ret = -ENODEV;
894 goto out;
895 }
896 pos += len;
897 size -= len;
898 }
899
900out:
901 vfree(buf);
902 return ret;
903}
904
905static void fd_free_prot(struct se_device *dev)
906{
907 struct fd_dev *fd_dev = FD_DEV(dev);
908
909 if (!fd_dev->fd_prot_file)
910 return;
911
912 filp_close(fd_dev->fd_prot_file, NULL);
913 fd_dev->fd_prot_file = NULL;
914}
915
916static struct sbc_ops fd_sbc_ops = {
917 .execute_rw = fd_execute_rw,
918 .execute_sync_cache = fd_execute_sync_cache,
919 .execute_write_same = fd_execute_write_same,
920 .execute_write_same_unmap = fd_execute_write_same_unmap,
921 .execute_unmap = fd_execute_unmap,
922};
923
924static sense_reason_t
925fd_parse_cdb(struct se_cmd *cmd)
926{
927 return sbc_parse_cdb(cmd, &fd_sbc_ops);
928}
929
930static struct se_subsystem_api fileio_template = {
931 .name = "fileio",
932 .inquiry_prod = "FILEIO",
933 .inquiry_rev = FD_VERSION,
934 .owner = THIS_MODULE,
935 .transport_type = TRANSPORT_PLUGIN_VHBA_PDEV,
936 .attach_hba = fd_attach_hba,
937 .detach_hba = fd_detach_hba,
938 .alloc_device = fd_alloc_device,
939 .configure_device = fd_configure_device,
940 .free_device = fd_free_device,
941 .parse_cdb = fd_parse_cdb,
942 .set_configfs_dev_params = fd_set_configfs_dev_params,
943 .show_configfs_dev_params = fd_show_configfs_dev_params,
944 .get_device_type = sbc_get_device_type,
945 .get_blocks = fd_get_blocks,
946 .init_prot = fd_init_prot,
947 .format_prot = fd_format_prot,
948 .free_prot = fd_free_prot,
949};
950
951static int __init fileio_module_init(void)
952{
953 return transport_subsystem_register(&fileio_template);
954}
955
956static void __exit fileio_module_exit(void)
957{
958 transport_subsystem_release(&fileio_template);
959}
960
961MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
962MODULE_AUTHOR("nab@Linux-iSCSI.org");
963MODULE_LICENSE("GPL");
964
965module_init(fileio_module_init);
966module_exit(fileio_module_exit);