Loading...
Note: File does not exist in v3.1.
1/*
2 * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
3 * Copyright (C) 2014 Red Hat, Inc.
4 * Copyright (C) 2015 Arrikto, Inc.
5 * Copyright (C) 2017 Chinamobile, Inc.
6 *
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms and conditions of the GNU General Public License,
9 * version 2, as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * more details.
15 *
16 * You should have received a copy of the GNU General Public License along with
17 * this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 */
20
21#include <linux/spinlock.h>
22#include <linux/module.h>
23#include <linux/idr.h>
24#include <linux/kernel.h>
25#include <linux/timer.h>
26#include <linux/parser.h>
27#include <linux/vmalloc.h>
28#include <linux/uio_driver.h>
29#include <linux/radix-tree.h>
30#include <linux/stringify.h>
31#include <linux/bitops.h>
32#include <linux/highmem.h>
33#include <linux/configfs.h>
34#include <linux/mutex.h>
35#include <linux/workqueue.h>
36#include <net/genetlink.h>
37#include <scsi/scsi_common.h>
38#include <scsi/scsi_proto.h>
39#include <target/target_core_base.h>
40#include <target/target_core_fabric.h>
41#include <target/target_core_backend.h>
42
43#include <linux/target_core_user.h>
44
45/*
46 * Define a shared-memory interface for LIO to pass SCSI commands and
47 * data to userspace for processing. This is to allow backends that
48 * are too complex for in-kernel support to be possible.
49 *
50 * It uses the UIO framework to do a lot of the device-creation and
51 * introspection work for us.
52 *
53 * See the .h file for how the ring is laid out. Note that while the
54 * command ring is defined, the particulars of the data area are
55 * not. Offset values in the command entry point to other locations
56 * internal to the mmap()ed area. There is separate space outside the
57 * command ring for data buffers. This leaves maximum flexibility for
58 * moving buffer allocations, or even page flipping or other
59 * allocation techniques, without altering the command ring layout.
60 *
61 * SECURITY:
62 * The user process must be assumed to be malicious. There's no way to
63 * prevent it breaking the command ring protocol if it wants, but in
64 * order to prevent other issues we must only ever read *data* from
65 * the shared memory area, not offsets or sizes. This applies to
66 * command ring entries as well as the mailbox. Extra code needed for
67 * this may have a 'UAM' comment.
68 */
69
70#define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
71
72/* For cmd area, the size is fixed 8MB */
73#define CMDR_SIZE (8 * 1024 * 1024)
74
75/*
76 * For data area, the block size is PAGE_SIZE and
77 * the total size is 256K * PAGE_SIZE.
78 */
79#define DATA_BLOCK_SIZE PAGE_SIZE
80#define DATA_BLOCK_SHIFT PAGE_SHIFT
81#define DATA_BLOCK_BITS_DEF (256 * 1024)
82#define DATA_SIZE (DATA_BLOCK_BITS * DATA_BLOCK_SIZE)
83
84#define TCMU_MBS_TO_BLOCKS(_mbs) (_mbs << (20 - DATA_BLOCK_SHIFT))
85#define TCMU_BLOCKS_TO_MBS(_blocks) (_blocks >> (20 - DATA_BLOCK_SHIFT))
86
87/* The total size of the ring is 8M + 256K * PAGE_SIZE */
88#define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
89
90/*
91 * Default number of global data blocks(512K * PAGE_SIZE)
92 * when the unmap thread will be started.
93 */
94#define TCMU_GLOBAL_MAX_BLOCKS_DEF (512 * 1024)
95
96static u8 tcmu_kern_cmd_reply_supported;
97
98static struct device *tcmu_root_device;
99
100struct tcmu_hba {
101 u32 host_id;
102};
103
104#define TCMU_CONFIG_LEN 256
105
106struct tcmu_nl_cmd {
107 /* wake up thread waiting for reply */
108 struct completion complete;
109 int cmd;
110 int status;
111};
112
113struct tcmu_dev {
114 struct list_head node;
115 struct kref kref;
116
117 struct se_device se_dev;
118
119 char *name;
120 struct se_hba *hba;
121
122#define TCMU_DEV_BIT_OPEN 0
123#define TCMU_DEV_BIT_BROKEN 1
124#define TCMU_DEV_BIT_BLOCKED 2
125 unsigned long flags;
126
127 struct uio_info uio_info;
128
129 struct inode *inode;
130
131 struct tcmu_mailbox *mb_addr;
132 size_t dev_size;
133 u32 cmdr_size;
134 u32 cmdr_last_cleaned;
135 /* Offset of data area from start of mb */
136 /* Must add data_off and mb_addr to get the address */
137 size_t data_off;
138 size_t data_size;
139 uint32_t max_blocks;
140 size_t ring_size;
141
142 struct mutex cmdr_lock;
143 struct list_head cmdr_queue;
144
145 uint32_t dbi_max;
146 uint32_t dbi_thresh;
147 unsigned long *data_bitmap;
148 struct radix_tree_root data_blocks;
149
150 struct idr commands;
151
152 struct timer_list cmd_timer;
153 unsigned int cmd_time_out;
154
155 struct timer_list qfull_timer;
156 int qfull_time_out;
157
158 struct list_head timedout_entry;
159
160 spinlock_t nl_cmd_lock;
161 struct tcmu_nl_cmd curr_nl_cmd;
162 /* wake up threads waiting on curr_nl_cmd */
163 wait_queue_head_t nl_cmd_wq;
164
165 char dev_config[TCMU_CONFIG_LEN];
166
167 int nl_reply_supported;
168};
169
170#define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
171
172#define CMDR_OFF sizeof(struct tcmu_mailbox)
173
174struct tcmu_cmd {
175 struct se_cmd *se_cmd;
176 struct tcmu_dev *tcmu_dev;
177 struct list_head cmdr_queue_entry;
178
179 uint16_t cmd_id;
180
181 /* Can't use se_cmd when cleaning up expired cmds, because if
182 cmd has been completed then accessing se_cmd is off limits */
183 uint32_t dbi_cnt;
184 uint32_t dbi_cur;
185 uint32_t *dbi;
186
187 unsigned long deadline;
188
189#define TCMU_CMD_BIT_EXPIRED 0
190 unsigned long flags;
191};
192/*
193 * To avoid dead lock the mutex lock order should always be:
194 *
195 * mutex_lock(&root_udev_mutex);
196 * ...
197 * mutex_lock(&tcmu_dev->cmdr_lock);
198 * mutex_unlock(&tcmu_dev->cmdr_lock);
199 * ...
200 * mutex_unlock(&root_udev_mutex);
201 */
202static DEFINE_MUTEX(root_udev_mutex);
203static LIST_HEAD(root_udev);
204
205static DEFINE_SPINLOCK(timed_out_udevs_lock);
206static LIST_HEAD(timed_out_udevs);
207
208static struct kmem_cache *tcmu_cmd_cache;
209
210static atomic_t global_db_count = ATOMIC_INIT(0);
211static struct delayed_work tcmu_unmap_work;
212static int tcmu_global_max_blocks = TCMU_GLOBAL_MAX_BLOCKS_DEF;
213
214static int tcmu_set_global_max_data_area(const char *str,
215 const struct kernel_param *kp)
216{
217 int ret, max_area_mb;
218
219 ret = kstrtoint(str, 10, &max_area_mb);
220 if (ret)
221 return -EINVAL;
222
223 if (max_area_mb <= 0) {
224 pr_err("global_max_data_area must be larger than 0.\n");
225 return -EINVAL;
226 }
227
228 tcmu_global_max_blocks = TCMU_MBS_TO_BLOCKS(max_area_mb);
229 if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
230 schedule_delayed_work(&tcmu_unmap_work, 0);
231 else
232 cancel_delayed_work_sync(&tcmu_unmap_work);
233
234 return 0;
235}
236
237static int tcmu_get_global_max_data_area(char *buffer,
238 const struct kernel_param *kp)
239{
240 return sprintf(buffer, "%d", TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
241}
242
243static const struct kernel_param_ops tcmu_global_max_data_area_op = {
244 .set = tcmu_set_global_max_data_area,
245 .get = tcmu_get_global_max_data_area,
246};
247
248module_param_cb(global_max_data_area_mb, &tcmu_global_max_data_area_op, NULL,
249 S_IWUSR | S_IRUGO);
250MODULE_PARM_DESC(global_max_data_area_mb,
251 "Max MBs allowed to be allocated to all the tcmu device's "
252 "data areas.");
253
254/* multicast group */
255enum tcmu_multicast_groups {
256 TCMU_MCGRP_CONFIG,
257};
258
259static const struct genl_multicast_group tcmu_mcgrps[] = {
260 [TCMU_MCGRP_CONFIG] = { .name = "config", },
261};
262
263static struct nla_policy tcmu_attr_policy[TCMU_ATTR_MAX+1] = {
264 [TCMU_ATTR_DEVICE] = { .type = NLA_STRING },
265 [TCMU_ATTR_MINOR] = { .type = NLA_U32 },
266 [TCMU_ATTR_CMD_STATUS] = { .type = NLA_S32 },
267 [TCMU_ATTR_DEVICE_ID] = { .type = NLA_U32 },
268 [TCMU_ATTR_SUPP_KERN_CMD_REPLY] = { .type = NLA_U8 },
269};
270
271static int tcmu_genl_cmd_done(struct genl_info *info, int completed_cmd)
272{
273 struct se_device *dev;
274 struct tcmu_dev *udev;
275 struct tcmu_nl_cmd *nl_cmd;
276 int dev_id, rc, ret = 0;
277 bool is_removed = (completed_cmd == TCMU_CMD_REMOVED_DEVICE);
278
279 if (!info->attrs[TCMU_ATTR_CMD_STATUS] ||
280 !info->attrs[TCMU_ATTR_DEVICE_ID]) {
281 printk(KERN_ERR "TCMU_ATTR_CMD_STATUS or TCMU_ATTR_DEVICE_ID not set, doing nothing\n");
282 return -EINVAL;
283 }
284
285 dev_id = nla_get_u32(info->attrs[TCMU_ATTR_DEVICE_ID]);
286 rc = nla_get_s32(info->attrs[TCMU_ATTR_CMD_STATUS]);
287
288 dev = target_find_device(dev_id, !is_removed);
289 if (!dev) {
290 printk(KERN_ERR "tcmu nl cmd %u/%u completion could not find device with dev id %u.\n",
291 completed_cmd, rc, dev_id);
292 return -ENODEV;
293 }
294 udev = TCMU_DEV(dev);
295
296 spin_lock(&udev->nl_cmd_lock);
297 nl_cmd = &udev->curr_nl_cmd;
298
299 pr_debug("genl cmd done got id %d curr %d done %d rc %d\n", dev_id,
300 nl_cmd->cmd, completed_cmd, rc);
301
302 if (nl_cmd->cmd != completed_cmd) {
303 printk(KERN_ERR "Mismatched commands (Expecting reply for %d. Current %d).\n",
304 completed_cmd, nl_cmd->cmd);
305 ret = -EINVAL;
306 } else {
307 nl_cmd->status = rc;
308 }
309
310 spin_unlock(&udev->nl_cmd_lock);
311 if (!is_removed)
312 target_undepend_item(&dev->dev_group.cg_item);
313 if (!ret)
314 complete(&nl_cmd->complete);
315 return ret;
316}
317
318static int tcmu_genl_rm_dev_done(struct sk_buff *skb, struct genl_info *info)
319{
320 return tcmu_genl_cmd_done(info, TCMU_CMD_REMOVED_DEVICE);
321}
322
323static int tcmu_genl_add_dev_done(struct sk_buff *skb, struct genl_info *info)
324{
325 return tcmu_genl_cmd_done(info, TCMU_CMD_ADDED_DEVICE);
326}
327
328static int tcmu_genl_reconfig_dev_done(struct sk_buff *skb,
329 struct genl_info *info)
330{
331 return tcmu_genl_cmd_done(info, TCMU_CMD_RECONFIG_DEVICE);
332}
333
334static int tcmu_genl_set_features(struct sk_buff *skb, struct genl_info *info)
335{
336 if (info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]) {
337 tcmu_kern_cmd_reply_supported =
338 nla_get_u8(info->attrs[TCMU_ATTR_SUPP_KERN_CMD_REPLY]);
339 printk(KERN_INFO "tcmu daemon: command reply support %u.\n",
340 tcmu_kern_cmd_reply_supported);
341 }
342
343 return 0;
344}
345
346static const struct genl_ops tcmu_genl_ops[] = {
347 {
348 .cmd = TCMU_CMD_SET_FEATURES,
349 .flags = GENL_ADMIN_PERM,
350 .policy = tcmu_attr_policy,
351 .doit = tcmu_genl_set_features,
352 },
353 {
354 .cmd = TCMU_CMD_ADDED_DEVICE_DONE,
355 .flags = GENL_ADMIN_PERM,
356 .policy = tcmu_attr_policy,
357 .doit = tcmu_genl_add_dev_done,
358 },
359 {
360 .cmd = TCMU_CMD_REMOVED_DEVICE_DONE,
361 .flags = GENL_ADMIN_PERM,
362 .policy = tcmu_attr_policy,
363 .doit = tcmu_genl_rm_dev_done,
364 },
365 {
366 .cmd = TCMU_CMD_RECONFIG_DEVICE_DONE,
367 .flags = GENL_ADMIN_PERM,
368 .policy = tcmu_attr_policy,
369 .doit = tcmu_genl_reconfig_dev_done,
370 },
371};
372
373/* Our generic netlink family */
374static struct genl_family tcmu_genl_family __ro_after_init = {
375 .module = THIS_MODULE,
376 .hdrsize = 0,
377 .name = "TCM-USER",
378 .version = 2,
379 .maxattr = TCMU_ATTR_MAX,
380 .mcgrps = tcmu_mcgrps,
381 .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
382 .netnsok = true,
383 .ops = tcmu_genl_ops,
384 .n_ops = ARRAY_SIZE(tcmu_genl_ops),
385};
386
387#define tcmu_cmd_set_dbi_cur(cmd, index) ((cmd)->dbi_cur = (index))
388#define tcmu_cmd_reset_dbi_cur(cmd) tcmu_cmd_set_dbi_cur(cmd, 0)
389#define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
390#define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
391
392static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
393{
394 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
395 uint32_t i;
396
397 for (i = 0; i < len; i++)
398 clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
399}
400
401static inline bool tcmu_get_empty_block(struct tcmu_dev *udev,
402 struct tcmu_cmd *tcmu_cmd)
403{
404 struct page *page;
405 int ret, dbi;
406
407 dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
408 if (dbi == udev->dbi_thresh)
409 return false;
410
411 page = radix_tree_lookup(&udev->data_blocks, dbi);
412 if (!page) {
413 if (atomic_add_return(1, &global_db_count) >
414 tcmu_global_max_blocks)
415 schedule_delayed_work(&tcmu_unmap_work, 0);
416
417 /* try to get new page from the mm */
418 page = alloc_page(GFP_KERNEL);
419 if (!page)
420 goto err_alloc;
421
422 ret = radix_tree_insert(&udev->data_blocks, dbi, page);
423 if (ret)
424 goto err_insert;
425 }
426
427 if (dbi > udev->dbi_max)
428 udev->dbi_max = dbi;
429
430 set_bit(dbi, udev->data_bitmap);
431 tcmu_cmd_set_dbi(tcmu_cmd, dbi);
432
433 return true;
434err_insert:
435 __free_page(page);
436err_alloc:
437 atomic_dec(&global_db_count);
438 return false;
439}
440
441static bool tcmu_get_empty_blocks(struct tcmu_dev *udev,
442 struct tcmu_cmd *tcmu_cmd)
443{
444 int i;
445
446 for (i = tcmu_cmd->dbi_cur; i < tcmu_cmd->dbi_cnt; i++) {
447 if (!tcmu_get_empty_block(udev, tcmu_cmd))
448 return false;
449 }
450 return true;
451}
452
453static inline struct page *
454tcmu_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
455{
456 return radix_tree_lookup(&udev->data_blocks, dbi);
457}
458
459static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
460{
461 kfree(tcmu_cmd->dbi);
462 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
463}
464
465static inline size_t tcmu_cmd_get_data_length(struct tcmu_cmd *tcmu_cmd)
466{
467 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
468 size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
469
470 if (se_cmd->se_cmd_flags & SCF_BIDI) {
471 BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
472 data_length += round_up(se_cmd->t_bidi_data_sg->length,
473 DATA_BLOCK_SIZE);
474 }
475
476 return data_length;
477}
478
479static inline uint32_t tcmu_cmd_get_block_cnt(struct tcmu_cmd *tcmu_cmd)
480{
481 size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
482
483 return data_length / DATA_BLOCK_SIZE;
484}
485
486static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
487{
488 struct se_device *se_dev = se_cmd->se_dev;
489 struct tcmu_dev *udev = TCMU_DEV(se_dev);
490 struct tcmu_cmd *tcmu_cmd;
491
492 tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_KERNEL);
493 if (!tcmu_cmd)
494 return NULL;
495
496 INIT_LIST_HEAD(&tcmu_cmd->cmdr_queue_entry);
497 tcmu_cmd->se_cmd = se_cmd;
498 tcmu_cmd->tcmu_dev = udev;
499
500 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
501 tcmu_cmd->dbi_cnt = tcmu_cmd_get_block_cnt(tcmu_cmd);
502 tcmu_cmd->dbi = kcalloc(tcmu_cmd->dbi_cnt, sizeof(uint32_t),
503 GFP_KERNEL);
504 if (!tcmu_cmd->dbi) {
505 kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
506 return NULL;
507 }
508
509 return tcmu_cmd;
510}
511
512static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
513{
514 unsigned long offset = offset_in_page(vaddr);
515 void *start = vaddr - offset;
516
517 size = round_up(size+offset, PAGE_SIZE);
518
519 while (size) {
520 flush_dcache_page(virt_to_page(start));
521 start += PAGE_SIZE;
522 size -= PAGE_SIZE;
523 }
524}
525
526/*
527 * Some ring helper functions. We don't assume size is a power of 2 so
528 * we can't use circ_buf.h.
529 */
530static inline size_t spc_used(size_t head, size_t tail, size_t size)
531{
532 int diff = head - tail;
533
534 if (diff >= 0)
535 return diff;
536 else
537 return size + diff;
538}
539
540static inline size_t spc_free(size_t head, size_t tail, size_t size)
541{
542 /* Keep 1 byte unused or we can't tell full from empty */
543 return (size - spc_used(head, tail, size) - 1);
544}
545
546static inline size_t head_to_end(size_t head, size_t size)
547{
548 return size - head;
549}
550
551static inline void new_iov(struct iovec **iov, int *iov_cnt)
552{
553 struct iovec *iovec;
554
555 if (*iov_cnt != 0)
556 (*iov)++;
557 (*iov_cnt)++;
558
559 iovec = *iov;
560 memset(iovec, 0, sizeof(struct iovec));
561}
562
563#define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
564
565/* offset is relative to mb_addr */
566static inline size_t get_block_offset_user(struct tcmu_dev *dev,
567 int dbi, int remaining)
568{
569 return dev->data_off + dbi * DATA_BLOCK_SIZE +
570 DATA_BLOCK_SIZE - remaining;
571}
572
573static inline size_t iov_tail(struct iovec *iov)
574{
575 return (size_t)iov->iov_base + iov->iov_len;
576}
577
578static void scatter_data_area(struct tcmu_dev *udev,
579 struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
580 unsigned int data_nents, struct iovec **iov,
581 int *iov_cnt, bool copy_data)
582{
583 int i, dbi;
584 int block_remaining = 0;
585 void *from, *to = NULL;
586 size_t copy_bytes, to_offset, offset;
587 struct scatterlist *sg;
588 struct page *page;
589
590 for_each_sg(data_sg, sg, data_nents, i) {
591 int sg_remaining = sg->length;
592 from = kmap_atomic(sg_page(sg)) + sg->offset;
593 while (sg_remaining > 0) {
594 if (block_remaining == 0) {
595 if (to)
596 kunmap_atomic(to);
597
598 block_remaining = DATA_BLOCK_SIZE;
599 dbi = tcmu_cmd_get_dbi(tcmu_cmd);
600 page = tcmu_get_block_page(udev, dbi);
601 to = kmap_atomic(page);
602 }
603
604 /*
605 * Covert to virtual offset of the ring data area.
606 */
607 to_offset = get_block_offset_user(udev, dbi,
608 block_remaining);
609
610 /*
611 * The following code will gather and map the blocks
612 * to the same iovec when the blocks are all next to
613 * each other.
614 */
615 copy_bytes = min_t(size_t, sg_remaining,
616 block_remaining);
617 if (*iov_cnt != 0 &&
618 to_offset == iov_tail(*iov)) {
619 /*
620 * Will append to the current iovec, because
621 * the current block page is next to the
622 * previous one.
623 */
624 (*iov)->iov_len += copy_bytes;
625 } else {
626 /*
627 * Will allocate a new iovec because we are
628 * first time here or the current block page
629 * is not next to the previous one.
630 */
631 new_iov(iov, iov_cnt);
632 (*iov)->iov_base = (void __user *)to_offset;
633 (*iov)->iov_len = copy_bytes;
634 }
635
636 if (copy_data) {
637 offset = DATA_BLOCK_SIZE - block_remaining;
638 memcpy(to + offset,
639 from + sg->length - sg_remaining,
640 copy_bytes);
641 tcmu_flush_dcache_range(to, copy_bytes);
642 }
643
644 sg_remaining -= copy_bytes;
645 block_remaining -= copy_bytes;
646 }
647 kunmap_atomic(from - sg->offset);
648 }
649
650 if (to)
651 kunmap_atomic(to);
652}
653
654static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
655 bool bidi)
656{
657 struct se_cmd *se_cmd = cmd->se_cmd;
658 int i, dbi;
659 int block_remaining = 0;
660 void *from = NULL, *to;
661 size_t copy_bytes, offset;
662 struct scatterlist *sg, *data_sg;
663 struct page *page;
664 unsigned int data_nents;
665 uint32_t count = 0;
666
667 if (!bidi) {
668 data_sg = se_cmd->t_data_sg;
669 data_nents = se_cmd->t_data_nents;
670 } else {
671
672 /*
673 * For bidi case, the first count blocks are for Data-Out
674 * buffer blocks, and before gathering the Data-In buffer
675 * the Data-Out buffer blocks should be discarded.
676 */
677 count = DIV_ROUND_UP(se_cmd->data_length, DATA_BLOCK_SIZE);
678
679 data_sg = se_cmd->t_bidi_data_sg;
680 data_nents = se_cmd->t_bidi_data_nents;
681 }
682
683 tcmu_cmd_set_dbi_cur(cmd, count);
684
685 for_each_sg(data_sg, sg, data_nents, i) {
686 int sg_remaining = sg->length;
687 to = kmap_atomic(sg_page(sg)) + sg->offset;
688 while (sg_remaining > 0) {
689 if (block_remaining == 0) {
690 if (from)
691 kunmap_atomic(from);
692
693 block_remaining = DATA_BLOCK_SIZE;
694 dbi = tcmu_cmd_get_dbi(cmd);
695 page = tcmu_get_block_page(udev, dbi);
696 from = kmap_atomic(page);
697 }
698 copy_bytes = min_t(size_t, sg_remaining,
699 block_remaining);
700 offset = DATA_BLOCK_SIZE - block_remaining;
701 tcmu_flush_dcache_range(from, copy_bytes);
702 memcpy(to + sg->length - sg_remaining, from + offset,
703 copy_bytes);
704
705 sg_remaining -= copy_bytes;
706 block_remaining -= copy_bytes;
707 }
708 kunmap_atomic(to - sg->offset);
709 }
710 if (from)
711 kunmap_atomic(from);
712}
713
714static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
715{
716 return thresh - bitmap_weight(bitmap, thresh);
717}
718
719/*
720 * We can't queue a command until we have space available on the cmd ring *and*
721 * space available on the data area.
722 *
723 * Called with ring lock held.
724 */
725static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
726 size_t cmd_size, size_t data_needed)
727{
728 struct tcmu_mailbox *mb = udev->mb_addr;
729 uint32_t blocks_needed = (data_needed + DATA_BLOCK_SIZE - 1)
730 / DATA_BLOCK_SIZE;
731 size_t space, cmd_needed;
732 u32 cmd_head;
733
734 tcmu_flush_dcache_range(mb, sizeof(*mb));
735
736 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
737
738 /*
739 * If cmd end-of-ring space is too small then we need space for a NOP plus
740 * original cmd - cmds are internally contiguous.
741 */
742 if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
743 cmd_needed = cmd_size;
744 else
745 cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
746
747 space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
748 if (space < cmd_needed) {
749 pr_debug("no cmd space: %u %u %u\n", cmd_head,
750 udev->cmdr_last_cleaned, udev->cmdr_size);
751 return false;
752 }
753
754 /* try to check and get the data blocks as needed */
755 space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
756 if ((space * DATA_BLOCK_SIZE) < data_needed) {
757 unsigned long blocks_left =
758 (udev->max_blocks - udev->dbi_thresh) + space;
759
760 if (blocks_left < blocks_needed) {
761 pr_debug("no data space: only %lu available, but ask for %zu\n",
762 blocks_left * DATA_BLOCK_SIZE,
763 data_needed);
764 return false;
765 }
766
767 udev->dbi_thresh += blocks_needed;
768 if (udev->dbi_thresh > udev->max_blocks)
769 udev->dbi_thresh = udev->max_blocks;
770 }
771
772 return tcmu_get_empty_blocks(udev, cmd);
773}
774
775static inline size_t tcmu_cmd_get_base_cmd_size(size_t iov_cnt)
776{
777 return max(offsetof(struct tcmu_cmd_entry, req.iov[iov_cnt]),
778 sizeof(struct tcmu_cmd_entry));
779}
780
781static inline size_t tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
782 size_t base_command_size)
783{
784 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
785 size_t command_size;
786
787 command_size = base_command_size +
788 round_up(scsi_command_size(se_cmd->t_task_cdb),
789 TCMU_OP_ALIGN_SIZE);
790
791 WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
792
793 return command_size;
794}
795
796static int tcmu_setup_cmd_timer(struct tcmu_cmd *tcmu_cmd, unsigned int tmo,
797 struct timer_list *timer)
798{
799 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
800 int cmd_id;
801
802 if (tcmu_cmd->cmd_id)
803 goto setup_timer;
804
805 cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 1, USHRT_MAX, GFP_NOWAIT);
806 if (cmd_id < 0) {
807 pr_err("tcmu: Could not allocate cmd id.\n");
808 return cmd_id;
809 }
810 tcmu_cmd->cmd_id = cmd_id;
811
812 pr_debug("allocated cmd %u for dev %s tmo %lu\n", tcmu_cmd->cmd_id,
813 udev->name, tmo / MSEC_PER_SEC);
814
815setup_timer:
816 if (!tmo)
817 return 0;
818
819 tcmu_cmd->deadline = round_jiffies_up(jiffies + msecs_to_jiffies(tmo));
820 mod_timer(timer, tcmu_cmd->deadline);
821 return 0;
822}
823
824static int add_to_cmdr_queue(struct tcmu_cmd *tcmu_cmd)
825{
826 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
827 unsigned int tmo;
828 int ret;
829
830 /*
831 * For backwards compat if qfull_time_out is not set use
832 * cmd_time_out and if that's not set use the default time out.
833 */
834 if (!udev->qfull_time_out)
835 return -ETIMEDOUT;
836 else if (udev->qfull_time_out > 0)
837 tmo = udev->qfull_time_out;
838 else if (udev->cmd_time_out)
839 tmo = udev->cmd_time_out;
840 else
841 tmo = TCMU_TIME_OUT;
842
843 ret = tcmu_setup_cmd_timer(tcmu_cmd, tmo, &udev->qfull_timer);
844 if (ret)
845 return ret;
846
847 list_add_tail(&tcmu_cmd->cmdr_queue_entry, &udev->cmdr_queue);
848 pr_debug("adding cmd %u on dev %s to ring space wait queue\n",
849 tcmu_cmd->cmd_id, udev->name);
850 return 0;
851}
852
853/**
854 * queue_cmd_ring - queue cmd to ring or internally
855 * @tcmu_cmd: cmd to queue
856 * @scsi_err: TCM error code if failure (-1) returned.
857 *
858 * Returns:
859 * -1 we cannot queue internally or to the ring.
860 * 0 success
861 * 1 internally queued to wait for ring memory to free.
862 */
863static sense_reason_t queue_cmd_ring(struct tcmu_cmd *tcmu_cmd, int *scsi_err)
864{
865 struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
866 struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
867 size_t base_command_size, command_size;
868 struct tcmu_mailbox *mb;
869 struct tcmu_cmd_entry *entry;
870 struct iovec *iov;
871 int iov_cnt, ret;
872 uint32_t cmd_head;
873 uint64_t cdb_off;
874 bool copy_to_data_area;
875 size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
876
877 *scsi_err = TCM_NO_SENSE;
878
879 if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags)) {
880 *scsi_err = TCM_LUN_BUSY;
881 return -1;
882 }
883
884 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
885 *scsi_err = TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
886 return -1;
887 }
888
889 /*
890 * Must be a certain minimum size for response sense info, but
891 * also may be larger if the iov array is large.
892 *
893 * We prepare as many iovs as possbile for potential uses here,
894 * because it's expensive to tell how many regions are freed in
895 * the bitmap & global data pool, as the size calculated here
896 * will only be used to do the checks.
897 *
898 * The size will be recalculated later as actually needed to save
899 * cmd area memories.
900 */
901 base_command_size = tcmu_cmd_get_base_cmd_size(tcmu_cmd->dbi_cnt);
902 command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
903
904 if (!list_empty(&udev->cmdr_queue))
905 goto queue;
906
907 mb = udev->mb_addr;
908 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
909 if ((command_size > (udev->cmdr_size / 2)) ||
910 data_length > udev->data_size) {
911 pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
912 "cmd ring/data area\n", command_size, data_length,
913 udev->cmdr_size, udev->data_size);
914 *scsi_err = TCM_INVALID_CDB_FIELD;
915 return -1;
916 }
917
918 if (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
919 /*
920 * Don't leave commands partially setup because the unmap
921 * thread might need the blocks to make forward progress.
922 */
923 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
924 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
925 goto queue;
926 }
927
928 /* Insert a PAD if end-of-ring space is too small */
929 if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
930 size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
931
932 entry = (void *) mb + CMDR_OFF + cmd_head;
933 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
934 tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
935 entry->hdr.cmd_id = 0; /* not used for PAD */
936 entry->hdr.kflags = 0;
937 entry->hdr.uflags = 0;
938 tcmu_flush_dcache_range(entry, sizeof(*entry));
939
940 UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
941 tcmu_flush_dcache_range(mb, sizeof(*mb));
942
943 cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
944 WARN_ON(cmd_head != 0);
945 }
946
947 entry = (void *) mb + CMDR_OFF + cmd_head;
948 memset(entry, 0, command_size);
949 tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
950
951 /* Handle allocating space from the data area */
952 tcmu_cmd_reset_dbi_cur(tcmu_cmd);
953 iov = &entry->req.iov[0];
954 iov_cnt = 0;
955 copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
956 || se_cmd->se_cmd_flags & SCF_BIDI);
957 scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
958 se_cmd->t_data_nents, &iov, &iov_cnt,
959 copy_to_data_area);
960 entry->req.iov_cnt = iov_cnt;
961
962 /* Handle BIDI commands */
963 iov_cnt = 0;
964 if (se_cmd->se_cmd_flags & SCF_BIDI) {
965 iov++;
966 scatter_data_area(udev, tcmu_cmd, se_cmd->t_bidi_data_sg,
967 se_cmd->t_bidi_data_nents, &iov, &iov_cnt,
968 false);
969 }
970 entry->req.iov_bidi_cnt = iov_cnt;
971
972 ret = tcmu_setup_cmd_timer(tcmu_cmd, udev->cmd_time_out,
973 &udev->cmd_timer);
974 if (ret) {
975 tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cnt);
976 mutex_unlock(&udev->cmdr_lock);
977
978 *scsi_err = TCM_OUT_OF_RESOURCES;
979 return -1;
980 }
981 entry->hdr.cmd_id = tcmu_cmd->cmd_id;
982
983 /*
984 * Recalaulate the command's base size and size according
985 * to the actual needs
986 */
987 base_command_size = tcmu_cmd_get_base_cmd_size(entry->req.iov_cnt +
988 entry->req.iov_bidi_cnt);
989 command_size = tcmu_cmd_get_cmd_size(tcmu_cmd, base_command_size);
990
991 tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
992
993 /* All offsets relative to mb_addr, not start of entry! */
994 cdb_off = CMDR_OFF + cmd_head + base_command_size;
995 memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
996 entry->req.cdb_off = cdb_off;
997 tcmu_flush_dcache_range(entry, sizeof(*entry));
998
999 UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
1000 tcmu_flush_dcache_range(mb, sizeof(*mb));
1001
1002 /* TODO: only if FLUSH and FUA? */
1003 uio_event_notify(&udev->uio_info);
1004
1005 return 0;
1006
1007queue:
1008 if (add_to_cmdr_queue(tcmu_cmd)) {
1009 *scsi_err = TCM_OUT_OF_RESOURCES;
1010 return -1;
1011 }
1012
1013 return 1;
1014}
1015
1016static sense_reason_t
1017tcmu_queue_cmd(struct se_cmd *se_cmd)
1018{
1019 struct se_device *se_dev = se_cmd->se_dev;
1020 struct tcmu_dev *udev = TCMU_DEV(se_dev);
1021 struct tcmu_cmd *tcmu_cmd;
1022 sense_reason_t scsi_ret;
1023 int ret;
1024
1025 tcmu_cmd = tcmu_alloc_cmd(se_cmd);
1026 if (!tcmu_cmd)
1027 return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
1028
1029 mutex_lock(&udev->cmdr_lock);
1030 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1031 mutex_unlock(&udev->cmdr_lock);
1032 if (ret < 0)
1033 tcmu_free_cmd(tcmu_cmd);
1034 return scsi_ret;
1035}
1036
1037static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
1038{
1039 struct se_cmd *se_cmd = cmd->se_cmd;
1040 struct tcmu_dev *udev = cmd->tcmu_dev;
1041
1042 /*
1043 * cmd has been completed already from timeout, just reclaim
1044 * data area space and free cmd
1045 */
1046 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1047 goto out;
1048
1049 tcmu_cmd_reset_dbi_cur(cmd);
1050
1051 if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
1052 pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
1053 cmd->se_cmd);
1054 entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
1055 } else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
1056 transport_copy_sense_to_cmd(se_cmd, entry->rsp.sense_buffer);
1057 } else if (se_cmd->se_cmd_flags & SCF_BIDI) {
1058 /* Get Data-In buffer before clean up */
1059 gather_data_area(udev, cmd, true);
1060 } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
1061 gather_data_area(udev, cmd, false);
1062 } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
1063 /* TODO: */
1064 } else if (se_cmd->data_direction != DMA_NONE) {
1065 pr_warn("TCMU: data direction was %d!\n",
1066 se_cmd->data_direction);
1067 }
1068
1069 target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
1070
1071out:
1072 cmd->se_cmd = NULL;
1073 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1074 tcmu_free_cmd(cmd);
1075}
1076
1077static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
1078{
1079 struct tcmu_mailbox *mb;
1080 int handled = 0;
1081
1082 if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
1083 pr_err("ring broken, not handling completions\n");
1084 return 0;
1085 }
1086
1087 mb = udev->mb_addr;
1088 tcmu_flush_dcache_range(mb, sizeof(*mb));
1089
1090 while (udev->cmdr_last_cleaned != READ_ONCE(mb->cmd_tail)) {
1091
1092 struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
1093 struct tcmu_cmd *cmd;
1094
1095 tcmu_flush_dcache_range(entry, sizeof(*entry));
1096
1097 if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
1098 UPDATE_HEAD(udev->cmdr_last_cleaned,
1099 tcmu_hdr_get_len(entry->hdr.len_op),
1100 udev->cmdr_size);
1101 continue;
1102 }
1103 WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
1104
1105 cmd = idr_remove(&udev->commands, entry->hdr.cmd_id);
1106 if (!cmd) {
1107 pr_err("cmd_id %u not found, ring is broken\n",
1108 entry->hdr.cmd_id);
1109 set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
1110 break;
1111 }
1112
1113 tcmu_handle_completion(cmd, entry);
1114
1115 UPDATE_HEAD(udev->cmdr_last_cleaned,
1116 tcmu_hdr_get_len(entry->hdr.len_op),
1117 udev->cmdr_size);
1118
1119 handled++;
1120 }
1121
1122 if (mb->cmd_tail == mb->cmd_head) {
1123 /* no more pending commands */
1124 del_timer(&udev->cmd_timer);
1125
1126 if (list_empty(&udev->cmdr_queue)) {
1127 /*
1128 * no more pending or waiting commands so try to
1129 * reclaim blocks if needed.
1130 */
1131 if (atomic_read(&global_db_count) >
1132 tcmu_global_max_blocks)
1133 schedule_delayed_work(&tcmu_unmap_work, 0);
1134 }
1135 }
1136
1137 return handled;
1138}
1139
1140static int tcmu_check_expired_cmd(int id, void *p, void *data)
1141{
1142 struct tcmu_cmd *cmd = p;
1143 struct tcmu_dev *udev = cmd->tcmu_dev;
1144 u8 scsi_status;
1145 struct se_cmd *se_cmd;
1146 bool is_running;
1147
1148 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
1149 return 0;
1150
1151 if (!time_after(jiffies, cmd->deadline))
1152 return 0;
1153
1154 is_running = list_empty(&cmd->cmdr_queue_entry);
1155 se_cmd = cmd->se_cmd;
1156
1157 if (is_running) {
1158 /*
1159 * If cmd_time_out is disabled but qfull is set deadline
1160 * will only reflect the qfull timeout. Ignore it.
1161 */
1162 if (!udev->cmd_time_out)
1163 return 0;
1164
1165 set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
1166 /*
1167 * target_complete_cmd will translate this to LUN COMM FAILURE
1168 */
1169 scsi_status = SAM_STAT_CHECK_CONDITION;
1170 } else {
1171 list_del_init(&cmd->cmdr_queue_entry);
1172
1173 idr_remove(&udev->commands, id);
1174 tcmu_free_cmd(cmd);
1175 scsi_status = SAM_STAT_TASK_SET_FULL;
1176 }
1177
1178 pr_debug("Timing out cmd %u on dev %s that is %s.\n",
1179 id, udev->name, is_running ? "inflight" : "queued");
1180
1181 target_complete_cmd(se_cmd, scsi_status);
1182 return 0;
1183}
1184
1185static void tcmu_device_timedout(struct tcmu_dev *udev)
1186{
1187 spin_lock(&timed_out_udevs_lock);
1188 if (list_empty(&udev->timedout_entry))
1189 list_add_tail(&udev->timedout_entry, &timed_out_udevs);
1190 spin_unlock(&timed_out_udevs_lock);
1191
1192 schedule_delayed_work(&tcmu_unmap_work, 0);
1193}
1194
1195static void tcmu_cmd_timedout(struct timer_list *t)
1196{
1197 struct tcmu_dev *udev = from_timer(udev, t, cmd_timer);
1198
1199 pr_debug("%s cmd timeout has expired\n", udev->name);
1200 tcmu_device_timedout(udev);
1201}
1202
1203static void tcmu_qfull_timedout(struct timer_list *t)
1204{
1205 struct tcmu_dev *udev = from_timer(udev, t, qfull_timer);
1206
1207 pr_debug("%s qfull timeout has expired\n", udev->name);
1208 tcmu_device_timedout(udev);
1209}
1210
1211static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
1212{
1213 struct tcmu_hba *tcmu_hba;
1214
1215 tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
1216 if (!tcmu_hba)
1217 return -ENOMEM;
1218
1219 tcmu_hba->host_id = host_id;
1220 hba->hba_ptr = tcmu_hba;
1221
1222 return 0;
1223}
1224
1225static void tcmu_detach_hba(struct se_hba *hba)
1226{
1227 kfree(hba->hba_ptr);
1228 hba->hba_ptr = NULL;
1229}
1230
1231static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
1232{
1233 struct tcmu_dev *udev;
1234
1235 udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
1236 if (!udev)
1237 return NULL;
1238 kref_init(&udev->kref);
1239
1240 udev->name = kstrdup(name, GFP_KERNEL);
1241 if (!udev->name) {
1242 kfree(udev);
1243 return NULL;
1244 }
1245
1246 udev->hba = hba;
1247 udev->cmd_time_out = TCMU_TIME_OUT;
1248 udev->qfull_time_out = -1;
1249
1250 udev->max_blocks = DATA_BLOCK_BITS_DEF;
1251 mutex_init(&udev->cmdr_lock);
1252
1253 INIT_LIST_HEAD(&udev->timedout_entry);
1254 INIT_LIST_HEAD(&udev->cmdr_queue);
1255 idr_init(&udev->commands);
1256
1257 timer_setup(&udev->qfull_timer, tcmu_qfull_timedout, 0);
1258 timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
1259
1260 init_waitqueue_head(&udev->nl_cmd_wq);
1261 spin_lock_init(&udev->nl_cmd_lock);
1262
1263 INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
1264
1265 return &udev->se_dev;
1266}
1267
1268static bool run_cmdr_queue(struct tcmu_dev *udev, bool fail)
1269{
1270 struct tcmu_cmd *tcmu_cmd, *tmp_cmd;
1271 LIST_HEAD(cmds);
1272 bool drained = true;
1273 sense_reason_t scsi_ret;
1274 int ret;
1275
1276 if (list_empty(&udev->cmdr_queue))
1277 return true;
1278
1279 pr_debug("running %s's cmdr queue forcefail %d\n", udev->name, fail);
1280
1281 list_splice_init(&udev->cmdr_queue, &cmds);
1282
1283 list_for_each_entry_safe(tcmu_cmd, tmp_cmd, &cmds, cmdr_queue_entry) {
1284 list_del_init(&tcmu_cmd->cmdr_queue_entry);
1285
1286 pr_debug("removing cmd %u on dev %s from queue\n",
1287 tcmu_cmd->cmd_id, udev->name);
1288
1289 if (fail) {
1290 idr_remove(&udev->commands, tcmu_cmd->cmd_id);
1291 /*
1292 * We were not able to even start the command, so
1293 * fail with busy to allow a retry in case runner
1294 * was only temporarily down. If the device is being
1295 * removed then LIO core will do the right thing and
1296 * fail the retry.
1297 */
1298 target_complete_cmd(tcmu_cmd->se_cmd, SAM_STAT_BUSY);
1299 tcmu_free_cmd(tcmu_cmd);
1300 continue;
1301 }
1302
1303 ret = queue_cmd_ring(tcmu_cmd, &scsi_ret);
1304 if (ret < 0) {
1305 pr_debug("cmd %u on dev %s failed with %u\n",
1306 tcmu_cmd->cmd_id, udev->name, scsi_ret);
1307
1308 idr_remove(&udev->commands, tcmu_cmd->cmd_id);
1309 /*
1310 * Ignore scsi_ret for now. target_complete_cmd
1311 * drops it.
1312 */
1313 target_complete_cmd(tcmu_cmd->se_cmd,
1314 SAM_STAT_CHECK_CONDITION);
1315 tcmu_free_cmd(tcmu_cmd);
1316 } else if (ret > 0) {
1317 pr_debug("ran out of space during cmdr queue run\n");
1318 /*
1319 * cmd was requeued, so just put all cmds back in
1320 * the queue
1321 */
1322 list_splice_tail(&cmds, &udev->cmdr_queue);
1323 drained = false;
1324 goto done;
1325 }
1326 }
1327 if (list_empty(&udev->cmdr_queue))
1328 del_timer(&udev->qfull_timer);
1329done:
1330 return drained;
1331}
1332
1333static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
1334{
1335 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1336
1337 mutex_lock(&udev->cmdr_lock);
1338 tcmu_handle_completions(udev);
1339 run_cmdr_queue(udev, false);
1340 mutex_unlock(&udev->cmdr_lock);
1341
1342 return 0;
1343}
1344
1345/*
1346 * mmap code from uio.c. Copied here because we want to hook mmap()
1347 * and this stuff must come along.
1348 */
1349static int tcmu_find_mem_index(struct vm_area_struct *vma)
1350{
1351 struct tcmu_dev *udev = vma->vm_private_data;
1352 struct uio_info *info = &udev->uio_info;
1353
1354 if (vma->vm_pgoff < MAX_UIO_MAPS) {
1355 if (info->mem[vma->vm_pgoff].size == 0)
1356 return -1;
1357 return (int)vma->vm_pgoff;
1358 }
1359 return -1;
1360}
1361
1362static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
1363{
1364 struct page *page;
1365
1366 mutex_lock(&udev->cmdr_lock);
1367 page = tcmu_get_block_page(udev, dbi);
1368 if (likely(page)) {
1369 mutex_unlock(&udev->cmdr_lock);
1370 return page;
1371 }
1372
1373 /*
1374 * Userspace messed up and passed in a address not in the
1375 * data iov passed to it.
1376 */
1377 pr_err("Invalid addr to data block mapping (dbi %u) on device %s\n",
1378 dbi, udev->name);
1379 page = NULL;
1380 mutex_unlock(&udev->cmdr_lock);
1381
1382 return page;
1383}
1384
1385static int tcmu_vma_fault(struct vm_fault *vmf)
1386{
1387 struct tcmu_dev *udev = vmf->vma->vm_private_data;
1388 struct uio_info *info = &udev->uio_info;
1389 struct page *page;
1390 unsigned long offset;
1391 void *addr;
1392
1393 int mi = tcmu_find_mem_index(vmf->vma);
1394 if (mi < 0)
1395 return VM_FAULT_SIGBUS;
1396
1397 /*
1398 * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
1399 * to use mem[N].
1400 */
1401 offset = (vmf->pgoff - mi) << PAGE_SHIFT;
1402
1403 if (offset < udev->data_off) {
1404 /* For the vmalloc()ed cmd area pages */
1405 addr = (void *)(unsigned long)info->mem[mi].addr + offset;
1406 page = vmalloc_to_page(addr);
1407 } else {
1408 uint32_t dbi;
1409
1410 /* For the dynamically growing data area pages */
1411 dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
1412 page = tcmu_try_get_block_page(udev, dbi);
1413 if (!page)
1414 return VM_FAULT_SIGBUS;
1415 }
1416
1417 get_page(page);
1418 vmf->page = page;
1419 return 0;
1420}
1421
1422static const struct vm_operations_struct tcmu_vm_ops = {
1423 .fault = tcmu_vma_fault,
1424};
1425
1426static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
1427{
1428 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1429
1430 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
1431 vma->vm_ops = &tcmu_vm_ops;
1432
1433 vma->vm_private_data = udev;
1434
1435 /* Ensure the mmap is exactly the right size */
1436 if (vma_pages(vma) != (udev->ring_size >> PAGE_SHIFT))
1437 return -EINVAL;
1438
1439 return 0;
1440}
1441
1442static int tcmu_open(struct uio_info *info, struct inode *inode)
1443{
1444 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1445
1446 /* O_EXCL not supported for char devs, so fake it? */
1447 if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
1448 return -EBUSY;
1449
1450 udev->inode = inode;
1451 kref_get(&udev->kref);
1452
1453 pr_debug("open\n");
1454
1455 return 0;
1456}
1457
1458static void tcmu_dev_call_rcu(struct rcu_head *p)
1459{
1460 struct se_device *dev = container_of(p, struct se_device, rcu_head);
1461 struct tcmu_dev *udev = TCMU_DEV(dev);
1462
1463 kfree(udev->uio_info.name);
1464 kfree(udev->name);
1465 kfree(udev);
1466}
1467
1468static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1469{
1470 if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1471 kmem_cache_free(tcmu_cmd_cache, cmd);
1472 return 0;
1473 }
1474 return -EINVAL;
1475}
1476
1477static void tcmu_blocks_release(struct radix_tree_root *blocks,
1478 int start, int end)
1479{
1480 int i;
1481 struct page *page;
1482
1483 for (i = start; i < end; i++) {
1484 page = radix_tree_delete(blocks, i);
1485 if (page) {
1486 __free_page(page);
1487 atomic_dec(&global_db_count);
1488 }
1489 }
1490}
1491
1492static void tcmu_dev_kref_release(struct kref *kref)
1493{
1494 struct tcmu_dev *udev = container_of(kref, struct tcmu_dev, kref);
1495 struct se_device *dev = &udev->se_dev;
1496 struct tcmu_cmd *cmd;
1497 bool all_expired = true;
1498 int i;
1499
1500 vfree(udev->mb_addr);
1501 udev->mb_addr = NULL;
1502
1503 spin_lock_bh(&timed_out_udevs_lock);
1504 if (!list_empty(&udev->timedout_entry))
1505 list_del(&udev->timedout_entry);
1506 spin_unlock_bh(&timed_out_udevs_lock);
1507
1508 /* Upper layer should drain all requests before calling this */
1509 mutex_lock(&udev->cmdr_lock);
1510 idr_for_each_entry(&udev->commands, cmd, i) {
1511 if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1512 all_expired = false;
1513 }
1514 idr_destroy(&udev->commands);
1515 WARN_ON(!all_expired);
1516
1517 tcmu_blocks_release(&udev->data_blocks, 0, udev->dbi_max + 1);
1518 kfree(udev->data_bitmap);
1519 mutex_unlock(&udev->cmdr_lock);
1520
1521 call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1522}
1523
1524static int tcmu_release(struct uio_info *info, struct inode *inode)
1525{
1526 struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
1527
1528 clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
1529
1530 pr_debug("close\n");
1531 /* release ref from open */
1532 kref_put(&udev->kref, tcmu_dev_kref_release);
1533 return 0;
1534}
1535
1536static void tcmu_init_genl_cmd_reply(struct tcmu_dev *udev, int cmd)
1537{
1538 struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1539
1540 if (!tcmu_kern_cmd_reply_supported)
1541 return;
1542
1543 if (udev->nl_reply_supported <= 0)
1544 return;
1545
1546relock:
1547 spin_lock(&udev->nl_cmd_lock);
1548
1549 if (nl_cmd->cmd != TCMU_CMD_UNSPEC) {
1550 spin_unlock(&udev->nl_cmd_lock);
1551 pr_debug("sleeping for open nl cmd\n");
1552 wait_event(udev->nl_cmd_wq, (nl_cmd->cmd == TCMU_CMD_UNSPEC));
1553 goto relock;
1554 }
1555
1556 memset(nl_cmd, 0, sizeof(*nl_cmd));
1557 nl_cmd->cmd = cmd;
1558 init_completion(&nl_cmd->complete);
1559
1560 spin_unlock(&udev->nl_cmd_lock);
1561}
1562
1563static int tcmu_wait_genl_cmd_reply(struct tcmu_dev *udev)
1564{
1565 struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
1566 int ret;
1567 DEFINE_WAIT(__wait);
1568
1569 if (!tcmu_kern_cmd_reply_supported)
1570 return 0;
1571
1572 if (udev->nl_reply_supported <= 0)
1573 return 0;
1574
1575 pr_debug("sleeping for nl reply\n");
1576 wait_for_completion(&nl_cmd->complete);
1577
1578 spin_lock(&udev->nl_cmd_lock);
1579 nl_cmd->cmd = TCMU_CMD_UNSPEC;
1580 ret = nl_cmd->status;
1581 nl_cmd->status = 0;
1582 spin_unlock(&udev->nl_cmd_lock);
1583
1584 wake_up_all(&udev->nl_cmd_wq);
1585
1586 return ret;
1587}
1588
1589static int tcmu_netlink_event(struct tcmu_dev *udev, enum tcmu_genl_cmd cmd,
1590 int reconfig_attr, const void *reconfig_data)
1591{
1592 struct sk_buff *skb;
1593 void *msg_header;
1594 int ret = -ENOMEM;
1595
1596 skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
1597 if (!skb)
1598 return ret;
1599
1600 msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
1601 if (!msg_header)
1602 goto free_skb;
1603
1604 ret = nla_put_string(skb, TCMU_ATTR_DEVICE, udev->uio_info.name);
1605 if (ret < 0)
1606 goto free_skb;
1607
1608 ret = nla_put_u32(skb, TCMU_ATTR_MINOR, udev->uio_info.uio_dev->minor);
1609 if (ret < 0)
1610 goto free_skb;
1611
1612 ret = nla_put_u32(skb, TCMU_ATTR_DEVICE_ID, udev->se_dev.dev_index);
1613 if (ret < 0)
1614 goto free_skb;
1615
1616 if (cmd == TCMU_CMD_RECONFIG_DEVICE) {
1617 switch (reconfig_attr) {
1618 case TCMU_ATTR_DEV_CFG:
1619 ret = nla_put_string(skb, reconfig_attr, reconfig_data);
1620 break;
1621 case TCMU_ATTR_DEV_SIZE:
1622 ret = nla_put_u64_64bit(skb, reconfig_attr,
1623 *((u64 *)reconfig_data),
1624 TCMU_ATTR_PAD);
1625 break;
1626 case TCMU_ATTR_WRITECACHE:
1627 ret = nla_put_u8(skb, reconfig_attr,
1628 *((u8 *)reconfig_data));
1629 break;
1630 default:
1631 BUG();
1632 }
1633
1634 if (ret < 0)
1635 goto free_skb;
1636 }
1637
1638 genlmsg_end(skb, msg_header);
1639
1640 tcmu_init_genl_cmd_reply(udev, cmd);
1641
1642 ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
1643 TCMU_MCGRP_CONFIG, GFP_KERNEL);
1644 /* We don't care if no one is listening */
1645 if (ret == -ESRCH)
1646 ret = 0;
1647 if (!ret)
1648 ret = tcmu_wait_genl_cmd_reply(udev);
1649
1650 return ret;
1651free_skb:
1652 nlmsg_free(skb);
1653 return ret;
1654}
1655
1656static int tcmu_update_uio_info(struct tcmu_dev *udev)
1657{
1658 struct tcmu_hba *hba = udev->hba->hba_ptr;
1659 struct uio_info *info;
1660 size_t size, used;
1661 char *str;
1662
1663 info = &udev->uio_info;
1664 size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
1665 udev->dev_config);
1666 size += 1; /* for \0 */
1667 str = kmalloc(size, GFP_KERNEL);
1668 if (!str)
1669 return -ENOMEM;
1670
1671 used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
1672 if (udev->dev_config[0])
1673 snprintf(str + used, size - used, "/%s", udev->dev_config);
1674
1675 /* If the old string exists, free it */
1676 kfree(info->name);
1677 info->name = str;
1678
1679 return 0;
1680}
1681
1682static int tcmu_configure_device(struct se_device *dev)
1683{
1684 struct tcmu_dev *udev = TCMU_DEV(dev);
1685 struct uio_info *info;
1686 struct tcmu_mailbox *mb;
1687 int ret = 0;
1688
1689 ret = tcmu_update_uio_info(udev);
1690 if (ret)
1691 return ret;
1692
1693 info = &udev->uio_info;
1694
1695 udev->data_bitmap = kzalloc(BITS_TO_LONGS(udev->max_blocks) *
1696 sizeof(unsigned long), GFP_KERNEL);
1697 if (!udev->data_bitmap) {
1698 ret = -ENOMEM;
1699 goto err_bitmap_alloc;
1700 }
1701
1702 udev->mb_addr = vzalloc(CMDR_SIZE);
1703 if (!udev->mb_addr) {
1704 ret = -ENOMEM;
1705 goto err_vzalloc;
1706 }
1707
1708 /* mailbox fits in first part of CMDR space */
1709 udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
1710 udev->data_off = CMDR_SIZE;
1711 udev->data_size = udev->max_blocks * DATA_BLOCK_SIZE;
1712 udev->dbi_thresh = 0; /* Default in Idle state */
1713
1714 /* Initialise the mailbox of the ring buffer */
1715 mb = udev->mb_addr;
1716 mb->version = TCMU_MAILBOX_VERSION;
1717 mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC;
1718 mb->cmdr_off = CMDR_OFF;
1719 mb->cmdr_size = udev->cmdr_size;
1720
1721 WARN_ON(!PAGE_ALIGNED(udev->data_off));
1722 WARN_ON(udev->data_size % PAGE_SIZE);
1723 WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
1724
1725 info->version = __stringify(TCMU_MAILBOX_VERSION);
1726
1727 info->mem[0].name = "tcm-user command & data buffer";
1728 info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
1729 info->mem[0].size = udev->ring_size = udev->data_size + CMDR_SIZE;
1730 info->mem[0].memtype = UIO_MEM_NONE;
1731
1732 info->irqcontrol = tcmu_irqcontrol;
1733 info->irq = UIO_IRQ_CUSTOM;
1734
1735 info->mmap = tcmu_mmap;
1736 info->open = tcmu_open;
1737 info->release = tcmu_release;
1738
1739 ret = uio_register_device(tcmu_root_device, info);
1740 if (ret)
1741 goto err_register;
1742
1743 /* User can set hw_block_size before enable the device */
1744 if (dev->dev_attrib.hw_block_size == 0)
1745 dev->dev_attrib.hw_block_size = 512;
1746 /* Other attributes can be configured in userspace */
1747 if (!dev->dev_attrib.hw_max_sectors)
1748 dev->dev_attrib.hw_max_sectors = 128;
1749 if (!dev->dev_attrib.emulate_write_cache)
1750 dev->dev_attrib.emulate_write_cache = 0;
1751 dev->dev_attrib.hw_queue_depth = 128;
1752
1753 /* If user didn't explicitly disable netlink reply support, use
1754 * module scope setting.
1755 */
1756 if (udev->nl_reply_supported >= 0)
1757 udev->nl_reply_supported = tcmu_kern_cmd_reply_supported;
1758
1759 /*
1760 * Get a ref incase userspace does a close on the uio device before
1761 * LIO has initiated tcmu_free_device.
1762 */
1763 kref_get(&udev->kref);
1764
1765 ret = tcmu_netlink_event(udev, TCMU_CMD_ADDED_DEVICE, 0, NULL);
1766 if (ret)
1767 goto err_netlink;
1768
1769 mutex_lock(&root_udev_mutex);
1770 list_add(&udev->node, &root_udev);
1771 mutex_unlock(&root_udev_mutex);
1772
1773 return 0;
1774
1775err_netlink:
1776 kref_put(&udev->kref, tcmu_dev_kref_release);
1777 uio_unregister_device(&udev->uio_info);
1778err_register:
1779 vfree(udev->mb_addr);
1780 udev->mb_addr = NULL;
1781err_vzalloc:
1782 kfree(udev->data_bitmap);
1783 udev->data_bitmap = NULL;
1784err_bitmap_alloc:
1785 kfree(info->name);
1786 info->name = NULL;
1787
1788 return ret;
1789}
1790
1791static bool tcmu_dev_configured(struct tcmu_dev *udev)
1792{
1793 return udev->uio_info.uio_dev ? true : false;
1794}
1795
1796static void tcmu_free_device(struct se_device *dev)
1797{
1798 struct tcmu_dev *udev = TCMU_DEV(dev);
1799
1800 /* release ref from init */
1801 kref_put(&udev->kref, tcmu_dev_kref_release);
1802}
1803
1804static void tcmu_destroy_device(struct se_device *dev)
1805{
1806 struct tcmu_dev *udev = TCMU_DEV(dev);
1807
1808 del_timer_sync(&udev->cmd_timer);
1809 del_timer_sync(&udev->qfull_timer);
1810
1811 mutex_lock(&root_udev_mutex);
1812 list_del(&udev->node);
1813 mutex_unlock(&root_udev_mutex);
1814
1815 tcmu_netlink_event(udev, TCMU_CMD_REMOVED_DEVICE, 0, NULL);
1816
1817 uio_unregister_device(&udev->uio_info);
1818
1819 /* release ref from configure */
1820 kref_put(&udev->kref, tcmu_dev_kref_release);
1821}
1822
1823static void tcmu_unblock_dev(struct tcmu_dev *udev)
1824{
1825 mutex_lock(&udev->cmdr_lock);
1826 clear_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags);
1827 mutex_unlock(&udev->cmdr_lock);
1828}
1829
1830static void tcmu_block_dev(struct tcmu_dev *udev)
1831{
1832 mutex_lock(&udev->cmdr_lock);
1833
1834 if (test_and_set_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
1835 goto unlock;
1836
1837 /* complete IO that has executed successfully */
1838 tcmu_handle_completions(udev);
1839 /* fail IO waiting to be queued */
1840 run_cmdr_queue(udev, true);
1841
1842unlock:
1843 mutex_unlock(&udev->cmdr_lock);
1844}
1845
1846static void tcmu_reset_ring(struct tcmu_dev *udev, u8 err_level)
1847{
1848 struct tcmu_mailbox *mb;
1849 struct tcmu_cmd *cmd;
1850 int i;
1851
1852 mutex_lock(&udev->cmdr_lock);
1853
1854 idr_for_each_entry(&udev->commands, cmd, i) {
1855 if (!list_empty(&cmd->cmdr_queue_entry))
1856 continue;
1857
1858 pr_debug("removing cmd %u on dev %s from ring (is expired %d)\n",
1859 cmd->cmd_id, udev->name,
1860 test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags));
1861
1862 idr_remove(&udev->commands, i);
1863 if (!test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1864 if (err_level == 1) {
1865 /*
1866 * Userspace was not able to start the
1867 * command or it is retryable.
1868 */
1869 target_complete_cmd(cmd->se_cmd, SAM_STAT_BUSY);
1870 } else {
1871 /* hard failure */
1872 target_complete_cmd(cmd->se_cmd,
1873 SAM_STAT_CHECK_CONDITION);
1874 }
1875 }
1876 tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
1877 tcmu_free_cmd(cmd);
1878 }
1879
1880 mb = udev->mb_addr;
1881 tcmu_flush_dcache_range(mb, sizeof(*mb));
1882 pr_debug("mb last %u head %u tail %u\n", udev->cmdr_last_cleaned,
1883 mb->cmd_tail, mb->cmd_head);
1884
1885 udev->cmdr_last_cleaned = 0;
1886 mb->cmd_tail = 0;
1887 mb->cmd_head = 0;
1888 tcmu_flush_dcache_range(mb, sizeof(*mb));
1889
1890 del_timer(&udev->cmd_timer);
1891
1892 mutex_unlock(&udev->cmdr_lock);
1893}
1894
1895enum {
1896 Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_hw_max_sectors,
1897 Opt_nl_reply_supported, Opt_max_data_area_mb, Opt_err,
1898};
1899
1900static match_table_t tokens = {
1901 {Opt_dev_config, "dev_config=%s"},
1902 {Opt_dev_size, "dev_size=%u"},
1903 {Opt_hw_block_size, "hw_block_size=%u"},
1904 {Opt_hw_max_sectors, "hw_max_sectors=%u"},
1905 {Opt_nl_reply_supported, "nl_reply_supported=%d"},
1906 {Opt_max_data_area_mb, "max_data_area_mb=%u"},
1907 {Opt_err, NULL}
1908};
1909
1910static int tcmu_set_dev_attrib(substring_t *arg, u32 *dev_attrib)
1911{
1912 unsigned long tmp_ul;
1913 char *arg_p;
1914 int ret;
1915
1916 arg_p = match_strdup(arg);
1917 if (!arg_p)
1918 return -ENOMEM;
1919
1920 ret = kstrtoul(arg_p, 0, &tmp_ul);
1921 kfree(arg_p);
1922 if (ret < 0) {
1923 pr_err("kstrtoul() failed for dev attrib\n");
1924 return ret;
1925 }
1926 if (!tmp_ul) {
1927 pr_err("dev attrib must be nonzero\n");
1928 return -EINVAL;
1929 }
1930 *dev_attrib = tmp_ul;
1931 return 0;
1932}
1933
1934static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
1935 const char *page, ssize_t count)
1936{
1937 struct tcmu_dev *udev = TCMU_DEV(dev);
1938 char *orig, *ptr, *opts, *arg_p;
1939 substring_t args[MAX_OPT_ARGS];
1940 int ret = 0, token, tmpval;
1941
1942 opts = kstrdup(page, GFP_KERNEL);
1943 if (!opts)
1944 return -ENOMEM;
1945
1946 orig = opts;
1947
1948 while ((ptr = strsep(&opts, ",\n")) != NULL) {
1949 if (!*ptr)
1950 continue;
1951
1952 token = match_token(ptr, tokens, args);
1953 switch (token) {
1954 case Opt_dev_config:
1955 if (match_strlcpy(udev->dev_config, &args[0],
1956 TCMU_CONFIG_LEN) == 0) {
1957 ret = -EINVAL;
1958 break;
1959 }
1960 pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
1961 break;
1962 case Opt_dev_size:
1963 arg_p = match_strdup(&args[0]);
1964 if (!arg_p) {
1965 ret = -ENOMEM;
1966 break;
1967 }
1968 ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
1969 kfree(arg_p);
1970 if (ret < 0)
1971 pr_err("kstrtoul() failed for dev_size=\n");
1972 break;
1973 case Opt_hw_block_size:
1974 ret = tcmu_set_dev_attrib(&args[0],
1975 &(dev->dev_attrib.hw_block_size));
1976 break;
1977 case Opt_hw_max_sectors:
1978 ret = tcmu_set_dev_attrib(&args[0],
1979 &(dev->dev_attrib.hw_max_sectors));
1980 break;
1981 case Opt_nl_reply_supported:
1982 arg_p = match_strdup(&args[0]);
1983 if (!arg_p) {
1984 ret = -ENOMEM;
1985 break;
1986 }
1987 ret = kstrtoint(arg_p, 0, &udev->nl_reply_supported);
1988 kfree(arg_p);
1989 if (ret < 0)
1990 pr_err("kstrtoint() failed for nl_reply_supported=\n");
1991 break;
1992 case Opt_max_data_area_mb:
1993 if (dev->export_count) {
1994 pr_err("Unable to set max_data_area_mb while exports exist\n");
1995 ret = -EINVAL;
1996 break;
1997 }
1998
1999 arg_p = match_strdup(&args[0]);
2000 if (!arg_p) {
2001 ret = -ENOMEM;
2002 break;
2003 }
2004 ret = kstrtoint(arg_p, 0, &tmpval);
2005 kfree(arg_p);
2006 if (ret < 0) {
2007 pr_err("kstrtoint() failed for max_data_area_mb=\n");
2008 break;
2009 }
2010
2011 if (tmpval <= 0) {
2012 pr_err("Invalid max_data_area %d\n", tmpval);
2013 ret = -EINVAL;
2014 break;
2015 }
2016
2017 udev->max_blocks = TCMU_MBS_TO_BLOCKS(tmpval);
2018 if (udev->max_blocks > tcmu_global_max_blocks) {
2019 pr_err("%d is too large. Adjusting max_data_area_mb to global limit of %u\n",
2020 tmpval,
2021 TCMU_BLOCKS_TO_MBS(tcmu_global_max_blocks));
2022 udev->max_blocks = tcmu_global_max_blocks;
2023 }
2024 break;
2025 default:
2026 break;
2027 }
2028
2029 if (ret)
2030 break;
2031 }
2032
2033 kfree(orig);
2034 return (!ret) ? count : ret;
2035}
2036
2037static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
2038{
2039 struct tcmu_dev *udev = TCMU_DEV(dev);
2040 ssize_t bl = 0;
2041
2042 bl = sprintf(b + bl, "Config: %s ",
2043 udev->dev_config[0] ? udev->dev_config : "NULL");
2044 bl += sprintf(b + bl, "Size: %zu ", udev->dev_size);
2045 bl += sprintf(b + bl, "MaxDataAreaMB: %u\n",
2046 TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2047
2048 return bl;
2049}
2050
2051static sector_t tcmu_get_blocks(struct se_device *dev)
2052{
2053 struct tcmu_dev *udev = TCMU_DEV(dev);
2054
2055 return div_u64(udev->dev_size - dev->dev_attrib.block_size,
2056 dev->dev_attrib.block_size);
2057}
2058
2059static sense_reason_t
2060tcmu_parse_cdb(struct se_cmd *cmd)
2061{
2062 return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
2063}
2064
2065static ssize_t tcmu_cmd_time_out_show(struct config_item *item, char *page)
2066{
2067 struct se_dev_attrib *da = container_of(to_config_group(item),
2068 struct se_dev_attrib, da_group);
2069 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2070
2071 return snprintf(page, PAGE_SIZE, "%lu\n", udev->cmd_time_out / MSEC_PER_SEC);
2072}
2073
2074static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *page,
2075 size_t count)
2076{
2077 struct se_dev_attrib *da = container_of(to_config_group(item),
2078 struct se_dev_attrib, da_group);
2079 struct tcmu_dev *udev = container_of(da->da_dev,
2080 struct tcmu_dev, se_dev);
2081 u32 val;
2082 int ret;
2083
2084 if (da->da_dev->export_count) {
2085 pr_err("Unable to set tcmu cmd_time_out while exports exist\n");
2086 return -EINVAL;
2087 }
2088
2089 ret = kstrtou32(page, 0, &val);
2090 if (ret < 0)
2091 return ret;
2092
2093 udev->cmd_time_out = val * MSEC_PER_SEC;
2094 return count;
2095}
2096CONFIGFS_ATTR(tcmu_, cmd_time_out);
2097
2098static ssize_t tcmu_qfull_time_out_show(struct config_item *item, char *page)
2099{
2100 struct se_dev_attrib *da = container_of(to_config_group(item),
2101 struct se_dev_attrib, da_group);
2102 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2103
2104 return snprintf(page, PAGE_SIZE, "%ld\n", udev->qfull_time_out <= 0 ?
2105 udev->qfull_time_out :
2106 udev->qfull_time_out / MSEC_PER_SEC);
2107}
2108
2109static ssize_t tcmu_qfull_time_out_store(struct config_item *item,
2110 const char *page, size_t count)
2111{
2112 struct se_dev_attrib *da = container_of(to_config_group(item),
2113 struct se_dev_attrib, da_group);
2114 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2115 s32 val;
2116 int ret;
2117
2118 ret = kstrtos32(page, 0, &val);
2119 if (ret < 0)
2120 return ret;
2121
2122 if (val >= 0) {
2123 udev->qfull_time_out = val * MSEC_PER_SEC;
2124 } else if (val == -1) {
2125 udev->qfull_time_out = val;
2126 } else {
2127 printk(KERN_ERR "Invalid qfull timeout value %d\n", val);
2128 return -EINVAL;
2129 }
2130 return count;
2131}
2132CONFIGFS_ATTR(tcmu_, qfull_time_out);
2133
2134static ssize_t tcmu_max_data_area_mb_show(struct config_item *item, char *page)
2135{
2136 struct se_dev_attrib *da = container_of(to_config_group(item),
2137 struct se_dev_attrib, da_group);
2138 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2139
2140 return snprintf(page, PAGE_SIZE, "%u\n",
2141 TCMU_BLOCKS_TO_MBS(udev->max_blocks));
2142}
2143CONFIGFS_ATTR_RO(tcmu_, max_data_area_mb);
2144
2145static ssize_t tcmu_dev_config_show(struct config_item *item, char *page)
2146{
2147 struct se_dev_attrib *da = container_of(to_config_group(item),
2148 struct se_dev_attrib, da_group);
2149 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2150
2151 return snprintf(page, PAGE_SIZE, "%s\n", udev->dev_config);
2152}
2153
2154static ssize_t tcmu_dev_config_store(struct config_item *item, const char *page,
2155 size_t count)
2156{
2157 struct se_dev_attrib *da = container_of(to_config_group(item),
2158 struct se_dev_attrib, da_group);
2159 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2160 int ret, len;
2161
2162 len = strlen(page);
2163 if (!len || len > TCMU_CONFIG_LEN - 1)
2164 return -EINVAL;
2165
2166 /* Check if device has been configured before */
2167 if (tcmu_dev_configured(udev)) {
2168 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
2169 TCMU_ATTR_DEV_CFG, page);
2170 if (ret) {
2171 pr_err("Unable to reconfigure device\n");
2172 return ret;
2173 }
2174 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2175
2176 ret = tcmu_update_uio_info(udev);
2177 if (ret)
2178 return ret;
2179 return count;
2180 }
2181 strlcpy(udev->dev_config, page, TCMU_CONFIG_LEN);
2182
2183 return count;
2184}
2185CONFIGFS_ATTR(tcmu_, dev_config);
2186
2187static ssize_t tcmu_dev_size_show(struct config_item *item, char *page)
2188{
2189 struct se_dev_attrib *da = container_of(to_config_group(item),
2190 struct se_dev_attrib, da_group);
2191 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2192
2193 return snprintf(page, PAGE_SIZE, "%zu\n", udev->dev_size);
2194}
2195
2196static ssize_t tcmu_dev_size_store(struct config_item *item, const char *page,
2197 size_t count)
2198{
2199 struct se_dev_attrib *da = container_of(to_config_group(item),
2200 struct se_dev_attrib, da_group);
2201 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2202 u64 val;
2203 int ret;
2204
2205 ret = kstrtou64(page, 0, &val);
2206 if (ret < 0)
2207 return ret;
2208
2209 /* Check if device has been configured before */
2210 if (tcmu_dev_configured(udev)) {
2211 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
2212 TCMU_ATTR_DEV_SIZE, &val);
2213 if (ret) {
2214 pr_err("Unable to reconfigure device\n");
2215 return ret;
2216 }
2217 }
2218 udev->dev_size = val;
2219 return count;
2220}
2221CONFIGFS_ATTR(tcmu_, dev_size);
2222
2223static ssize_t tcmu_nl_reply_supported_show(struct config_item *item,
2224 char *page)
2225{
2226 struct se_dev_attrib *da = container_of(to_config_group(item),
2227 struct se_dev_attrib, da_group);
2228 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2229
2230 return snprintf(page, PAGE_SIZE, "%d\n", udev->nl_reply_supported);
2231}
2232
2233static ssize_t tcmu_nl_reply_supported_store(struct config_item *item,
2234 const char *page, size_t count)
2235{
2236 struct se_dev_attrib *da = container_of(to_config_group(item),
2237 struct se_dev_attrib, da_group);
2238 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2239 s8 val;
2240 int ret;
2241
2242 ret = kstrtos8(page, 0, &val);
2243 if (ret < 0)
2244 return ret;
2245
2246 udev->nl_reply_supported = val;
2247 return count;
2248}
2249CONFIGFS_ATTR(tcmu_, nl_reply_supported);
2250
2251static ssize_t tcmu_emulate_write_cache_show(struct config_item *item,
2252 char *page)
2253{
2254 struct se_dev_attrib *da = container_of(to_config_group(item),
2255 struct se_dev_attrib, da_group);
2256
2257 return snprintf(page, PAGE_SIZE, "%i\n", da->emulate_write_cache);
2258}
2259
2260static ssize_t tcmu_emulate_write_cache_store(struct config_item *item,
2261 const char *page, size_t count)
2262{
2263 struct se_dev_attrib *da = container_of(to_config_group(item),
2264 struct se_dev_attrib, da_group);
2265 struct tcmu_dev *udev = TCMU_DEV(da->da_dev);
2266 u8 val;
2267 int ret;
2268
2269 ret = kstrtou8(page, 0, &val);
2270 if (ret < 0)
2271 return ret;
2272
2273 /* Check if device has been configured before */
2274 if (tcmu_dev_configured(udev)) {
2275 ret = tcmu_netlink_event(udev, TCMU_CMD_RECONFIG_DEVICE,
2276 TCMU_ATTR_WRITECACHE, &val);
2277 if (ret) {
2278 pr_err("Unable to reconfigure device\n");
2279 return ret;
2280 }
2281 }
2282
2283 da->emulate_write_cache = val;
2284 return count;
2285}
2286CONFIGFS_ATTR(tcmu_, emulate_write_cache);
2287
2288static ssize_t tcmu_block_dev_show(struct config_item *item, char *page)
2289{
2290 struct se_device *se_dev = container_of(to_config_group(item),
2291 struct se_device,
2292 dev_action_group);
2293 struct tcmu_dev *udev = TCMU_DEV(se_dev);
2294
2295 if (test_bit(TCMU_DEV_BIT_BLOCKED, &udev->flags))
2296 return snprintf(page, PAGE_SIZE, "%s\n", "blocked");
2297 else
2298 return snprintf(page, PAGE_SIZE, "%s\n", "unblocked");
2299}
2300
2301static ssize_t tcmu_block_dev_store(struct config_item *item, const char *page,
2302 size_t count)
2303{
2304 struct se_device *se_dev = container_of(to_config_group(item),
2305 struct se_device,
2306 dev_action_group);
2307 struct tcmu_dev *udev = TCMU_DEV(se_dev);
2308 u8 val;
2309 int ret;
2310
2311 ret = kstrtou8(page, 0, &val);
2312 if (ret < 0)
2313 return ret;
2314
2315 if (val > 1) {
2316 pr_err("Invalid block value %d\n", val);
2317 return -EINVAL;
2318 }
2319
2320 if (!val)
2321 tcmu_unblock_dev(udev);
2322 else
2323 tcmu_block_dev(udev);
2324 return count;
2325}
2326CONFIGFS_ATTR(tcmu_, block_dev);
2327
2328static ssize_t tcmu_reset_ring_store(struct config_item *item, const char *page,
2329 size_t count)
2330{
2331 struct se_device *se_dev = container_of(to_config_group(item),
2332 struct se_device,
2333 dev_action_group);
2334 struct tcmu_dev *udev = TCMU_DEV(se_dev);
2335 u8 val;
2336 int ret;
2337
2338 ret = kstrtou8(page, 0, &val);
2339 if (ret < 0)
2340 return ret;
2341
2342 if (val != 1 && val != 2) {
2343 pr_err("Invalid reset ring value %d\n", val);
2344 return -EINVAL;
2345 }
2346
2347 tcmu_reset_ring(udev, val);
2348 return count;
2349}
2350CONFIGFS_ATTR_WO(tcmu_, reset_ring);
2351
2352static struct configfs_attribute *tcmu_attrib_attrs[] = {
2353 &tcmu_attr_cmd_time_out,
2354 &tcmu_attr_qfull_time_out,
2355 &tcmu_attr_max_data_area_mb,
2356 &tcmu_attr_dev_config,
2357 &tcmu_attr_dev_size,
2358 &tcmu_attr_emulate_write_cache,
2359 &tcmu_attr_nl_reply_supported,
2360 NULL,
2361};
2362
2363static struct configfs_attribute **tcmu_attrs;
2364
2365static struct configfs_attribute *tcmu_action_attrs[] = {
2366 &tcmu_attr_block_dev,
2367 &tcmu_attr_reset_ring,
2368 NULL,
2369};
2370
2371static struct target_backend_ops tcmu_ops = {
2372 .name = "user",
2373 .owner = THIS_MODULE,
2374 .transport_flags = TRANSPORT_FLAG_PASSTHROUGH,
2375 .attach_hba = tcmu_attach_hba,
2376 .detach_hba = tcmu_detach_hba,
2377 .alloc_device = tcmu_alloc_device,
2378 .configure_device = tcmu_configure_device,
2379 .destroy_device = tcmu_destroy_device,
2380 .free_device = tcmu_free_device,
2381 .parse_cdb = tcmu_parse_cdb,
2382 .set_configfs_dev_params = tcmu_set_configfs_dev_params,
2383 .show_configfs_dev_params = tcmu_show_configfs_dev_params,
2384 .get_device_type = sbc_get_device_type,
2385 .get_blocks = tcmu_get_blocks,
2386 .tb_dev_action_attrs = tcmu_action_attrs,
2387};
2388
2389static void find_free_blocks(void)
2390{
2391 struct tcmu_dev *udev;
2392 loff_t off;
2393 u32 start, end, block, total_freed = 0;
2394
2395 if (atomic_read(&global_db_count) <= tcmu_global_max_blocks)
2396 return;
2397
2398 mutex_lock(&root_udev_mutex);
2399 list_for_each_entry(udev, &root_udev, node) {
2400 mutex_lock(&udev->cmdr_lock);
2401
2402 /* Try to complete the finished commands first */
2403 tcmu_handle_completions(udev);
2404
2405 /* Skip the udevs in idle */
2406 if (!udev->dbi_thresh) {
2407 mutex_unlock(&udev->cmdr_lock);
2408 continue;
2409 }
2410
2411 end = udev->dbi_max + 1;
2412 block = find_last_bit(udev->data_bitmap, end);
2413 if (block == udev->dbi_max) {
2414 /*
2415 * The last bit is dbi_max, so it is not possible
2416 * reclaim any blocks.
2417 */
2418 mutex_unlock(&udev->cmdr_lock);
2419 continue;
2420 } else if (block == end) {
2421 /* The current udev will goto idle state */
2422 udev->dbi_thresh = start = 0;
2423 udev->dbi_max = 0;
2424 } else {
2425 udev->dbi_thresh = start = block + 1;
2426 udev->dbi_max = block;
2427 }
2428
2429 /* Here will truncate the data area from off */
2430 off = udev->data_off + start * DATA_BLOCK_SIZE;
2431 unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
2432
2433 /* Release the block pages */
2434 tcmu_blocks_release(&udev->data_blocks, start, end);
2435 mutex_unlock(&udev->cmdr_lock);
2436
2437 total_freed += end - start;
2438 pr_debug("Freed %u blocks (total %u) from %s.\n", end - start,
2439 total_freed, udev->name);
2440 }
2441 mutex_unlock(&root_udev_mutex);
2442
2443 if (atomic_read(&global_db_count) > tcmu_global_max_blocks)
2444 schedule_delayed_work(&tcmu_unmap_work, msecs_to_jiffies(5000));
2445}
2446
2447static void check_timedout_devices(void)
2448{
2449 struct tcmu_dev *udev, *tmp_dev;
2450 LIST_HEAD(devs);
2451
2452 spin_lock_bh(&timed_out_udevs_lock);
2453 list_splice_init(&timed_out_udevs, &devs);
2454
2455 list_for_each_entry_safe(udev, tmp_dev, &devs, timedout_entry) {
2456 list_del_init(&udev->timedout_entry);
2457 spin_unlock_bh(&timed_out_udevs_lock);
2458
2459 mutex_lock(&udev->cmdr_lock);
2460 idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
2461 mutex_unlock(&udev->cmdr_lock);
2462
2463 spin_lock_bh(&timed_out_udevs_lock);
2464 }
2465
2466 spin_unlock_bh(&timed_out_udevs_lock);
2467}
2468
2469static void tcmu_unmap_work_fn(struct work_struct *work)
2470{
2471 check_timedout_devices();
2472 find_free_blocks();
2473}
2474
2475static int __init tcmu_module_init(void)
2476{
2477 int ret, i, k, len = 0;
2478
2479 BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
2480
2481 INIT_DELAYED_WORK(&tcmu_unmap_work, tcmu_unmap_work_fn);
2482
2483 tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
2484 sizeof(struct tcmu_cmd),
2485 __alignof__(struct tcmu_cmd),
2486 0, NULL);
2487 if (!tcmu_cmd_cache)
2488 return -ENOMEM;
2489
2490 tcmu_root_device = root_device_register("tcm_user");
2491 if (IS_ERR(tcmu_root_device)) {
2492 ret = PTR_ERR(tcmu_root_device);
2493 goto out_free_cache;
2494 }
2495
2496 ret = genl_register_family(&tcmu_genl_family);
2497 if (ret < 0) {
2498 goto out_unreg_device;
2499 }
2500
2501 for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2502 len += sizeof(struct configfs_attribute *);
2503 }
2504 for (i = 0; tcmu_attrib_attrs[i] != NULL; i++) {
2505 len += sizeof(struct configfs_attribute *);
2506 }
2507 len += sizeof(struct configfs_attribute *);
2508
2509 tcmu_attrs = kzalloc(len, GFP_KERNEL);
2510 if (!tcmu_attrs) {
2511 ret = -ENOMEM;
2512 goto out_unreg_genl;
2513 }
2514
2515 for (i = 0; passthrough_attrib_attrs[i] != NULL; i++) {
2516 tcmu_attrs[i] = passthrough_attrib_attrs[i];
2517 }
2518 for (k = 0; tcmu_attrib_attrs[k] != NULL; k++) {
2519 tcmu_attrs[i] = tcmu_attrib_attrs[k];
2520 i++;
2521 }
2522 tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
2523
2524 ret = transport_backend_register(&tcmu_ops);
2525 if (ret)
2526 goto out_attrs;
2527
2528 return 0;
2529
2530out_attrs:
2531 kfree(tcmu_attrs);
2532out_unreg_genl:
2533 genl_unregister_family(&tcmu_genl_family);
2534out_unreg_device:
2535 root_device_unregister(tcmu_root_device);
2536out_free_cache:
2537 kmem_cache_destroy(tcmu_cmd_cache);
2538
2539 return ret;
2540}
2541
2542static void __exit tcmu_module_exit(void)
2543{
2544 cancel_delayed_work_sync(&tcmu_unmap_work);
2545 target_backend_unregister(&tcmu_ops);
2546 kfree(tcmu_attrs);
2547 genl_unregister_family(&tcmu_genl_family);
2548 root_device_unregister(tcmu_root_device);
2549 kmem_cache_destroy(tcmu_cmd_cache);
2550}
2551
2552MODULE_DESCRIPTION("TCM USER subsystem plugin");
2553MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
2554MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
2555MODULE_LICENSE("GPL");
2556
2557module_init(tcmu_module_init);
2558module_exit(tcmu_module_exit);