source: src/linux/universal/linux-4.9/drivers/target/target_core_user.c @ 31885

Last change on this file since 31885 was 31885, checked in by brainslayer, 5 weeks ago

update

File size: 31.3 KB
Line 
1/*
2 * Copyright (C) 2013 Shaohua Li <shli@kernel.org>
3 * Copyright (C) 2014 Red Hat, Inc.
4 * Copyright (C) 2015 Arrikto, Inc.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20#include <linux/spinlock.h>
21#include <linux/module.h>
22#include <linux/idr.h>
23#include <linux/kernel.h>
24#include <linux/timer.h>
25#include <linux/parser.h>
26#include <linux/vmalloc.h>
27#include <linux/uio_driver.h>
28#include <linux/stringify.h>
29#include <linux/bitops.h>
30#include <net/genetlink.h>
31#include <scsi/scsi_common.h>
32#include <scsi/scsi_proto.h>
33#include <target/target_core_base.h>
34#include <target/target_core_fabric.h>
35#include <target/target_core_backend.h>
36
37#include <linux/target_core_user.h>
38
39/*
40 * Define a shared-memory interface for LIO to pass SCSI commands and
41 * data to userspace for processing. This is to allow backends that
42 * are too complex for in-kernel support to be possible.
43 *
44 * It uses the UIO framework to do a lot of the device-creation and
45 * introspection work for us.
46 *
47 * See the .h file for how the ring is laid out. Note that while the
48 * command ring is defined, the particulars of the data area are
49 * not. Offset values in the command entry point to other locations
50 * internal to the mmap()ed area. There is separate space outside the
51 * command ring for data buffers. This leaves maximum flexibility for
52 * moving buffer allocations, or even page flipping or other
53 * allocation techniques, without altering the command ring layout.
54 *
55 * SECURITY:
56 * The user process must be assumed to be malicious. There's no way to
57 * prevent it breaking the command ring protocol if it wants, but in
58 * order to prevent other issues we must only ever read *data* from
59 * the shared memory area, not offsets or sizes. This applies to
60 * command ring entries as well as the mailbox. Extra code needed for
61 * this may have a 'UAM' comment.
62 */
63
64
65#define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
66
67#define DATA_BLOCK_BITS 256
68#define DATA_BLOCK_SIZE 4096
69
70#define CMDR_SIZE (16 * 4096)
71#define DATA_SIZE (DATA_BLOCK_BITS * DATA_BLOCK_SIZE)
72
73#define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
74
75static struct device *tcmu_root_device;
76
77struct tcmu_hba {
78        u32 host_id;
79};
80
81#define TCMU_CONFIG_LEN 256
82
83struct tcmu_dev {
84        struct se_device se_dev;
85
86        char *name;
87        struct se_hba *hba;
88
89#define TCMU_DEV_BIT_OPEN 0
90#define TCMU_DEV_BIT_BROKEN 1
91        unsigned long flags;
92
93        struct uio_info uio_info;
94
95        struct tcmu_mailbox *mb_addr;
96        size_t dev_size;
97        u32 cmdr_size;
98        u32 cmdr_last_cleaned;
99        /* Offset of data area from start of mb */
100        /* Must add data_off and mb_addr to get the address */
101        size_t data_off;
102        size_t data_size;
103
104        DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
105
106        wait_queue_head_t wait_cmdr;
107        /* TODO should this be a mutex? */
108        spinlock_t cmdr_lock;
109
110        struct idr commands;
111        spinlock_t commands_lock;
112
113        struct timer_list timeout;
114
115        char dev_config[TCMU_CONFIG_LEN];
116};
117
118#define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev, se_dev)
119
120#define CMDR_OFF sizeof(struct tcmu_mailbox)
121
122struct tcmu_cmd {
123        struct se_cmd *se_cmd;
124        struct tcmu_dev *tcmu_dev;
125
126        uint16_t cmd_id;
127
128        /* Can't use se_cmd when cleaning up expired cmds, because if
129           cmd has been completed then accessing se_cmd is off limits */
130        DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
131
132        unsigned long deadline;
133
134#define TCMU_CMD_BIT_EXPIRED 0
135        unsigned long flags;
136};
137
138static struct kmem_cache *tcmu_cmd_cache;
139
140/* multicast group */
141enum tcmu_multicast_groups {
142        TCMU_MCGRP_CONFIG,
143};
144
145static const struct genl_multicast_group tcmu_mcgrps[] = {
146        [TCMU_MCGRP_CONFIG] = { .name = "config", },
147};
148
149/* Our generic netlink family */
150static struct genl_family tcmu_genl_family = {
151        .id = GENL_ID_GENERATE,
152        .hdrsize = 0,
153        .name = "TCM-USER",
154        .version = 1,
155        .maxattr = TCMU_ATTR_MAX,
156        .mcgrps = tcmu_mcgrps,
157        .n_mcgrps = ARRAY_SIZE(tcmu_mcgrps),
158        .netnsok = true,
159};
160
161static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
162{
163        struct se_device *se_dev = se_cmd->se_dev;
164        struct tcmu_dev *udev = TCMU_DEV(se_dev);
165        struct tcmu_cmd *tcmu_cmd;
166        int cmd_id;
167
168        tcmu_cmd = kmem_cache_zalloc(tcmu_cmd_cache, GFP_KERNEL);
169        if (!tcmu_cmd)
170                return NULL;
171
172        tcmu_cmd->se_cmd = se_cmd;
173        tcmu_cmd->tcmu_dev = udev;
174        tcmu_cmd->deadline = jiffies + msecs_to_jiffies(TCMU_TIME_OUT);
175
176        idr_preload(GFP_KERNEL);
177        spin_lock_irq(&udev->commands_lock);
178        cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 0,
179                USHRT_MAX, GFP_NOWAIT);
180        spin_unlock_irq(&udev->commands_lock);
181        idr_preload_end();
182
183        if (cmd_id < 0) {
184                kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
185                return NULL;
186        }
187        tcmu_cmd->cmd_id = cmd_id;
188
189        return tcmu_cmd;
190}
191
192static inline void tcmu_flush_dcache_range(void *vaddr, size_t size)
193{
194        unsigned long offset = offset_in_page(vaddr);
195
196        size = round_up(size+offset, PAGE_SIZE);
197        vaddr -= offset;
198
199        while (size) {
200                flush_dcache_page(virt_to_page(vaddr));
201                size -= PAGE_SIZE;
202        }
203}
204
205/*
206 * Some ring helper functions. We don't assume size is a power of 2 so
207 * we can't use circ_buf.h.
208 */
209static inline size_t spc_used(size_t head, size_t tail, size_t size)
210{
211        int diff = head - tail;
212
213        if (diff >= 0)
214                return diff;
215        else
216                return size + diff;
217}
218
219static inline size_t spc_free(size_t head, size_t tail, size_t size)
220{
221        /* Keep 1 byte unused or we can't tell full from empty */
222        return (size - spc_used(head, tail, size) - 1);
223}
224
225static inline size_t head_to_end(size_t head, size_t size)
226{
227        return size - head;
228}
229
230static inline void new_iov(struct iovec **iov, int *iov_cnt,
231                           struct tcmu_dev *udev)
232{
233        struct iovec *iovec;
234
235        if (*iov_cnt != 0)
236                (*iov)++;
237        (*iov_cnt)++;
238
239        iovec = *iov;
240        memset(iovec, 0, sizeof(struct iovec));
241}
242
243#define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
244
245/* offset is relative to mb_addr */
246static inline size_t get_block_offset(struct tcmu_dev *dev,
247                int block, int remaining)
248{
249        return dev->data_off + block * DATA_BLOCK_SIZE +
250                DATA_BLOCK_SIZE - remaining;
251}
252
253static inline size_t iov_tail(struct tcmu_dev *udev, struct iovec *iov)
254{
255        return (size_t)iov->iov_base + iov->iov_len;
256}
257
258static void alloc_and_scatter_data_area(struct tcmu_dev *udev,
259        struct scatterlist *data_sg, unsigned int data_nents,
260        struct iovec **iov, int *iov_cnt, bool copy_data)
261{
262        int i, block;
263        int block_remaining = 0;
264        void *from, *to;
265        size_t copy_bytes, to_offset;
266        struct scatterlist *sg;
267
268        for_each_sg(data_sg, sg, data_nents, i) {
269                int sg_remaining = sg->length;
270                from = kmap_atomic(sg_page(sg)) + sg->offset;
271                while (sg_remaining > 0) {
272                        if (block_remaining == 0) {
273                                block = find_first_zero_bit(udev->data_bitmap,
274                                                DATA_BLOCK_BITS);
275                                block_remaining = DATA_BLOCK_SIZE;
276                                set_bit(block, udev->data_bitmap);
277                        }
278                        copy_bytes = min_t(size_t, sg_remaining,
279                                        block_remaining);
280                        to_offset = get_block_offset(udev, block,
281                                        block_remaining);
282                        to = (void *)udev->mb_addr + to_offset;
283                        if (*iov_cnt != 0 &&
284                            to_offset == iov_tail(udev, *iov)) {
285                                (*iov)->iov_len += copy_bytes;
286                        } else {
287                                new_iov(iov, iov_cnt, udev);
288                                (*iov)->iov_base = (void __user *) to_offset;
289                                (*iov)->iov_len = copy_bytes;
290                        }
291                        if (copy_data) {
292                                memcpy(to, from + sg->length - sg_remaining,
293                                        copy_bytes);
294                                tcmu_flush_dcache_range(to, copy_bytes);
295                        }
296                        sg_remaining -= copy_bytes;
297                        block_remaining -= copy_bytes;
298                }
299                kunmap_atomic(from - sg->offset);
300        }
301}
302
303static void free_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd)
304{
305        bitmap_xor(udev->data_bitmap, udev->data_bitmap, cmd->data_bitmap,
306                   DATA_BLOCK_BITS);
307}
308
309static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
310                             bool bidi)
311{
312        struct se_cmd *se_cmd = cmd->se_cmd;
313        int i, block;
314        int block_remaining = 0;
315        void *from, *to;
316        size_t copy_bytes, from_offset;
317        struct scatterlist *sg, *data_sg;
318        unsigned int data_nents;
319        DECLARE_BITMAP(bitmap, DATA_BLOCK_BITS);
320
321        bitmap_copy(bitmap, cmd->data_bitmap, DATA_BLOCK_BITS);
322
323        if (!bidi) {
324                data_sg = se_cmd->t_data_sg;
325                data_nents = se_cmd->t_data_nents;
326        } else {
327                uint32_t count;
328
329                /*
330                 * For bidi case, the first count blocks are for Data-Out
331                 * buffer blocks, and before gathering the Data-In buffer
332                 * the Data-Out buffer blocks should be discarded.
333                 */
334                count = DIV_ROUND_UP(se_cmd->data_length, DATA_BLOCK_SIZE);
335                while (count--) {
336                        block = find_first_bit(bitmap, DATA_BLOCK_BITS);
337                        clear_bit(block, bitmap);
338                }
339
340                data_sg = se_cmd->t_bidi_data_sg;
341                data_nents = se_cmd->t_bidi_data_nents;
342        }
343
344        for_each_sg(data_sg, sg, data_nents, i) {
345                int sg_remaining = sg->length;
346                to = kmap_atomic(sg_page(sg)) + sg->offset;
347                while (sg_remaining > 0) {
348                        if (block_remaining == 0) {
349                                block = find_first_bit(bitmap,
350                                                DATA_BLOCK_BITS);
351                                block_remaining = DATA_BLOCK_SIZE;
352                                clear_bit(block, bitmap);
353                        }
354                        copy_bytes = min_t(size_t, sg_remaining,
355                                        block_remaining);
356                        from_offset = get_block_offset(udev, block,
357                                        block_remaining);
358                        from = (void *) udev->mb_addr + from_offset;
359                        tcmu_flush_dcache_range(from, copy_bytes);
360                        memcpy(to + sg->length - sg_remaining, from,
361                                        copy_bytes);
362
363                        sg_remaining -= copy_bytes;
364                        block_remaining -= copy_bytes;
365                }
366                kunmap_atomic(to - sg->offset);
367        }
368}
369
370static inline size_t spc_bitmap_free(unsigned long *bitmap)
371{
372        return DATA_BLOCK_SIZE * (DATA_BLOCK_BITS -
373                        bitmap_weight(bitmap, DATA_BLOCK_BITS));
374}
375
376/*
377 * We can't queue a command until we have space available on the cmd ring *and*
378 * space available on the data area.
379 *
380 * Called with ring lock held.
381 */
382static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t data_needed)
383{
384        struct tcmu_mailbox *mb = udev->mb_addr;
385        size_t space, cmd_needed;
386        u32 cmd_head;
387
388        tcmu_flush_dcache_range(mb, sizeof(*mb));
389
390        cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
391
392        /*
393         * If cmd end-of-ring space is too small then we need space for a NOP plus
394         * original cmd - cmds are internally contiguous.
395         */
396        if (head_to_end(cmd_head, udev->cmdr_size) >= cmd_size)
397                cmd_needed = cmd_size;
398        else
399                cmd_needed = cmd_size + head_to_end(cmd_head, udev->cmdr_size);
400
401        space = spc_free(cmd_head, udev->cmdr_last_cleaned, udev->cmdr_size);
402        if (space < cmd_needed) {
403                pr_debug("no cmd space: %u %u %u\n", cmd_head,
404                       udev->cmdr_last_cleaned, udev->cmdr_size);
405                return false;
406        }
407
408        space = spc_bitmap_free(udev->data_bitmap);
409        if (space < data_needed) {
410                pr_debug("no data space: only %zu available, but ask for %zu\n",
411                                space, data_needed);
412                return false;
413        }
414
415        return true;
416}
417
418static inline size_t tcmu_cmd_get_data_length(struct tcmu_cmd *tcmu_cmd)
419{
420        struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
421        size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
422
423        if (se_cmd->se_cmd_flags & SCF_BIDI) {
424                BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
425                data_length += round_up(se_cmd->t_bidi_data_sg->length,
426                                DATA_BLOCK_SIZE);
427        }
428
429        return data_length;
430}
431
432static inline uint32_t tcmu_cmd_get_block_cnt(struct tcmu_cmd *tcmu_cmd)
433{
434        size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
435
436        return data_length / DATA_BLOCK_SIZE;
437}
438
439static sense_reason_t
440tcmu_queue_cmd_ring(struct tcmu_cmd *tcmu_cmd)
441{
442        struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
443        struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
444        size_t base_command_size, command_size;
445        struct tcmu_mailbox *mb;
446        struct tcmu_cmd_entry *entry;
447        struct iovec *iov;
448        int iov_cnt;
449        uint32_t cmd_head;
450        uint64_t cdb_off;
451        bool copy_to_data_area;
452        size_t data_length = tcmu_cmd_get_data_length(tcmu_cmd);
453        DECLARE_BITMAP(old_bitmap, DATA_BLOCK_BITS);
454
455        if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
456                return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
457
458        /*
459         * Must be a certain minimum size for response sense info, but
460         * also may be larger if the iov array is large.
461         *
462         * We prepare way too many iovs for potential uses here, because it's
463         * expensive to tell how many regions are freed in the bitmap
464        */
465        base_command_size = max(offsetof(struct tcmu_cmd_entry,
466                                req.iov[tcmu_cmd_get_block_cnt(tcmu_cmd)]),
467                                sizeof(struct tcmu_cmd_entry));
468        command_size = base_command_size
469                + round_up(scsi_command_size(se_cmd->t_task_cdb), TCMU_OP_ALIGN_SIZE);
470
471        WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
472
473        spin_lock_irq(&udev->cmdr_lock);
474
475        mb = udev->mb_addr;
476        cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
477        if ((command_size > (udev->cmdr_size / 2)) ||
478            data_length > udev->data_size) {
479                pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
480                        "cmd ring/data area\n", command_size, data_length,
481                        udev->cmdr_size, udev->data_size);
482                spin_unlock_irq(&udev->cmdr_lock);
483                return TCM_INVALID_CDB_FIELD;
484        }
485
486        while (!is_ring_space_avail(udev, command_size, data_length)) {
487                int ret;
488                DEFINE_WAIT(__wait);
489
490                prepare_to_wait(&udev->wait_cmdr, &__wait, TASK_INTERRUPTIBLE);
491
492                pr_debug("sleeping for ring space\n");
493                spin_unlock_irq(&udev->cmdr_lock);
494                ret = schedule_timeout(msecs_to_jiffies(TCMU_TIME_OUT));
495                finish_wait(&udev->wait_cmdr, &__wait);
496                if (!ret) {
497                        pr_warn("tcmu: command timed out\n");
498                        return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
499                }
500
501                spin_lock_irq(&udev->cmdr_lock);
502
503                /* We dropped cmdr_lock, cmd_head is stale */
504                cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
505        }
506
507        /* Insert a PAD if end-of-ring space is too small */
508        if (head_to_end(cmd_head, udev->cmdr_size) < command_size) {
509                size_t pad_size = head_to_end(cmd_head, udev->cmdr_size);
510
511                entry = (void *) mb + CMDR_OFF + cmd_head;
512                tcmu_flush_dcache_range(entry, sizeof(*entry));
513                tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_PAD);
514                tcmu_hdr_set_len(&entry->hdr.len_op, pad_size);
515                entry->hdr.cmd_id = 0; /* not used for PAD */
516                entry->hdr.kflags = 0;
517                entry->hdr.uflags = 0;
518
519                UPDATE_HEAD(mb->cmd_head, pad_size, udev->cmdr_size);
520
521                cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
522                WARN_ON(cmd_head != 0);
523        }
524
525        entry = (void *) mb + CMDR_OFF + cmd_head;
526        tcmu_flush_dcache_range(entry, sizeof(*entry));
527        tcmu_hdr_set_op(&entry->hdr.len_op, TCMU_OP_CMD);
528        tcmu_hdr_set_len(&entry->hdr.len_op, command_size);
529        entry->hdr.cmd_id = tcmu_cmd->cmd_id;
530        entry->hdr.kflags = 0;
531        entry->hdr.uflags = 0;
532
533        bitmap_copy(old_bitmap, udev->data_bitmap, DATA_BLOCK_BITS);
534
535        /* Handle allocating space from the data area */
536        iov = &entry->req.iov[0];
537        iov_cnt = 0;
538        copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
539                || se_cmd->se_cmd_flags & SCF_BIDI);
540        alloc_and_scatter_data_area(udev, se_cmd->t_data_sg,
541                se_cmd->t_data_nents, &iov, &iov_cnt, copy_to_data_area);
542        entry->req.iov_cnt = iov_cnt;
543        entry->req.iov_dif_cnt = 0;
544
545        /* Handle BIDI commands */
546        if (se_cmd->se_cmd_flags & SCF_BIDI) {
547                iov_cnt = 0;
548                iov++;
549                alloc_and_scatter_data_area(udev, se_cmd->t_bidi_data_sg,
550                                se_cmd->t_bidi_data_nents, &iov, &iov_cnt,
551                                false);
552                entry->req.iov_bidi_cnt = iov_cnt;
553        }
554        /* cmd's data_bitmap is what changed in process */
555        bitmap_xor(tcmu_cmd->data_bitmap, old_bitmap, udev->data_bitmap,
556                        DATA_BLOCK_BITS);
557
558        /* All offsets relative to mb_addr, not start of entry! */
559        cdb_off = CMDR_OFF + cmd_head + base_command_size;
560        memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
561        entry->req.cdb_off = cdb_off;
562        tcmu_flush_dcache_range(entry, sizeof(*entry));
563
564        UPDATE_HEAD(mb->cmd_head, command_size, udev->cmdr_size);
565        tcmu_flush_dcache_range(mb, sizeof(*mb));
566
567        spin_unlock_irq(&udev->cmdr_lock);
568
569        /* TODO: only if FLUSH and FUA? */
570        uio_event_notify(&udev->uio_info);
571
572        mod_timer(&udev->timeout,
573                round_jiffies_up(jiffies + msecs_to_jiffies(TCMU_TIME_OUT)));
574
575        return TCM_NO_SENSE;
576}
577
578static sense_reason_t
579tcmu_queue_cmd(struct se_cmd *se_cmd)
580{
581        struct se_device *se_dev = se_cmd->se_dev;
582        struct tcmu_dev *udev = TCMU_DEV(se_dev);
583        struct tcmu_cmd *tcmu_cmd;
584        int ret;
585
586        tcmu_cmd = tcmu_alloc_cmd(se_cmd);
587        if (!tcmu_cmd)
588                return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
589
590        ret = tcmu_queue_cmd_ring(tcmu_cmd);
591        if (ret != TCM_NO_SENSE) {
592                pr_err("TCMU: Could not queue command\n");
593                spin_lock_irq(&udev->commands_lock);
594                idr_remove(&udev->commands, tcmu_cmd->cmd_id);
595                spin_unlock_irq(&udev->commands_lock);
596
597                kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
598        }
599
600        return ret;
601}
602
603static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *entry)
604{
605        struct se_cmd *se_cmd = cmd->se_cmd;
606        struct tcmu_dev *udev = cmd->tcmu_dev;
607
608        if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
609                /*
610                 * cmd has been completed already from timeout, just reclaim
611                 * data area space and free cmd
612                 */
613                free_data_area(udev, cmd);
614
615                kmem_cache_free(tcmu_cmd_cache, cmd);
616                return;
617        }
618
619        if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
620                free_data_area(udev, cmd);
621                pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
622                        cmd->se_cmd);
623                entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
624        } else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
625                memcpy(se_cmd->sense_buffer, entry->rsp.sense_buffer,
626                               se_cmd->scsi_sense_length);
627                free_data_area(udev, cmd);
628        } else if (se_cmd->se_cmd_flags & SCF_BIDI) {
629                /* Get Data-In buffer before clean up */
630                gather_data_area(udev, cmd, true);
631                free_data_area(udev, cmd);
632        } else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
633                gather_data_area(udev, cmd, false);
634                free_data_area(udev, cmd);
635        } else if (se_cmd->data_direction == DMA_TO_DEVICE) {
636                free_data_area(udev, cmd);
637        } else if (se_cmd->data_direction != DMA_NONE) {
638                pr_warn("TCMU: data direction was %d!\n",
639                        se_cmd->data_direction);
640        }
641
642        target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
643        cmd->se_cmd = NULL;
644
645        kmem_cache_free(tcmu_cmd_cache, cmd);
646}
647
648static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
649{
650        struct tcmu_mailbox *mb;
651        unsigned long flags;
652        int handled = 0;
653
654        if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags)) {
655                pr_err("ring broken, not handling completions\n");
656                return 0;
657        }
658
659        spin_lock_irqsave(&udev->cmdr_lock, flags);
660
661        mb = udev->mb_addr;
662        tcmu_flush_dcache_range(mb, sizeof(*mb));
663
664        while (udev->cmdr_last_cleaned != ACCESS_ONCE(mb->cmd_tail)) {
665
666                struct tcmu_cmd_entry *entry = (void *) mb + CMDR_OFF + udev->cmdr_last_cleaned;
667                struct tcmu_cmd *cmd;
668
669                tcmu_flush_dcache_range(entry, sizeof(*entry));
670
671                if (tcmu_hdr_get_op(entry->hdr.len_op) == TCMU_OP_PAD) {
672                        UPDATE_HEAD(udev->cmdr_last_cleaned,
673                                    tcmu_hdr_get_len(entry->hdr.len_op),
674                                    udev->cmdr_size);
675                        continue;
676                }
677                WARN_ON(tcmu_hdr_get_op(entry->hdr.len_op) != TCMU_OP_CMD);
678
679                spin_lock(&udev->commands_lock);
680                cmd = idr_find(&udev->commands, entry->hdr.cmd_id);
681                if (cmd)
682                        idr_remove(&udev->commands, cmd->cmd_id);
683                spin_unlock(&udev->commands_lock);
684
685                if (!cmd) {
686                        pr_err("cmd_id not found, ring is broken\n");
687                        set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
688                        break;
689                }
690
691                tcmu_handle_completion(cmd, entry);
692
693                UPDATE_HEAD(udev->cmdr_last_cleaned,
694                            tcmu_hdr_get_len(entry->hdr.len_op),
695                            udev->cmdr_size);
696
697                handled++;
698        }
699
700        if (mb->cmd_tail == mb->cmd_head)
701                del_timer(&udev->timeout); /* no more pending cmds */
702
703        spin_unlock_irqrestore(&udev->cmdr_lock, flags);
704
705        wake_up(&udev->wait_cmdr);
706
707        return handled;
708}
709
710static int tcmu_check_expired_cmd(int id, void *p, void *data)
711{
712        struct tcmu_cmd *cmd = p;
713
714        if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags))
715                return 0;
716
717        if (!time_after(jiffies, cmd->deadline))
718                return 0;
719
720        set_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags);
721        target_complete_cmd(cmd->se_cmd, SAM_STAT_CHECK_CONDITION);
722        cmd->se_cmd = NULL;
723
724        return 0;
725}
726
727static void tcmu_device_timedout(unsigned long data)
728{
729        struct tcmu_dev *udev = (struct tcmu_dev *)data;
730        unsigned long flags;
731        int handled;
732
733        handled = tcmu_handle_completions(udev);
734
735        pr_warn("%d completions handled from timeout\n", handled);
736
737        spin_lock_irqsave(&udev->commands_lock, flags);
738        idr_for_each(&udev->commands, tcmu_check_expired_cmd, NULL);
739        spin_unlock_irqrestore(&udev->commands_lock, flags);
740
741        /*
742         * We don't need to wakeup threads on wait_cmdr since they have their
743         * own timeout.
744         */
745}
746
747static int tcmu_attach_hba(struct se_hba *hba, u32 host_id)
748{
749        struct tcmu_hba *tcmu_hba;
750
751        tcmu_hba = kzalloc(sizeof(struct tcmu_hba), GFP_KERNEL);
752        if (!tcmu_hba)
753                return -ENOMEM;
754
755        tcmu_hba->host_id = host_id;
756        hba->hba_ptr = tcmu_hba;
757
758        return 0;
759}
760
761static void tcmu_detach_hba(struct se_hba *hba)
762{
763        kfree(hba->hba_ptr);
764        hba->hba_ptr = NULL;
765}
766
767static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
768{
769        struct tcmu_dev *udev;
770
771        udev = kzalloc(sizeof(struct tcmu_dev), GFP_KERNEL);
772        if (!udev)
773                return NULL;
774
775        udev->name = kstrdup(name, GFP_KERNEL);
776        if (!udev->name) {
777                kfree(udev);
778                return NULL;
779        }
780
781        udev->hba = hba;
782
783        init_waitqueue_head(&udev->wait_cmdr);
784        spin_lock_init(&udev->cmdr_lock);
785
786        idr_init(&udev->commands);
787        spin_lock_init(&udev->commands_lock);
788
789        setup_timer(&udev->timeout, tcmu_device_timedout,
790                (unsigned long)udev);
791
792        return &udev->se_dev;
793}
794
795static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
796{
797        struct tcmu_dev *tcmu_dev = container_of(info, struct tcmu_dev, uio_info);
798
799        tcmu_handle_completions(tcmu_dev);
800
801        return 0;
802}
803
804/*
805 * mmap code from uio.c. Copied here because we want to hook mmap()
806 * and this stuff must come along.
807 */
808static int tcmu_find_mem_index(struct vm_area_struct *vma)
809{
810        struct tcmu_dev *udev = vma->vm_private_data;
811        struct uio_info *info = &udev->uio_info;
812
813        if (vma->vm_pgoff < MAX_UIO_MAPS) {
814                if (info->mem[vma->vm_pgoff].size == 0)
815                        return -1;
816                return (int)vma->vm_pgoff;
817        }
818        return -1;
819}
820
821static int tcmu_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
822{
823        struct tcmu_dev *udev = vma->vm_private_data;
824        struct uio_info *info = &udev->uio_info;
825        struct page *page;
826        unsigned long offset;
827        void *addr;
828
829        int mi = tcmu_find_mem_index(vma);
830        if (mi < 0)
831                return VM_FAULT_SIGBUS;
832
833        /*
834         * We need to subtract mi because userspace uses offset = N*PAGE_SIZE
835         * to use mem[N].
836         */
837        offset = (vmf->pgoff - mi) << PAGE_SHIFT;
838
839        addr = (void *)(unsigned long)info->mem[mi].addr + offset;
840        if (info->mem[mi].memtype == UIO_MEM_LOGICAL)
841                page = virt_to_page(addr);
842        else
843                page = vmalloc_to_page(addr);
844        get_page(page);
845        vmf->page = page;
846        return 0;
847}
848
849static const struct vm_operations_struct tcmu_vm_ops = {
850        .fault = tcmu_vma_fault,
851};
852
853static int tcmu_mmap(struct uio_info *info, struct vm_area_struct *vma)
854{
855        struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
856
857        vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
858        vma->vm_ops = &tcmu_vm_ops;
859
860        vma->vm_private_data = udev;
861
862        /* Ensure the mmap is exactly the right size */
863        if (vma_pages(vma) != (TCMU_RING_SIZE >> PAGE_SHIFT))
864                return -EINVAL;
865
866        return 0;
867}
868
869static int tcmu_open(struct uio_info *info, struct inode *inode)
870{
871        struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
872
873        /* O_EXCL not supported for char devs, so fake it? */
874        if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
875                return -EBUSY;
876
877        pr_debug("open\n");
878
879        return 0;
880}
881
882static int tcmu_release(struct uio_info *info, struct inode *inode)
883{
884        struct tcmu_dev *udev = container_of(info, struct tcmu_dev, uio_info);
885
886        clear_bit(TCMU_DEV_BIT_OPEN, &udev->flags);
887
888        pr_debug("close\n");
889
890        return 0;
891}
892
893static int tcmu_netlink_event(enum tcmu_genl_cmd cmd, const char *name, int minor)
894{
895        struct sk_buff *skb;
896        void *msg_header;
897        int ret = -ENOMEM;
898
899        skb = genlmsg_new(NLMSG_GOODSIZE, GFP_KERNEL);
900        if (!skb)
901                return ret;
902
903        msg_header = genlmsg_put(skb, 0, 0, &tcmu_genl_family, 0, cmd);
904        if (!msg_header)
905                goto free_skb;
906
907        ret = nla_put_string(skb, TCMU_ATTR_DEVICE, name);
908        if (ret < 0)
909                goto free_skb;
910
911        ret = nla_put_u32(skb, TCMU_ATTR_MINOR, minor);
912        if (ret < 0)
913                goto free_skb;
914
915        genlmsg_end(skb, msg_header);
916
917        ret = genlmsg_multicast_allns(&tcmu_genl_family, skb, 0,
918                                TCMU_MCGRP_CONFIG, GFP_KERNEL);
919
920        /* We don't care if no one is listening */
921        if (ret == -ESRCH)
922                ret = 0;
923
924        return ret;
925free_skb:
926        nlmsg_free(skb);
927        return ret;
928}
929
930static int tcmu_configure_device(struct se_device *dev)
931{
932        struct tcmu_dev *udev = TCMU_DEV(dev);
933        struct tcmu_hba *hba = udev->hba->hba_ptr;
934        struct uio_info *info;
935        struct tcmu_mailbox *mb;
936        size_t size;
937        size_t used;
938        int ret = 0;
939        char *str;
940
941        info = &udev->uio_info;
942
943        size = snprintf(NULL, 0, "tcm-user/%u/%s/%s", hba->host_id, udev->name,
944                        udev->dev_config);
945        size += 1; /* for \0 */
946        str = kmalloc(size, GFP_KERNEL);
947        if (!str)
948                return -ENOMEM;
949
950        used = snprintf(str, size, "tcm-user/%u/%s", hba->host_id, udev->name);
951
952        if (udev->dev_config[0])
953                snprintf(str + used, size - used, "/%s", udev->dev_config);
954
955        info->name = str;
956
957        udev->mb_addr = vzalloc(TCMU_RING_SIZE);
958        if (!udev->mb_addr) {
959                ret = -ENOMEM;
960                goto err_vzalloc;
961        }
962
963        /* mailbox fits in first part of CMDR space */
964        udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
965        udev->data_off = CMDR_SIZE;
966        udev->data_size = TCMU_RING_SIZE - CMDR_SIZE;
967
968        mb = udev->mb_addr;
969        mb->version = TCMU_MAILBOX_VERSION;
970        mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC;
971        mb->cmdr_off = CMDR_OFF;
972        mb->cmdr_size = udev->cmdr_size;
973
974        WARN_ON(!PAGE_ALIGNED(udev->data_off));
975        WARN_ON(udev->data_size % PAGE_SIZE);
976        WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
977
978        info->version = __stringify(TCMU_MAILBOX_VERSION);
979
980        info->mem[0].name = "tcm-user command & data buffer";
981        info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
982        info->mem[0].size = TCMU_RING_SIZE;
983        info->mem[0].memtype = UIO_MEM_VIRTUAL;
984
985        info->irqcontrol = tcmu_irqcontrol;
986        info->irq = UIO_IRQ_CUSTOM;
987
988        info->mmap = tcmu_mmap;
989        info->open = tcmu_open;
990        info->release = tcmu_release;
991
992        ret = uio_register_device(tcmu_root_device, info);
993        if (ret)
994                goto err_register;
995
996        /* User can set hw_block_size before enable the device */
997        if (dev->dev_attrib.hw_block_size == 0)
998                dev->dev_attrib.hw_block_size = 512;
999        /* Other attributes can be configured in userspace */
1000        dev->dev_attrib.hw_max_sectors = 128;
1001        dev->dev_attrib.hw_queue_depth = 128;
1002
1003        ret = tcmu_netlink_event(TCMU_CMD_ADDED_DEVICE, udev->uio_info.name,
1004                                 udev->uio_info.uio_dev->minor);
1005        if (ret)
1006                goto err_netlink;
1007
1008        return 0;
1009
1010err_netlink:
1011        uio_unregister_device(&udev->uio_info);
1012err_register:
1013        vfree(udev->mb_addr);
1014err_vzalloc:
1015        kfree(info->name);
1016
1017        return ret;
1018}
1019
1020static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
1021{
1022        if (test_bit(TCMU_CMD_BIT_EXPIRED, &cmd->flags)) {
1023                kmem_cache_free(tcmu_cmd_cache, cmd);
1024                return 0;
1025        }
1026        return -EINVAL;
1027}
1028
1029static void tcmu_dev_call_rcu(struct rcu_head *p)
1030{
1031        struct se_device *dev = container_of(p, struct se_device, rcu_head);
1032        struct tcmu_dev *udev = TCMU_DEV(dev);
1033
1034        kfree(udev);
1035}
1036
1037static void tcmu_free_device(struct se_device *dev)
1038{
1039        struct tcmu_dev *udev = TCMU_DEV(dev);
1040        struct tcmu_cmd *cmd;
1041        bool all_expired = true;
1042        int i;
1043
1044        del_timer_sync(&udev->timeout);
1045
1046        vfree(udev->mb_addr);
1047
1048        /* Upper layer should drain all requests before calling this */
1049        spin_lock_irq(&udev->commands_lock);
1050        idr_for_each_entry(&udev->commands, cmd, i) {
1051                if (tcmu_check_and_free_pending_cmd(cmd) != 0)
1052                        all_expired = false;
1053        }
1054        idr_destroy(&udev->commands);
1055        spin_unlock_irq(&udev->commands_lock);
1056        WARN_ON(!all_expired);
1057
1058        /* Device was configured */
1059        if (udev->uio_info.uio_dev) {
1060                tcmu_netlink_event(TCMU_CMD_REMOVED_DEVICE, udev->uio_info.name,
1061                                   udev->uio_info.uio_dev->minor);
1062
1063                uio_unregister_device(&udev->uio_info);
1064                kfree(udev->uio_info.name);
1065                kfree(udev->name);
1066        }
1067        call_rcu(&dev->rcu_head, tcmu_dev_call_rcu);
1068}
1069
1070enum {
1071        Opt_dev_config, Opt_dev_size, Opt_hw_block_size, Opt_err,
1072};
1073
1074static match_table_t tokens = {
1075        {Opt_dev_config, "dev_config=%s"},
1076        {Opt_dev_size, "dev_size=%u"},
1077        {Opt_hw_block_size, "hw_block_size=%u"},
1078        {Opt_err, NULL}
1079};
1080
1081static ssize_t tcmu_set_configfs_dev_params(struct se_device *dev,
1082                const char *page, ssize_t count)
1083{
1084        struct tcmu_dev *udev = TCMU_DEV(dev);
1085        char *orig, *ptr, *opts, *arg_p;
1086        substring_t args[MAX_OPT_ARGS];
1087        int ret = 0, token;
1088        unsigned long tmp_ul;
1089
1090        opts = kstrdup(page, GFP_KERNEL);
1091        if (!opts)
1092                return -ENOMEM;
1093
1094        orig = opts;
1095
1096        while ((ptr = strsep(&opts, ",\n")) != NULL) {
1097                if (!*ptr)
1098                        continue;
1099
1100                token = match_token(ptr, tokens, args);
1101                switch (token) {
1102                case Opt_dev_config:
1103                        if (match_strlcpy(udev->dev_config, &args[0],
1104                                          TCMU_CONFIG_LEN) == 0) {
1105                                ret = -EINVAL;
1106                                break;
1107                        }
1108                        pr_debug("TCMU: Referencing Path: %s\n", udev->dev_config);
1109                        break;
1110                case Opt_dev_size:
1111                        arg_p = match_strdup(&args[0]);
1112                        if (!arg_p) {
1113                                ret = -ENOMEM;
1114                                break;
1115                        }
1116                        ret = kstrtoul(arg_p, 0, (unsigned long *) &udev->dev_size);
1117                        kfree(arg_p);
1118                        if (ret < 0)
1119                                pr_err("kstrtoul() failed for dev_size=\n");
1120                        break;
1121                case Opt_hw_block_size:
1122                        arg_p = match_strdup(&args[0]);
1123                        if (!arg_p) {
1124                                ret = -ENOMEM;
1125                                break;
1126                        }
1127                        ret = kstrtoul(arg_p, 0, &tmp_ul);
1128                        kfree(arg_p);
1129                        if (ret < 0) {
1130                                pr_err("kstrtoul() failed for hw_block_size=\n");
1131                                break;
1132                        }
1133                        if (!tmp_ul) {
1134                                pr_err("hw_block_size must be nonzero\n");
1135                                break;
1136                        }
1137                        dev->dev_attrib.hw_block_size = tmp_ul;
1138                        break;
1139                default:
1140                        break;
1141                }
1142        }
1143
1144        kfree(orig);
1145        return (!ret) ? count : ret;
1146}
1147
1148static ssize_t tcmu_show_configfs_dev_params(struct se_device *dev, char *b)
1149{
1150        struct tcmu_dev *udev = TCMU_DEV(dev);
1151        ssize_t bl = 0;
1152
1153        bl = sprintf(b + bl, "Config: %s ",
1154                     udev->dev_config[0] ? udev->dev_config : "NULL");
1155        bl += sprintf(b + bl, "Size: %zu\n", udev->dev_size);
1156
1157        return bl;
1158}
1159
1160static sector_t tcmu_get_blocks(struct se_device *dev)
1161{
1162        struct tcmu_dev *udev = TCMU_DEV(dev);
1163
1164        return div_u64(udev->dev_size - dev->dev_attrib.block_size,
1165                       dev->dev_attrib.block_size);
1166}
1167
1168static sense_reason_t
1169tcmu_parse_cdb(struct se_cmd *cmd)
1170{
1171        return passthrough_parse_cdb(cmd, tcmu_queue_cmd);
1172}
1173
1174static const struct target_backend_ops tcmu_ops = {
1175        .name                   = "user",
1176        .owner                  = THIS_MODULE,
1177        .transport_flags        = TRANSPORT_FLAG_PASSTHROUGH,
1178        .attach_hba             = tcmu_attach_hba,
1179        .detach_hba             = tcmu_detach_hba,
1180        .alloc_device           = tcmu_alloc_device,
1181        .configure_device       = tcmu_configure_device,
1182        .free_device            = tcmu_free_device,
1183        .parse_cdb              = tcmu_parse_cdb,
1184        .set_configfs_dev_params = tcmu_set_configfs_dev_params,
1185        .show_configfs_dev_params = tcmu_show_configfs_dev_params,
1186        .get_device_type        = sbc_get_device_type,
1187        .get_blocks             = tcmu_get_blocks,
1188        .tb_dev_attrib_attrs    = passthrough_attrib_attrs,
1189};
1190
1191static int __init tcmu_module_init(void)
1192{
1193        int ret;
1194
1195        BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
1196
1197        tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
1198                                sizeof(struct tcmu_cmd),
1199                                __alignof__(struct tcmu_cmd),
1200                                0, NULL);
1201        if (!tcmu_cmd_cache)
1202                return -ENOMEM;
1203
1204        tcmu_root_device = root_device_register("tcm_user");
1205        if (IS_ERR(tcmu_root_device)) {
1206                ret = PTR_ERR(tcmu_root_device);
1207                goto out_free_cache;
1208        }
1209
1210        ret = genl_register_family(&tcmu_genl_family);
1211        if (ret < 0) {
1212                goto out_unreg_device;
1213        }
1214
1215        ret = transport_backend_register(&tcmu_ops);
1216        if (ret)
1217                goto out_unreg_genl;
1218
1219        return 0;
1220
1221out_unreg_genl:
1222        genl_unregister_family(&tcmu_genl_family);
1223out_unreg_device:
1224        root_device_unregister(tcmu_root_device);
1225out_free_cache:
1226        kmem_cache_destroy(tcmu_cmd_cache);
1227
1228        return ret;
1229}
1230
1231static void __exit tcmu_module_exit(void)
1232{
1233        target_backend_unregister(&tcmu_ops);
1234        genl_unregister_family(&tcmu_genl_family);
1235        root_device_unregister(tcmu_root_device);
1236        kmem_cache_destroy(tcmu_cmd_cache);
1237}
1238
1239MODULE_DESCRIPTION("TCM USER subsystem plugin");
1240MODULE_AUTHOR("Shaohua Li <shli@kernel.org>");
1241MODULE_AUTHOR("Andy Grover <agrover@redhat.com>");
1242MODULE_LICENSE("GPL");
1243
1244module_init(tcmu_module_init);
1245module_exit(tcmu_module_exit);
Note: See TracBrowser for help on using the repository browser.