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 <scsi/scsi_proto.h>
24#include <asm/unaligned.h>
25
26#include <target/target_core_base.h>
27#include <target/target_core_backend.h>
28
29#include "target_core_file.h"
30
31static inline struct fd_dev *FD_DEV(struct se_device *dev)
32{
33 return container_of(dev, struct fd_dev, dev);
34}
35
36static int fd_attach_hba(struct se_hba *hba, u32 host_id)
37{
38 struct fd_host *fd_host;
39
40 fd_host = kzalloc(sizeof(struct fd_host), GFP_KERNEL);
41 if (!fd_host) {
42 pr_err("Unable to allocate memory for struct fd_host\n");
43 return -ENOMEM;
44 }
45
46 fd_host->fd_host_id = host_id;
47
48 hba->hba_ptr = fd_host;
49
50 pr_debug("CORE_HBA[%d] - TCM FILEIO HBA Driver %s on Generic"
51 " Target Core Stack %s\n", hba->hba_id, FD_VERSION,
52 TARGET_CORE_VERSION);
53 pr_debug("CORE_HBA[%d] - Attached FILEIO HBA: %u to Generic\n",
54 hba->hba_id, fd_host->fd_host_id);
55
56 return 0;
57}
58
59static void fd_detach_hba(struct se_hba *hba)
60{
61 struct fd_host *fd_host = hba->hba_ptr;
62
63 pr_debug("CORE_HBA[%d] - Detached FILEIO HBA: %u from Generic"
64 " Target Core\n", hba->hba_id, fd_host->fd_host_id);
65
66 kfree(fd_host);
67 hba->hba_ptr = NULL;
68}
69
70static struct se_device *fd_alloc_device(struct se_hba *hba, const char *name)
71{
72 struct fd_dev *fd_dev;
73 struct fd_host *fd_host = hba->hba_ptr;
74
75 fd_dev = kzalloc(sizeof(struct fd_dev), GFP_KERNEL);
76 if (!fd_dev) {
77 pr_err("Unable to allocate memory for struct fd_dev\n");
78 return NULL;
79 }
80
81 fd_dev->fd_host = fd_host;
82
83 pr_debug("FILEIO: Allocated fd_dev for %p\n", name);
84
85 return &fd_dev->dev;
86}
87
88static int fd_configure_device(struct se_device *dev)
89{
90 struct fd_dev *fd_dev = FD_DEV(dev);
91 struct fd_host *fd_host = dev->se_hba->hba_ptr;
92 struct file *file;
93 struct inode *inode = NULL;
94 int flags, ret = -EINVAL;
95
96 if (!(fd_dev->fbd_flags & FBDF_HAS_PATH)) {
97 pr_err("Missing fd_dev_name=\n");
98 return -EINVAL;
99 }
100
101 /*
102 * Use O_DSYNC by default instead of O_SYNC to forgo syncing
103 * of pure timestamp updates.
104 */
105 flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
106
107 /*
108 * Optionally allow fd_buffered_io=1 to be enabled for people
109 * who want use the fs buffer cache as an WriteCache mechanism.
110 *
111 * This means that in event of a hard failure, there is a risk
112 * of silent data-loss if the SCSI client has *not* performed a
113 * forced unit access (FUA) write, or issued SYNCHRONIZE_CACHE
114 * to write-out the entire device cache.
115 */
116 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
117 pr_debug("FILEIO: Disabling O_DSYNC, using buffered FILEIO\n");
118 flags &= ~O_DSYNC;
119 }
120
121 file = filp_open(fd_dev->fd_dev_name, flags, 0600);
122 if (IS_ERR(file)) {
123 pr_err("filp_open(%s) failed\n", fd_dev->fd_dev_name);
124 ret = PTR_ERR(file);
125 goto fail;
126 }
127 fd_dev->fd_file = file;
128 /*
129 * If using a block backend with this struct file, we extract
130 * fd_dev->fd_[block,dev]_size from struct block_device.
131 *
132 * Otherwise, we use the passed fd_size= from configfs
133 */
134 inode = file->f_mapping->host;
135 if (S_ISBLK(inode->i_mode)) {
136 struct request_queue *q = bdev_get_queue(inode->i_bdev);
137 unsigned long long dev_size;
138
139 fd_dev->fd_block_size = bdev_logical_block_size(inode->i_bdev);
140 /*
141 * Determine the number of bytes from i_size_read() minus
142 * one (1) logical sector from underlying struct block_device
143 */
144 dev_size = (i_size_read(file->f_mapping->host) -
145 fd_dev->fd_block_size);
146
147 pr_debug("FILEIO: Using size: %llu bytes from struct"
148 " block_device blocks: %llu logical_block_size: %d\n",
149 dev_size, div_u64(dev_size, fd_dev->fd_block_size),
150 fd_dev->fd_block_size);
151
152 if (target_configure_unmap_from_queue(&dev->dev_attrib, q))
153 pr_debug("IFILE: BLOCK Discard support available,"
154 " disabled by default\n");
155 /*
156 * Enable write same emulation for IBLOCK and use 0xFFFF as
157 * the smaller WRITE_SAME(10) only has a two-byte block count.
158 */
159 dev->dev_attrib.max_write_same_len = 0xFFFF;
160
161 if (blk_queue_nonrot(q))
162 dev->dev_attrib.is_nonrot = 1;
163 } else {
164 if (!(fd_dev->fbd_flags & FBDF_HAS_SIZE)) {
165 pr_err("FILEIO: Missing fd_dev_size="
166 " parameter, and no backing struct"
167 " block_device\n");
168 goto fail;
169 }
170
171 fd_dev->fd_block_size = FD_BLOCKSIZE;
172 /*
173 * Limit UNMAP emulation to 8k Number of LBAs (NoLB)
174 */
175 dev->dev_attrib.max_unmap_lba_count = 0x2000;
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 = 1;
181 dev->dev_attrib.unmap_granularity_alignment = 0;
182
183 /*
184 * Limit WRITE_SAME w/ UNMAP=0 emulation to 8k Number of LBAs (NoLB)
185 * based upon struct iovec limit for vfs_writev()
186 */
187 dev->dev_attrib.max_write_same_len = 0x1000;
188 }
189
190 dev->dev_attrib.hw_block_size = fd_dev->fd_block_size;
191 dev->dev_attrib.max_bytes_per_io = FD_MAX_BYTES;
192 dev->dev_attrib.hw_max_sectors = FD_MAX_BYTES / fd_dev->fd_block_size;
193 dev->dev_attrib.hw_queue_depth = FD_MAX_DEVICE_QUEUE_DEPTH;
194
195 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
196 pr_debug("FILEIO: Forcing setting of emulate_write_cache=1"
197 " with FDBD_HAS_BUFFERED_IO_WCE\n");
198 dev->dev_attrib.emulate_write_cache = 1;
199 }
200
201 fd_dev->fd_dev_id = fd_host->fd_host_dev_id_count++;
202 fd_dev->fd_queue_depth = dev->queue_depth;
203
204 pr_debug("CORE_FILE[%u] - Added TCM FILEIO Device ID: %u at %s,"
205 " %llu total bytes\n", fd_host->fd_host_id, fd_dev->fd_dev_id,
206 fd_dev->fd_dev_name, fd_dev->fd_dev_size);
207
208 return 0;
209fail:
210 if (fd_dev->fd_file) {
211 filp_close(fd_dev->fd_file, NULL);
212 fd_dev->fd_file = NULL;
213 }
214 return ret;
215}
216
217static void fd_dev_call_rcu(struct rcu_head *p)
218{
219 struct se_device *dev = container_of(p, struct se_device, rcu_head);
220 struct fd_dev *fd_dev = FD_DEV(dev);
221
222 kfree(fd_dev);
223}
224
225static void fd_free_device(struct se_device *dev)
226{
227 call_rcu(&dev->rcu_head, fd_dev_call_rcu);
228}
229
230static void fd_destroy_device(struct se_device *dev)
231{
232 struct fd_dev *fd_dev = FD_DEV(dev);
233
234 if (fd_dev->fd_file) {
235 filp_close(fd_dev->fd_file, NULL);
236 fd_dev->fd_file = NULL;
237 }
238}
239
240struct target_core_file_cmd {
241 unsigned long len;
242 struct se_cmd *cmd;
243 struct kiocb iocb;
244};
245
246static void cmd_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
247{
248 struct target_core_file_cmd *cmd;
249
250 cmd = container_of(iocb, struct target_core_file_cmd, iocb);
251
252 if (ret != cmd->len)
253 target_complete_cmd(cmd->cmd, SAM_STAT_CHECK_CONDITION);
254 else
255 target_complete_cmd(cmd->cmd, SAM_STAT_GOOD);
256
257 kfree(cmd);
258}
259
260static sense_reason_t
261fd_execute_rw_aio(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
262 enum dma_data_direction data_direction)
263{
264 int is_write = !(data_direction == DMA_FROM_DEVICE);
265 struct se_device *dev = cmd->se_dev;
266 struct fd_dev *fd_dev = FD_DEV(dev);
267 struct file *file = fd_dev->fd_file;
268 struct target_core_file_cmd *aio_cmd;
269 struct iov_iter iter = {};
270 struct scatterlist *sg;
271 struct bio_vec *bvec;
272 ssize_t len = 0;
273 int ret = 0, i;
274
275 aio_cmd = kmalloc(sizeof(struct target_core_file_cmd), GFP_KERNEL);
276 if (!aio_cmd)
277 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
278
279 bvec = kcalloc(sgl_nents, sizeof(struct bio_vec), GFP_KERNEL);
280 if (!bvec) {
281 kfree(aio_cmd);
282 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
283 }
284
285 for_each_sg(sgl, sg, sgl_nents, i) {
286 bvec[i].bv_page = sg_page(sg);
287 bvec[i].bv_len = sg->length;
288 bvec[i].bv_offset = sg->offset;
289
290 len += sg->length;
291 }
292
293 iov_iter_bvec(&iter, is_write, bvec, sgl_nents, len);
294
295 aio_cmd->cmd = cmd;
296 aio_cmd->len = len;
297 aio_cmd->iocb.ki_pos = cmd->t_task_lba * dev->dev_attrib.block_size;
298 aio_cmd->iocb.ki_filp = file;
299 aio_cmd->iocb.ki_complete = cmd_rw_aio_complete;
300 aio_cmd->iocb.ki_flags = IOCB_DIRECT;
301
302 if (is_write && (cmd->se_cmd_flags & SCF_FUA))
303 aio_cmd->iocb.ki_flags |= IOCB_DSYNC;
304
305 if (is_write)
306 ret = call_write_iter(file, &aio_cmd->iocb, &iter);
307 else
308 ret = call_read_iter(file, &aio_cmd->iocb, &iter);
309
310 kfree(bvec);
311
312 if (ret != -EIOCBQUEUED)
313 cmd_rw_aio_complete(&aio_cmd->iocb, ret, 0);
314
315 return 0;
316}
317
318static int fd_do_rw(struct se_cmd *cmd, struct file *fd,
319 u32 block_size, struct scatterlist *sgl,
320 u32 sgl_nents, u32 data_length, int is_write)
321{
322 struct scatterlist *sg;
323 struct iov_iter iter;
324 struct bio_vec *bvec;
325 ssize_t len = 0;
326 loff_t pos = (cmd->t_task_lba * block_size);
327 int ret = 0, i;
328
329 bvec = kcalloc(sgl_nents, sizeof(struct bio_vec), GFP_KERNEL);
330 if (!bvec) {
331 pr_err("Unable to allocate fd_do_readv iov[]\n");
332 return -ENOMEM;
333 }
334
335 for_each_sg(sgl, sg, sgl_nents, i) {
336 bvec[i].bv_page = sg_page(sg);
337 bvec[i].bv_len = sg->length;
338 bvec[i].bv_offset = sg->offset;
339
340 len += sg->length;
341 }
342
343 iov_iter_bvec(&iter, READ, bvec, sgl_nents, len);
344 if (is_write)
345 ret = vfs_iter_write(fd, &iter, &pos, 0);
346 else
347 ret = vfs_iter_read(fd, &iter, &pos, 0);
348
349 if (is_write) {
350 if (ret < 0 || ret != data_length) {
351 pr_err("%s() write returned %d\n", __func__, ret);
352 if (ret >= 0)
353 ret = -EINVAL;
354 }
355 } else {
356 /*
357 * Return zeros and GOOD status even if the READ did not return
358 * the expected virt_size for struct file w/o a backing struct
359 * block_device.
360 */
361 if (S_ISBLK(file_inode(fd)->i_mode)) {
362 if (ret < 0 || ret != data_length) {
363 pr_err("%s() returned %d, expecting %u for "
364 "S_ISBLK\n", __func__, ret,
365 data_length);
366 if (ret >= 0)
367 ret = -EINVAL;
368 }
369 } else {
370 if (ret < 0) {
371 pr_err("%s() returned %d for non S_ISBLK\n",
372 __func__, ret);
373 } else if (ret != data_length) {
374 /*
375 * Short read case:
376 * Probably some one truncate file under us.
377 * We must explicitly zero sg-pages to prevent
378 * expose uninizialized pages to userspace.
379 */
380 if (ret < data_length)
381 ret += iov_iter_zero(data_length - ret, &iter);
382 else
383 ret = -EINVAL;
384 }
385 }
386 }
387 kfree(bvec);
388 return ret;
389}
390
391static sense_reason_t
392fd_execute_sync_cache(struct se_cmd *cmd)
393{
394 struct se_device *dev = cmd->se_dev;
395 struct fd_dev *fd_dev = FD_DEV(dev);
396 int immed = (cmd->t_task_cdb[1] & 0x2);
397 loff_t start, end;
398 int ret;
399
400 /*
401 * If the Immediate bit is set, queue up the GOOD response
402 * for this SYNCHRONIZE_CACHE op
403 */
404 if (immed)
405 target_complete_cmd(cmd, SAM_STAT_GOOD);
406
407 /*
408 * Determine if we will be flushing the entire device.
409 */
410 if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
411 start = 0;
412 end = LLONG_MAX;
413 } else {
414 start = cmd->t_task_lba * dev->dev_attrib.block_size;
415 if (cmd->data_length)
416 end = start + cmd->data_length - 1;
417 else
418 end = LLONG_MAX;
419 }
420
421 ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
422 if (ret != 0)
423 pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
424
425 if (immed)
426 return 0;
427
428 if (ret)
429 target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
430 else
431 target_complete_cmd(cmd, SAM_STAT_GOOD);
432
433 return 0;
434}
435
436static sense_reason_t
437fd_execute_write_same(struct se_cmd *cmd)
438{
439 struct se_device *se_dev = cmd->se_dev;
440 struct fd_dev *fd_dev = FD_DEV(se_dev);
441 loff_t pos = cmd->t_task_lba * se_dev->dev_attrib.block_size;
442 sector_t nolb = sbc_get_write_same_sectors(cmd);
443 struct iov_iter iter;
444 struct bio_vec *bvec;
445 unsigned int len = 0, i;
446 ssize_t ret;
447
448 if (!nolb) {
449 target_complete_cmd(cmd, SAM_STAT_GOOD);
450 return 0;
451 }
452 if (cmd->prot_op) {
453 pr_err("WRITE_SAME: Protection information with FILEIO"
454 " backends not supported\n");
455 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
456 }
457
458 if (cmd->t_data_nents > 1 ||
459 cmd->t_data_sg[0].length != cmd->se_dev->dev_attrib.block_size) {
460 pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
461 " block_size: %u\n",
462 cmd->t_data_nents,
463 cmd->t_data_sg[0].length,
464 cmd->se_dev->dev_attrib.block_size);
465 return TCM_INVALID_CDB_FIELD;
466 }
467
468 bvec = kcalloc(nolb, sizeof(struct bio_vec), GFP_KERNEL);
469 if (!bvec)
470 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
471
472 for (i = 0; i < nolb; i++) {
473 bvec[i].bv_page = sg_page(&cmd->t_data_sg[0]);
474 bvec[i].bv_len = cmd->t_data_sg[0].length;
475 bvec[i].bv_offset = cmd->t_data_sg[0].offset;
476
477 len += se_dev->dev_attrib.block_size;
478 }
479
480 iov_iter_bvec(&iter, READ, bvec, nolb, len);
481 ret = vfs_iter_write(fd_dev->fd_file, &iter, &pos, 0);
482
483 kfree(bvec);
484 if (ret < 0 || ret != len) {
485 pr_err("vfs_iter_write() returned %zd for write same\n", ret);
486 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
487 }
488
489 target_complete_cmd(cmd, SAM_STAT_GOOD);
490 return 0;
491}
492
493static int
494fd_do_prot_fill(struct se_device *se_dev, sector_t lba, sector_t nolb,
495 void *buf, size_t bufsize)
496{
497 struct fd_dev *fd_dev = FD_DEV(se_dev);
498 struct file *prot_fd = fd_dev->fd_prot_file;
499 sector_t prot_length, prot;
500 loff_t pos = lba * se_dev->prot_length;
501
502 if (!prot_fd) {
503 pr_err("Unable to locate fd_dev->fd_prot_file\n");
504 return -ENODEV;
505 }
506
507 prot_length = nolb * se_dev->prot_length;
508
509 for (prot = 0; prot < prot_length;) {
510 sector_t len = min_t(sector_t, bufsize, prot_length - prot);
511 ssize_t ret = kernel_write(prot_fd, buf, len, &pos);
512
513 if (ret != len) {
514 pr_err("vfs_write to prot file failed: %zd\n", ret);
515 return ret < 0 ? ret : -ENODEV;
516 }
517 prot += ret;
518 }
519
520 return 0;
521}
522
523static int
524fd_do_prot_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
525{
526 void *buf;
527 int rc;
528
529 buf = (void *)__get_free_page(GFP_KERNEL);
530 if (!buf) {
531 pr_err("Unable to allocate FILEIO prot buf\n");
532 return -ENOMEM;
533 }
534 memset(buf, 0xff, PAGE_SIZE);
535
536 rc = fd_do_prot_fill(cmd->se_dev, lba, nolb, buf, PAGE_SIZE);
537
538 free_page((unsigned long)buf);
539
540 return rc;
541}
542
543static sense_reason_t
544fd_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
545{
546 struct file *file = FD_DEV(cmd->se_dev)->fd_file;
547 struct inode *inode = file->f_mapping->host;
548 int ret;
549
550 if (!nolb) {
551 return 0;
552 }
553
554 if (cmd->se_dev->dev_attrib.pi_prot_type) {
555 ret = fd_do_prot_unmap(cmd, lba, nolb);
556 if (ret)
557 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
558 }
559
560 if (S_ISBLK(inode->i_mode)) {
561 /* The backend is block device, use discard */
562 struct block_device *bdev = inode->i_bdev;
563 struct se_device *dev = cmd->se_dev;
564
565 ret = blkdev_issue_discard(bdev,
566 target_to_linux_sector(dev, lba),
567 target_to_linux_sector(dev, nolb),
568 GFP_KERNEL, 0);
569 if (ret < 0) {
570 pr_warn("FILEIO: blkdev_issue_discard() failed: %d\n",
571 ret);
572 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
573 }
574 } else {
575 /* The backend is normal file, use fallocate */
576 struct se_device *se_dev = cmd->se_dev;
577 loff_t pos = lba * se_dev->dev_attrib.block_size;
578 unsigned int len = nolb * se_dev->dev_attrib.block_size;
579 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
580
581 if (!file->f_op->fallocate)
582 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
583
584 ret = file->f_op->fallocate(file, mode, pos, len);
585 if (ret < 0) {
586 pr_warn("FILEIO: fallocate() failed: %d\n", ret);
587 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
588 }
589 }
590
591 return 0;
592}
593
594static sense_reason_t
595fd_execute_rw_buffered(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
596 enum dma_data_direction data_direction)
597{
598 struct se_device *dev = cmd->se_dev;
599 struct fd_dev *fd_dev = FD_DEV(dev);
600 struct file *file = fd_dev->fd_file;
601 struct file *pfile = fd_dev->fd_prot_file;
602 sense_reason_t rc;
603 int ret = 0;
604 /*
605 * Call vectorized fileio functions to map struct scatterlist
606 * physical memory addresses to struct iovec virtual memory.
607 */
608 if (data_direction == DMA_FROM_DEVICE) {
609 if (cmd->prot_type && dev->dev_attrib.pi_prot_type) {
610 ret = fd_do_rw(cmd, pfile, dev->prot_length,
611 cmd->t_prot_sg, cmd->t_prot_nents,
612 cmd->prot_length, 0);
613 if (ret < 0)
614 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
615 }
616
617 ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
618 sgl, sgl_nents, cmd->data_length, 0);
619
620 if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type &&
621 dev->dev_attrib.pi_prot_verify) {
622 u32 sectors = cmd->data_length >>
623 ilog2(dev->dev_attrib.block_size);
624
625 rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
626 0, cmd->t_prot_sg, 0);
627 if (rc)
628 return rc;
629 }
630 } else {
631 if (cmd->prot_type && dev->dev_attrib.pi_prot_type &&
632 dev->dev_attrib.pi_prot_verify) {
633 u32 sectors = cmd->data_length >>
634 ilog2(dev->dev_attrib.block_size);
635
636 rc = sbc_dif_verify(cmd, cmd->t_task_lba, sectors,
637 0, cmd->t_prot_sg, 0);
638 if (rc)
639 return rc;
640 }
641
642 ret = fd_do_rw(cmd, file, dev->dev_attrib.block_size,
643 sgl, sgl_nents, cmd->data_length, 1);
644 /*
645 * Perform implicit vfs_fsync_range() for fd_do_writev() ops
646 * for SCSI WRITEs with Forced Unit Access (FUA) set.
647 * Allow this to happen independent of WCE=0 setting.
648 */
649 if (ret > 0 && (cmd->se_cmd_flags & SCF_FUA)) {
650 loff_t start = cmd->t_task_lba *
651 dev->dev_attrib.block_size;
652 loff_t end;
653
654 if (cmd->data_length)
655 end = start + cmd->data_length - 1;
656 else
657 end = LLONG_MAX;
658
659 vfs_fsync_range(fd_dev->fd_file, start, end, 1);
660 }
661
662 if (ret > 0 && cmd->prot_type && dev->dev_attrib.pi_prot_type) {
663 ret = fd_do_rw(cmd, pfile, dev->prot_length,
664 cmd->t_prot_sg, cmd->t_prot_nents,
665 cmd->prot_length, 1);
666 if (ret < 0)
667 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
668 }
669 }
670
671 if (ret < 0)
672 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
673
674 target_complete_cmd(cmd, SAM_STAT_GOOD);
675 return 0;
676}
677
678static sense_reason_t
679fd_execute_rw(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
680 enum dma_data_direction data_direction)
681{
682 struct se_device *dev = cmd->se_dev;
683 struct fd_dev *fd_dev = FD_DEV(dev);
684
685 /*
686 * We are currently limited by the number of iovecs (2048) per
687 * single vfs_[writev,readv] call.
688 */
689 if (cmd->data_length > FD_MAX_BYTES) {
690 pr_err("FILEIO: Not able to process I/O of %u bytes due to"
691 "FD_MAX_BYTES: %u iovec count limitation\n",
692 cmd->data_length, FD_MAX_BYTES);
693 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
694 }
695
696 if (fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO)
697 return fd_execute_rw_aio(cmd, sgl, sgl_nents, data_direction);
698 return fd_execute_rw_buffered(cmd, sgl, sgl_nents, data_direction);
699}
700
701enum {
702 Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io,
703 Opt_fd_async_io, Opt_err
704};
705
706static match_table_t tokens = {
707 {Opt_fd_dev_name, "fd_dev_name=%s"},
708 {Opt_fd_dev_size, "fd_dev_size=%s"},
709 {Opt_fd_buffered_io, "fd_buffered_io=%d"},
710 {Opt_fd_async_io, "fd_async_io=%d"},
711 {Opt_err, NULL}
712};
713
714static ssize_t fd_set_configfs_dev_params(struct se_device *dev,
715 const char *page, ssize_t count)
716{
717 struct fd_dev *fd_dev = FD_DEV(dev);
718 char *orig, *ptr, *arg_p, *opts;
719 substring_t args[MAX_OPT_ARGS];
720 int ret = 0, arg, token;
721
722 opts = kstrdup(page, GFP_KERNEL);
723 if (!opts)
724 return -ENOMEM;
725
726 orig = opts;
727
728 while ((ptr = strsep(&opts, ",\n")) != NULL) {
729 if (!*ptr)
730 continue;
731
732 token = match_token(ptr, tokens, args);
733 switch (token) {
734 case Opt_fd_dev_name:
735 if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
736 FD_MAX_DEV_NAME) == 0) {
737 ret = -EINVAL;
738 break;
739 }
740 pr_debug("FILEIO: Referencing Path: %s\n",
741 fd_dev->fd_dev_name);
742 fd_dev->fbd_flags |= FBDF_HAS_PATH;
743 break;
744 case Opt_fd_dev_size:
745 arg_p = match_strdup(&args[0]);
746 if (!arg_p) {
747 ret = -ENOMEM;
748 break;
749 }
750 ret = kstrtoull(arg_p, 0, &fd_dev->fd_dev_size);
751 kfree(arg_p);
752 if (ret < 0) {
753 pr_err("kstrtoull() failed for"
754 " fd_dev_size=\n");
755 goto out;
756 }
757 pr_debug("FILEIO: Referencing Size: %llu"
758 " bytes\n", fd_dev->fd_dev_size);
759 fd_dev->fbd_flags |= FBDF_HAS_SIZE;
760 break;
761 case Opt_fd_buffered_io:
762 ret = match_int(args, &arg);
763 if (ret)
764 goto out;
765 if (arg != 1) {
766 pr_err("bogus fd_buffered_io=%d value\n", arg);
767 ret = -EINVAL;
768 goto out;
769 }
770
771 pr_debug("FILEIO: Using buffered I/O"
772 " operations for struct fd_dev\n");
773
774 fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
775 break;
776 case Opt_fd_async_io:
777 ret = match_int(args, &arg);
778 if (ret)
779 goto out;
780 if (arg != 1) {
781 pr_err("bogus fd_async_io=%d value\n", arg);
782 ret = -EINVAL;
783 goto out;
784 }
785
786 pr_debug("FILEIO: Using async I/O"
787 " operations for struct fd_dev\n");
788
789 fd_dev->fbd_flags |= FDBD_HAS_ASYNC_IO;
790 break;
791 default:
792 break;
793 }
794 }
795
796out:
797 kfree(orig);
798 return (!ret) ? count : ret;
799}
800
801static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b)
802{
803 struct fd_dev *fd_dev = FD_DEV(dev);
804 ssize_t bl = 0;
805
806 bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
807 bl += sprintf(b + bl, " File: %s Size: %llu Mode: %s Async: %d\n",
808 fd_dev->fd_dev_name, fd_dev->fd_dev_size,
809 (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
810 "Buffered-WCE" : "O_DSYNC",
811 !!(fd_dev->fbd_flags & FDBD_HAS_ASYNC_IO));
812 return bl;
813}
814
815static sector_t fd_get_blocks(struct se_device *dev)
816{
817 struct fd_dev *fd_dev = FD_DEV(dev);
818 struct file *f = fd_dev->fd_file;
819 struct inode *i = f->f_mapping->host;
820 unsigned long long dev_size;
821 /*
822 * When using a file that references an underlying struct block_device,
823 * ensure dev_size is always based on the current inode size in order
824 * to handle underlying block_device resize operations.
825 */
826 if (S_ISBLK(i->i_mode))
827 dev_size = i_size_read(i);
828 else
829 dev_size = fd_dev->fd_dev_size;
830
831 return div_u64(dev_size - dev->dev_attrib.block_size,
832 dev->dev_attrib.block_size);
833}
834
835static int fd_init_prot(struct se_device *dev)
836{
837 struct fd_dev *fd_dev = FD_DEV(dev);
838 struct file *prot_file, *file = fd_dev->fd_file;
839 struct inode *inode;
840 int ret, flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
841 char buf[FD_MAX_DEV_PROT_NAME];
842
843 if (!file) {
844 pr_err("Unable to locate fd_dev->fd_file\n");
845 return -ENODEV;
846 }
847
848 inode = file->f_mapping->host;
849 if (S_ISBLK(inode->i_mode)) {
850 pr_err("FILEIO Protection emulation only supported on"
851 " !S_ISBLK\n");
852 return -ENOSYS;
853 }
854
855 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE)
856 flags &= ~O_DSYNC;
857
858 snprintf(buf, FD_MAX_DEV_PROT_NAME, "%s.protection",
859 fd_dev->fd_dev_name);
860
861 prot_file = filp_open(buf, flags, 0600);
862 if (IS_ERR(prot_file)) {
863 pr_err("filp_open(%s) failed\n", buf);
864 ret = PTR_ERR(prot_file);
865 return ret;
866 }
867 fd_dev->fd_prot_file = prot_file;
868
869 return 0;
870}
871
872static int fd_format_prot(struct se_device *dev)
873{
874 unsigned char *buf;
875 int unit_size = FDBD_FORMAT_UNIT_SIZE * dev->dev_attrib.block_size;
876 int ret;
877
878 if (!dev->dev_attrib.pi_prot_type) {
879 pr_err("Unable to format_prot while pi_prot_type == 0\n");
880 return -ENODEV;
881 }
882
883 buf = vzalloc(unit_size);
884 if (!buf) {
885 pr_err("Unable to allocate FILEIO prot buf\n");
886 return -ENOMEM;
887 }
888
889 pr_debug("Using FILEIO prot_length: %llu\n",
890 (unsigned long long)(dev->transport->get_blocks(dev) + 1) *
891 dev->prot_length);
892
893 memset(buf, 0xff, unit_size);
894 ret = fd_do_prot_fill(dev, 0, dev->transport->get_blocks(dev) + 1,
895 buf, unit_size);
896 vfree(buf);
897 return ret;
898}
899
900static void fd_free_prot(struct se_device *dev)
901{
902 struct fd_dev *fd_dev = FD_DEV(dev);
903
904 if (!fd_dev->fd_prot_file)
905 return;
906
907 filp_close(fd_dev->fd_prot_file, NULL);
908 fd_dev->fd_prot_file = NULL;
909}
910
911static struct sbc_ops fd_sbc_ops = {
912 .execute_rw = fd_execute_rw,
913 .execute_sync_cache = fd_execute_sync_cache,
914 .execute_write_same = fd_execute_write_same,
915 .execute_unmap = fd_execute_unmap,
916};
917
918static sense_reason_t
919fd_parse_cdb(struct se_cmd *cmd)
920{
921 return sbc_parse_cdb(cmd, &fd_sbc_ops);
922}
923
924static const struct target_backend_ops fileio_ops = {
925 .name = "fileio",
926 .inquiry_prod = "FILEIO",
927 .inquiry_rev = FD_VERSION,
928 .owner = THIS_MODULE,
929 .attach_hba = fd_attach_hba,
930 .detach_hba = fd_detach_hba,
931 .alloc_device = fd_alloc_device,
932 .configure_device = fd_configure_device,
933 .destroy_device = fd_destroy_device,
934 .free_device = fd_free_device,
935 .parse_cdb = fd_parse_cdb,
936 .set_configfs_dev_params = fd_set_configfs_dev_params,
937 .show_configfs_dev_params = fd_show_configfs_dev_params,
938 .get_device_type = sbc_get_device_type,
939 .get_blocks = fd_get_blocks,
940 .init_prot = fd_init_prot,
941 .format_prot = fd_format_prot,
942 .free_prot = fd_free_prot,
943 .tb_dev_attrib_attrs = sbc_attrib_attrs,
944};
945
946static int __init fileio_module_init(void)
947{
948 return transport_backend_register(&fileio_ops);
949}
950
951static void __exit fileio_module_exit(void)
952{
953 target_backend_unregister(&fileio_ops);
954}
955
956MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
957MODULE_AUTHOR("nab@Linux-iSCSI.org");
958MODULE_LICENSE("GPL");
959
960module_init(fileio_module_init);
961module_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);