1 /*******************************************************************************
2 * Filename: target_core_file.c
4 * This file contains the Storage Engine <-> FILEIO transport specific functions
6 * (c) Copyright 2005-2012 RisingTide Systems LLC.
8 * Nicholas A. Bellinger <nab@kernel.org>
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.
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.
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.
24 ******************************************************************************/
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>
38 #include <target/target_core_base.h>
39 #include <target/target_core_backend.h>
41 #include "target_core_file.h"
43 static inline struct fd_dev *FD_DEV(struct se_device *dev)
45 return container_of(dev, struct fd_dev, dev);
48 /* fd_attach_hba(): (Part of se_subsystem_api_t template)
52 static int fd_attach_hba(struct se_hba *hba, u32 host_id)
54 struct fd_host *fd_host;
56 fd_host = kzalloc(sizeof(struct fd_host), GFP_KERNEL);
58 pr_err("Unable to allocate memory for struct fd_host\n");
62 fd_host->fd_host_id = host_id;
64 hba->hba_ptr = fd_host;
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);
75 static void fd_detach_hba(struct se_hba *hba)
77 struct fd_host *fd_host = hba->hba_ptr;
79 pr_debug("CORE_HBA[%d] - Detached FILEIO HBA: %u from Generic"
80 " Target Core\n", hba->hba_id, fd_host->fd_host_id);
86 static struct se_device *fd_alloc_device(struct se_hba *hba, const char *name)
88 struct fd_dev *fd_dev;
89 struct fd_host *fd_host = hba->hba_ptr;
91 fd_dev = kzalloc(sizeof(struct fd_dev), GFP_KERNEL);
93 pr_err("Unable to allocate memory for struct fd_dev\n");
97 fd_dev->fd_host = fd_host;
99 pr_debug("FILEIO: Allocated fd_dev for %p\n", name);
104 static int fd_configure_device(struct se_device *dev)
106 struct fd_dev *fd_dev = FD_DEV(dev);
107 struct fd_host *fd_host = dev->se_hba->hba_ptr;
109 struct inode *inode = NULL;
110 int flags, ret = -EINVAL;
112 if (!(fd_dev->fbd_flags & FBDF_HAS_PATH)) {
113 pr_err("Missing fd_dev_name=\n");
118 * Use O_DSYNC by default instead of O_SYNC to forgo syncing
119 * of pure timestamp updates.
121 flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;
124 * Optionally allow fd_buffered_io=1 to be enabled for people
125 * who want use the fs buffer cache as an WriteCache mechanism.
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.
132 if (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) {
133 pr_debug("FILEIO: Disabling O_DSYNC, using buffered FILEIO\n");
137 file = filp_open(fd_dev->fd_dev_name, flags, 0600);
139 pr_err("filp_open(%s) failed\n", fd_dev->fd_dev_name);
143 fd_dev->fd_file = file;
145 * If using a block backend with this struct file, we extract
146 * fd_dev->fd_[block,dev]_size from struct block_device.
148 * Otherwise, we use the passed fd_size= from configfs
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;
155 fd_dev->fd_block_size = bdev_logical_block_size(inode->i_bdev);
157 * Determine the number of bytes from i_size_read() minus
158 * one (1) logical sector from underlying struct block_device
160 dev_size = (i_size_read(file->f_mapping->host) -
161 fd_dev->fd_block_size);
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);
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
172 if (blk_queue_discard(q)) {
173 dev->dev_attrib.max_unmap_lba_count =
174 q->limits.max_discard_sectors;
176 * Currently hardcoded to 1 in Linux/SCSI code..
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");
187 * Enable write same emulation for IBLOCK and use 0xFFFF as
188 * the smaller WRITE_SAME(10) only has a two-byte block count.
190 dev->dev_attrib.max_write_same_len = 0xFFFF;
192 if (blk_queue_nonrot(q))
193 dev->dev_attrib.is_nonrot = 1;
195 if (!(fd_dev->fbd_flags & FBDF_HAS_SIZE)) {
196 pr_err("FILEIO: Missing fd_dev_size="
197 " parameter, and no backing struct"
202 fd_dev->fd_block_size = FD_BLOCKSIZE;
204 * Limit UNMAP emulation to 8k Number of LBAs (NoLB)
206 dev->dev_attrib.max_unmap_lba_count = 0x2000;
208 * Currently hardcoded to 1 in Linux/SCSI code..
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;
215 * Limit WRITE_SAME w/ UNMAP=0 emulation to 8k Number of LBAs (NoLB)
216 * based upon struct iovec limit for vfs_writev()
218 dev->dev_attrib.max_write_same_len = 0x1000;
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;
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;
232 fd_dev->fd_dev_id = fd_host->fd_host_dev_id_count++;
233 fd_dev->fd_queue_depth = dev->queue_depth;
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);
241 if (fd_dev->fd_file) {
242 filp_close(fd_dev->fd_file, NULL);
243 fd_dev->fd_file = NULL;
248 static void fd_free_device(struct se_device *dev)
250 struct fd_dev *fd_dev = FD_DEV(dev);
252 if (fd_dev->fd_file) {
253 filp_close(fd_dev->fd_file, NULL);
254 fd_dev->fd_file = NULL;
260 static int fd_do_rw(struct se_cmd *cmd, struct scatterlist *sgl,
261 u32 sgl_nents, int is_write)
263 struct se_device *se_dev = cmd->se_dev;
264 struct fd_dev *dev = FD_DEV(se_dev);
265 struct file *fd = dev->fd_file;
266 struct scatterlist *sg;
269 loff_t pos = (cmd->t_task_lba * se_dev->dev_attrib.block_size);
272 iov = kzalloc(sizeof(struct iovec) * sgl_nents, GFP_KERNEL);
274 pr_err("Unable to allocate fd_do_readv iov[]\n");
278 for_each_sg(sgl, sg, sgl_nents, i) {
279 iov[i].iov_len = sg->length;
280 iov[i].iov_base = kmap(sg_page(sg)) + sg->offset;
287 ret = vfs_writev(fd, &iov[0], sgl_nents, &pos);
289 ret = vfs_readv(fd, &iov[0], sgl_nents, &pos);
293 for_each_sg(sgl, sg, sgl_nents, i)
299 if (ret < 0 || ret != cmd->data_length) {
300 pr_err("%s() write returned %d\n", __func__, ret);
301 return (ret < 0 ? ret : -EINVAL);
305 * Return zeros and GOOD status even if the READ did not return
306 * the expected virt_size for struct file w/o a backing struct
309 if (S_ISBLK(file_inode(fd)->i_mode)) {
310 if (ret < 0 || ret != cmd->data_length) {
311 pr_err("%s() returned %d, expecting %u for "
312 "S_ISBLK\n", __func__, ret,
314 return (ret < 0 ? ret : -EINVAL);
318 pr_err("%s() returned %d for non S_ISBLK\n",
327 static sense_reason_t
328 fd_execute_sync_cache(struct se_cmd *cmd)
330 struct se_device *dev = cmd->se_dev;
331 struct fd_dev *fd_dev = FD_DEV(dev);
332 int immed = (cmd->t_task_cdb[1] & 0x2);
337 * If the Immediate bit is set, queue up the GOOD response
338 * for this SYNCHRONIZE_CACHE op
341 target_complete_cmd(cmd, SAM_STAT_GOOD);
344 * Determine if we will be flushing the entire device.
346 if (cmd->t_task_lba == 0 && cmd->data_length == 0) {
350 start = cmd->t_task_lba * dev->dev_attrib.block_size;
351 if (cmd->data_length)
352 end = start + cmd->data_length;
357 ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
359 pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
365 target_complete_cmd(cmd, SAM_STAT_CHECK_CONDITION);
367 target_complete_cmd(cmd, SAM_STAT_GOOD);
372 static unsigned char *
373 fd_setup_write_same_buf(struct se_cmd *cmd, struct scatterlist *sg,
376 struct se_device *se_dev = cmd->se_dev;
377 unsigned int block_size = se_dev->dev_attrib.block_size;
378 unsigned int i = 0, end;
379 unsigned char *buf, *p, *kmap_buf;
381 buf = kzalloc(min_t(unsigned int, len, PAGE_SIZE), GFP_KERNEL);
383 pr_err("Unable to allocate fd_execute_write_same buf\n");
387 kmap_buf = kmap(sg_page(sg)) + sg->offset;
389 pr_err("kmap() failed in fd_setup_write_same\n");
394 * Fill local *buf to contain multiple WRITE_SAME blocks up to
395 * min(len, PAGE_SIZE)
398 end = min_t(unsigned int, len, PAGE_SIZE);
401 memcpy(p, kmap_buf, block_size);
411 static sense_reason_t
412 fd_execute_write_same(struct se_cmd *cmd)
414 struct se_device *se_dev = cmd->se_dev;
415 struct fd_dev *fd_dev = FD_DEV(se_dev);
416 struct file *f = fd_dev->fd_file;
417 struct scatterlist *sg;
420 sector_t nolb = sbc_get_write_same_sectors(cmd);
421 loff_t pos = cmd->t_task_lba * se_dev->dev_attrib.block_size;
422 unsigned int len, len_tmp, iov_num;
427 target_complete_cmd(cmd, SAM_STAT_GOOD);
430 sg = &cmd->t_data_sg[0];
432 if (cmd->t_data_nents > 1 ||
433 sg->length != cmd->se_dev->dev_attrib.block_size) {
434 pr_err("WRITE_SAME: Illegal SGL t_data_nents: %u length: %u"
435 " block_size: %u\n", cmd->t_data_nents, sg->length,
436 cmd->se_dev->dev_attrib.block_size);
437 return TCM_INVALID_CDB_FIELD;
440 len = len_tmp = nolb * se_dev->dev_attrib.block_size;
441 iov_num = DIV_ROUND_UP(len, PAGE_SIZE);
443 buf = fd_setup_write_same_buf(cmd, sg, len);
445 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
447 iov = vzalloc(sizeof(struct iovec) * iov_num);
449 pr_err("Unable to allocate fd_execute_write_same iovecs\n");
451 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
454 * Map the single fabric received scatterlist block now populated
455 * in *buf into each iovec for I/O submission.
457 for (i = 0; i < iov_num; i++) {
458 iov[i].iov_base = buf;
459 iov[i].iov_len = min_t(unsigned int, len_tmp, PAGE_SIZE);
460 len_tmp -= iov[i].iov_len;
465 rc = vfs_writev(f, &iov[0], iov_num, &pos);
471 if (rc < 0 || rc != len) {
472 pr_err("vfs_writev() returned %d for write same\n", rc);
473 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
476 target_complete_cmd(cmd, SAM_STAT_GOOD);
480 static sense_reason_t
481 fd_do_unmap(struct se_cmd *cmd, void *priv, sector_t lba, sector_t nolb)
483 struct file *file = priv;
484 struct inode *inode = file->f_mapping->host;
487 if (S_ISBLK(inode->i_mode)) {
488 /* The backend is block device, use discard */
489 struct block_device *bdev = inode->i_bdev;
491 ret = blkdev_issue_discard(bdev, lba,
492 nolb, GFP_KERNEL, 0);
494 pr_warn("FILEIO: blkdev_issue_discard() failed: %d\n",
496 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
499 /* The backend is normal file, use fallocate */
500 struct se_device *se_dev = cmd->se_dev;
501 loff_t pos = lba * se_dev->dev_attrib.block_size;
502 unsigned int len = nolb * se_dev->dev_attrib.block_size;
503 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
505 if (!file->f_op->fallocate)
506 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
508 ret = file->f_op->fallocate(file, mode, pos, len);
510 pr_warn("FILEIO: fallocate() failed: %d\n", ret);
511 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
518 static sense_reason_t
519 fd_execute_write_same_unmap(struct se_cmd *cmd)
521 struct se_device *se_dev = cmd->se_dev;
522 struct fd_dev *fd_dev = FD_DEV(se_dev);
523 struct file *file = fd_dev->fd_file;
524 sector_t lba = cmd->t_task_lba;
525 sector_t nolb = sbc_get_write_same_sectors(cmd);
529 target_complete_cmd(cmd, SAM_STAT_GOOD);
533 ret = fd_do_unmap(cmd, file, lba, nolb);
537 target_complete_cmd(cmd, GOOD);
541 static sense_reason_t
542 fd_execute_unmap(struct se_cmd *cmd)
544 struct file *file = FD_DEV(cmd->se_dev)->fd_file;
546 return sbc_execute_unmap(cmd, fd_do_unmap, file);
549 static sense_reason_t
550 fd_execute_rw(struct se_cmd *cmd)
552 struct scatterlist *sgl = cmd->t_data_sg;
553 u32 sgl_nents = cmd->t_data_nents;
554 enum dma_data_direction data_direction = cmd->data_direction;
555 struct se_device *dev = cmd->se_dev;
558 * We are currently limited by the number of iovecs (2048) per
559 * single vfs_[writev,readv] call.
561 if (cmd->data_length > FD_MAX_BYTES) {
562 pr_err("FILEIO: Not able to process I/O of %u bytes due to"
563 "FD_MAX_BYTES: %u iovec count limitiation\n",
564 cmd->data_length, FD_MAX_BYTES);
565 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
568 * Call vectorized fileio functions to map struct scatterlist
569 * physical memory addresses to struct iovec virtual memory.
571 if (data_direction == DMA_FROM_DEVICE) {
572 ret = fd_do_rw(cmd, sgl, sgl_nents, 0);
574 ret = fd_do_rw(cmd, sgl, sgl_nents, 1);
576 * Perform implict vfs_fsync_range() for fd_do_writev() ops
577 * for SCSI WRITEs with Forced Unit Access (FUA) set.
578 * Allow this to happen independent of WCE=0 setting.
581 dev->dev_attrib.emulate_fua_write > 0 &&
582 (cmd->se_cmd_flags & SCF_FUA)) {
583 struct fd_dev *fd_dev = FD_DEV(dev);
584 loff_t start = cmd->t_task_lba *
585 dev->dev_attrib.block_size;
586 loff_t end = start + cmd->data_length;
588 vfs_fsync_range(fd_dev->fd_file, start, end, 1);
593 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
596 target_complete_cmd(cmd, SAM_STAT_GOOD);
601 Opt_fd_dev_name, Opt_fd_dev_size, Opt_fd_buffered_io, Opt_err
604 static match_table_t tokens = {
605 {Opt_fd_dev_name, "fd_dev_name=%s"},
606 {Opt_fd_dev_size, "fd_dev_size=%s"},
607 {Opt_fd_buffered_io, "fd_buffered_io=%d"},
611 static ssize_t fd_set_configfs_dev_params(struct se_device *dev,
612 const char *page, ssize_t count)
614 struct fd_dev *fd_dev = FD_DEV(dev);
615 char *orig, *ptr, *arg_p, *opts;
616 substring_t args[MAX_OPT_ARGS];
617 int ret = 0, arg, token;
619 opts = kstrdup(page, GFP_KERNEL);
625 while ((ptr = strsep(&opts, ",\n")) != NULL) {
629 token = match_token(ptr, tokens, args);
631 case Opt_fd_dev_name:
632 if (match_strlcpy(fd_dev->fd_dev_name, &args[0],
633 FD_MAX_DEV_NAME) == 0) {
637 pr_debug("FILEIO: Referencing Path: %s\n",
638 fd_dev->fd_dev_name);
639 fd_dev->fbd_flags |= FBDF_HAS_PATH;
641 case Opt_fd_dev_size:
642 arg_p = match_strdup(&args[0]);
647 ret = strict_strtoull(arg_p, 0, &fd_dev->fd_dev_size);
650 pr_err("strict_strtoull() failed for"
654 pr_debug("FILEIO: Referencing Size: %llu"
655 " bytes\n", fd_dev->fd_dev_size);
656 fd_dev->fbd_flags |= FBDF_HAS_SIZE;
658 case Opt_fd_buffered_io:
659 match_int(args, &arg);
661 pr_err("bogus fd_buffered_io=%d value\n", arg);
666 pr_debug("FILEIO: Using buffered I/O"
667 " operations for struct fd_dev\n");
669 fd_dev->fbd_flags |= FDBD_HAS_BUFFERED_IO_WCE;
678 return (!ret) ? count : ret;
681 static ssize_t fd_show_configfs_dev_params(struct se_device *dev, char *b)
683 struct fd_dev *fd_dev = FD_DEV(dev);
686 bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
687 bl += sprintf(b + bl, " File: %s Size: %llu Mode: %s\n",
688 fd_dev->fd_dev_name, fd_dev->fd_dev_size,
689 (fd_dev->fbd_flags & FDBD_HAS_BUFFERED_IO_WCE) ?
690 "Buffered-WCE" : "O_DSYNC");
694 static sector_t fd_get_blocks(struct se_device *dev)
696 struct fd_dev *fd_dev = FD_DEV(dev);
697 struct file *f = fd_dev->fd_file;
698 struct inode *i = f->f_mapping->host;
699 unsigned long long dev_size;
701 * When using a file that references an underlying struct block_device,
702 * ensure dev_size is always based on the current inode size in order
703 * to handle underlying block_device resize operations.
705 if (S_ISBLK(i->i_mode))
706 dev_size = i_size_read(i);
708 dev_size = fd_dev->fd_dev_size;
710 return div_u64(dev_size - dev->dev_attrib.block_size,
711 dev->dev_attrib.block_size);
714 static struct sbc_ops fd_sbc_ops = {
715 .execute_rw = fd_execute_rw,
716 .execute_sync_cache = fd_execute_sync_cache,
717 .execute_write_same = fd_execute_write_same,
718 .execute_write_same_unmap = fd_execute_write_same_unmap,
719 .execute_unmap = fd_execute_unmap,
722 static sense_reason_t
723 fd_parse_cdb(struct se_cmd *cmd)
725 return sbc_parse_cdb(cmd, &fd_sbc_ops);
728 static struct se_subsystem_api fileio_template = {
730 .inquiry_prod = "FILEIO",
731 .inquiry_rev = FD_VERSION,
732 .owner = THIS_MODULE,
733 .transport_type = TRANSPORT_PLUGIN_VHBA_PDEV,
734 .attach_hba = fd_attach_hba,
735 .detach_hba = fd_detach_hba,
736 .alloc_device = fd_alloc_device,
737 .configure_device = fd_configure_device,
738 .free_device = fd_free_device,
739 .parse_cdb = fd_parse_cdb,
740 .set_configfs_dev_params = fd_set_configfs_dev_params,
741 .show_configfs_dev_params = fd_show_configfs_dev_params,
742 .get_device_type = sbc_get_device_type,
743 .get_blocks = fd_get_blocks,
746 static int __init fileio_module_init(void)
748 return transport_subsystem_register(&fileio_template);
751 static void __exit fileio_module_exit(void)
753 transport_subsystem_release(&fileio_template);
756 MODULE_DESCRIPTION("TCM FILEIO subsystem plugin");
757 MODULE_AUTHOR("nab@Linux-iSCSI.org");
758 MODULE_LICENSE("GPL");
760 module_init(fileio_module_init);
761 module_exit(fileio_module_exit);