Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
[linux-2.6-microblaze.git] / drivers / scsi / libiscsi_tcp.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * iSCSI over TCP/IP Data-Path lib
4  *
5  * Copyright (C) 2004 Dmitry Yusupov
6  * Copyright (C) 2004 Alex Aizman
7  * Copyright (C) 2005 - 2006 Mike Christie
8  * Copyright (C) 2006 Red Hat, Inc.  All rights reserved.
9  * maintained by open-iscsi@googlegroups.com
10  *
11  * Credits:
12  *      Christoph Hellwig
13  *      FUJITA Tomonori
14  *      Arne Redlich
15  *      Zhenyu Wang
16  */
17
18 #include <crypto/hash.h>
19 #include <linux/types.h>
20 #include <linux/list.h>
21 #include <linux/inet.h>
22 #include <linux/slab.h>
23 #include <linux/file.h>
24 #include <linux/blkdev.h>
25 #include <linux/delay.h>
26 #include <linux/kfifo.h>
27 #include <linux/scatterlist.h>
28 #include <linux/module.h>
29 #include <net/tcp.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_host.h>
33 #include <scsi/scsi.h>
34 #include <scsi/scsi_transport_iscsi.h>
35 #include <trace/events/iscsi.h>
36
37 #include "iscsi_tcp.h"
38
39 MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
40               "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
41               "Alex Aizman <itn780@yahoo.com>");
42 MODULE_DESCRIPTION("iSCSI/TCP data-path");
43 MODULE_LICENSE("GPL");
44
45 static int iscsi_dbg_libtcp;
46 module_param_named(debug_libiscsi_tcp, iscsi_dbg_libtcp, int,
47                    S_IRUGO | S_IWUSR);
48 MODULE_PARM_DESC(debug_libiscsi_tcp, "Turn on debugging for libiscsi_tcp "
49                  "module. Set to 1 to turn on, and zero to turn off. Default "
50                  "is off.");
51
52 #define ISCSI_DBG_TCP(_conn, dbg_fmt, arg...)                   \
53         do {                                                    \
54                 if (iscsi_dbg_libtcp)                           \
55                         iscsi_conn_printk(KERN_INFO, _conn,     \
56                                              "%s " dbg_fmt,     \
57                                              __func__, ##arg);  \
58                 iscsi_dbg_trace(trace_iscsi_dbg_tcp,            \
59                                 &(_conn)->cls_conn->dev,        \
60                                 "%s " dbg_fmt, __func__, ##arg);\
61         } while (0);
62
63 static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
64                                    struct iscsi_segment *segment);
65
66 /*
67  * Scatterlist handling: inside the iscsi_segment, we
68  * remember an index into the scatterlist, and set data/size
69  * to the current scatterlist entry. For highmem pages, we
70  * kmap as needed.
71  *
72  * Note that the page is unmapped when we return from
73  * TCP's data_ready handler, so we may end up mapping and
74  * unmapping the same page repeatedly. The whole reason
75  * for this is that we shouldn't keep the page mapped
76  * outside the softirq.
77  */
78
79 /**
80  * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
81  * @segment: the buffer object
82  * @sg: scatterlist
83  * @offset: byte offset into that sg entry
84  *
85  * This function sets up the segment so that subsequent
86  * data is copied to the indicated sg entry, at the given
87  * offset.
88  */
89 static inline void
90 iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
91                           struct scatterlist *sg, unsigned int offset)
92 {
93         segment->sg = sg;
94         segment->sg_offset = offset;
95         segment->size = min(sg->length - offset,
96                             segment->total_size - segment->total_copied);
97         segment->data = NULL;
98 }
99
100 /**
101  * iscsi_tcp_segment_map - map the current S/G page
102  * @segment: iscsi_segment
103  * @recv: 1 if called from recv path
104  *
105  * We only need to possibly kmap data if scatter lists are being used,
106  * because the iscsi passthrough and internal IO paths will never use high
107  * mem pages.
108  */
109 static void iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
110 {
111         struct scatterlist *sg;
112
113         if (segment->data != NULL || !segment->sg)
114                 return;
115
116         sg = segment->sg;
117         BUG_ON(segment->sg_mapped);
118         BUG_ON(sg->length == 0);
119
120         /*
121          * We always map for the recv path.
122          *
123          * If the page count is greater than one it is ok to send
124          * to the network layer's zero copy send path. If not we
125          * have to go the slow sendmsg path.
126          *
127          * Same goes for slab pages: skb_can_coalesce() allows
128          * coalescing neighboring slab objects into a single frag which
129          * triggers one of hardened usercopy checks.
130          */
131         if (!recv && page_count(sg_page(sg)) >= 1 && !PageSlab(sg_page(sg)))
132                 return;
133
134         if (recv) {
135                 segment->atomic_mapped = true;
136                 segment->sg_mapped = kmap_atomic(sg_page(sg));
137         } else {
138                 segment->atomic_mapped = false;
139                 /* the xmit path can sleep with the page mapped so use kmap */
140                 segment->sg_mapped = kmap(sg_page(sg));
141         }
142
143         segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
144 }
145
146 void iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
147 {
148         if (segment->sg_mapped) {
149                 if (segment->atomic_mapped)
150                         kunmap_atomic(segment->sg_mapped);
151                 else
152                         kunmap(sg_page(segment->sg));
153                 segment->sg_mapped = NULL;
154                 segment->data = NULL;
155         }
156 }
157 EXPORT_SYMBOL_GPL(iscsi_tcp_segment_unmap);
158
159 /*
160  * Splice the digest buffer into the buffer
161  */
162 static inline void
163 iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
164 {
165         segment->data = digest;
166         segment->digest_len = ISCSI_DIGEST_SIZE;
167         segment->total_size += ISCSI_DIGEST_SIZE;
168         segment->size = ISCSI_DIGEST_SIZE;
169         segment->copied = 0;
170         segment->sg = NULL;
171         segment->hash = NULL;
172 }
173
174 /**
175  * iscsi_tcp_segment_done - check whether the segment is complete
176  * @tcp_conn: iscsi tcp connection
177  * @segment: iscsi segment to check
178  * @recv: set to one of this is called from the recv path
179  * @copied: number of bytes copied
180  *
181  * Check if we're done receiving this segment. If the receive
182  * buffer is full but we expect more data, move on to the
183  * next entry in the scatterlist.
184  *
185  * If the amount of data we received isn't a multiple of 4,
186  * we will transparently receive the pad bytes, too.
187  *
188  * This function must be re-entrant.
189  */
190 int iscsi_tcp_segment_done(struct iscsi_tcp_conn *tcp_conn,
191                            struct iscsi_segment *segment, int recv,
192                            unsigned copied)
193 {
194         struct scatterlist sg;
195         unsigned int pad;
196
197         ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copied %u %u size %u %s\n",
198                       segment->copied, copied, segment->size,
199                       recv ? "recv" : "xmit");
200         if (segment->hash && copied) {
201                 /*
202                  * If a segment is kmapd we must unmap it before sending
203                  * to the crypto layer since that will try to kmap it again.
204                  */
205                 iscsi_tcp_segment_unmap(segment);
206
207                 if (!segment->data) {
208                         sg_init_table(&sg, 1);
209                         sg_set_page(&sg, sg_page(segment->sg), copied,
210                                     segment->copied + segment->sg_offset +
211                                                         segment->sg->offset);
212                 } else
213                         sg_init_one(&sg, segment->data + segment->copied,
214                                     copied);
215                 ahash_request_set_crypt(segment->hash, &sg, NULL, copied);
216                 crypto_ahash_update(segment->hash);
217         }
218
219         segment->copied += copied;
220         if (segment->copied < segment->size) {
221                 iscsi_tcp_segment_map(segment, recv);
222                 return 0;
223         }
224
225         segment->total_copied += segment->copied;
226         segment->copied = 0;
227         segment->size = 0;
228
229         /* Unmap the current scatterlist page, if there is one. */
230         iscsi_tcp_segment_unmap(segment);
231
232         /* Do we have more scatterlist entries? */
233         ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "total copied %u total size %u\n",
234                       segment->total_copied, segment->total_size);
235         if (segment->total_copied < segment->total_size) {
236                 /* Proceed to the next entry in the scatterlist. */
237                 iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
238                                           0);
239                 iscsi_tcp_segment_map(segment, recv);
240                 BUG_ON(segment->size == 0);
241                 return 0;
242         }
243
244         /* Do we need to handle padding? */
245         if (!(tcp_conn->iscsi_conn->session->tt->caps & CAP_PADDING_OFFLOAD)) {
246                 pad = iscsi_padding(segment->total_copied);
247                 if (pad != 0) {
248                         ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
249                                       "consume %d pad bytes\n", pad);
250                         segment->total_size += pad;
251                         segment->size = pad;
252                         segment->data = segment->padbuf;
253                         return 0;
254                 }
255         }
256
257         /*
258          * Set us up for transferring the data digest. hdr digest
259          * is completely handled in hdr done function.
260          */
261         if (segment->hash) {
262                 ahash_request_set_crypt(segment->hash, NULL,
263                                         segment->digest, 0);
264                 crypto_ahash_final(segment->hash);
265                 iscsi_tcp_segment_splice_digest(segment,
266                                  recv ? segment->recv_digest : segment->digest);
267                 return 0;
268         }
269
270         return 1;
271 }
272 EXPORT_SYMBOL_GPL(iscsi_tcp_segment_done);
273
274 /**
275  * iscsi_tcp_segment_recv - copy data to segment
276  * @tcp_conn: the iSCSI TCP connection
277  * @segment: the buffer to copy to
278  * @ptr: data pointer
279  * @len: amount of data available
280  *
281  * This function copies up to @len bytes to the
282  * given buffer, and returns the number of bytes
283  * consumed, which can actually be less than @len.
284  *
285  * If hash digest is enabled, the function will update the
286  * hash while copying.
287  * Combining these two operations doesn't buy us a lot (yet),
288  * but in the future we could implement combined copy+crc,
289  * just way we do for network layer checksums.
290  */
291 static int
292 iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
293                        struct iscsi_segment *segment, const void *ptr,
294                        unsigned int len)
295 {
296         unsigned int copy = 0, copied = 0;
297
298         while (!iscsi_tcp_segment_done(tcp_conn, segment, 1, copy)) {
299                 if (copied == len) {
300                         ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
301                                       "copied %d bytes\n", len);
302                         break;
303                 }
304
305                 copy = min(len - copied, segment->size - segment->copied);
306                 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copying %d\n", copy);
307                 memcpy(segment->data + segment->copied, ptr + copied, copy);
308                 copied += copy;
309         }
310         return copied;
311 }
312
313 inline void
314 iscsi_tcp_dgst_header(struct ahash_request *hash, const void *hdr,
315                       size_t hdrlen, unsigned char digest[ISCSI_DIGEST_SIZE])
316 {
317         struct scatterlist sg;
318
319         sg_init_one(&sg, hdr, hdrlen);
320         ahash_request_set_crypt(hash, &sg, digest, hdrlen);
321         crypto_ahash_digest(hash);
322 }
323 EXPORT_SYMBOL_GPL(iscsi_tcp_dgst_header);
324
325 static inline int
326 iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
327                       struct iscsi_segment *segment)
328 {
329         if (!segment->digest_len)
330                 return 1;
331
332         if (memcmp(segment->recv_digest, segment->digest,
333                    segment->digest_len)) {
334                 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "digest mismatch\n");
335                 return 0;
336         }
337
338         return 1;
339 }
340
341 /*
342  * Helper function to set up segment buffer
343  */
344 static inline void
345 __iscsi_segment_init(struct iscsi_segment *segment, size_t size,
346                      iscsi_segment_done_fn_t *done, struct ahash_request *hash)
347 {
348         memset(segment, 0, sizeof(*segment));
349         segment->total_size = size;
350         segment->done = done;
351
352         if (hash) {
353                 segment->hash = hash;
354                 crypto_ahash_init(hash);
355         }
356 }
357
358 inline void
359 iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
360                           size_t size, iscsi_segment_done_fn_t *done,
361                           struct ahash_request *hash)
362 {
363         __iscsi_segment_init(segment, size, done, hash);
364         segment->data = data;
365         segment->size = size;
366 }
367 EXPORT_SYMBOL_GPL(iscsi_segment_init_linear);
368
369 inline int
370 iscsi_segment_seek_sg(struct iscsi_segment *segment,
371                       struct scatterlist *sg_list, unsigned int sg_count,
372                       unsigned int offset, size_t size,
373                       iscsi_segment_done_fn_t *done,
374                       struct ahash_request *hash)
375 {
376         struct scatterlist *sg;
377         unsigned int i;
378
379         __iscsi_segment_init(segment, size, done, hash);
380         for_each_sg(sg_list, sg, sg_count, i) {
381                 if (offset < sg->length) {
382                         iscsi_tcp_segment_init_sg(segment, sg, offset);
383                         return 0;
384                 }
385                 offset -= sg->length;
386         }
387
388         return ISCSI_ERR_DATA_OFFSET;
389 }
390 EXPORT_SYMBOL_GPL(iscsi_segment_seek_sg);
391
392 /**
393  * iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
394  * @tcp_conn: iscsi connection to prep for
395  *
396  * This function always passes NULL for the hash argument, because when this
397  * function is called we do not yet know the final size of the header and want
398  * to delay the digest processing until we know that.
399  */
400 void iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
401 {
402         ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
403                       "(%s)\n", tcp_conn->iscsi_conn->hdrdgst_en ?
404                       "digest enabled" : "digest disabled");
405         iscsi_segment_init_linear(&tcp_conn->in.segment,
406                                 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
407                                 iscsi_tcp_hdr_recv_done, NULL);
408 }
409 EXPORT_SYMBOL_GPL(iscsi_tcp_hdr_recv_prep);
410
411 /*
412  * Handle incoming reply to any other type of command
413  */
414 static int
415 iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
416                          struct iscsi_segment *segment)
417 {
418         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
419         int rc = 0;
420
421         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
422                 return ISCSI_ERR_DATA_DGST;
423
424         rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
425                         conn->data, tcp_conn->in.datalen);
426         if (rc)
427                 return rc;
428
429         iscsi_tcp_hdr_recv_prep(tcp_conn);
430         return 0;
431 }
432
433 static void
434 iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
435 {
436         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
437         struct ahash_request *rx_hash = NULL;
438
439         if (conn->datadgst_en &&
440             !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
441                 rx_hash = tcp_conn->rx_hash;
442
443         iscsi_segment_init_linear(&tcp_conn->in.segment,
444                                 conn->data, tcp_conn->in.datalen,
445                                 iscsi_tcp_data_recv_done, rx_hash);
446 }
447
448 /**
449  * iscsi_tcp_cleanup_task - free tcp_task resources
450  * @task: iscsi task
451  *
452  * must be called with session back_lock
453  */
454 void iscsi_tcp_cleanup_task(struct iscsi_task *task)
455 {
456         struct iscsi_tcp_task *tcp_task = task->dd_data;
457         struct iscsi_r2t_info *r2t;
458
459         /* nothing to do for mgmt */
460         if (!task->sc)
461                 return;
462
463         spin_lock_bh(&tcp_task->queue2pool);
464         /* flush task's r2t queues */
465         while (kfifo_out(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
466                 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
467                             sizeof(void*));
468                 ISCSI_DBG_TCP(task->conn, "pending r2t dropped\n");
469         }
470
471         r2t = tcp_task->r2t;
472         if (r2t != NULL) {
473                 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
474                             sizeof(void*));
475                 tcp_task->r2t = NULL;
476         }
477         spin_unlock_bh(&tcp_task->queue2pool);
478 }
479 EXPORT_SYMBOL_GPL(iscsi_tcp_cleanup_task);
480
481 /**
482  * iscsi_tcp_data_in - SCSI Data-In Response processing
483  * @conn: iscsi connection
484  * @task: scsi command task
485  */
486 static int iscsi_tcp_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
487 {
488         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
489         struct iscsi_tcp_task *tcp_task = task->dd_data;
490         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
491         int datasn = be32_to_cpu(rhdr->datasn);
492         unsigned total_in_length = task->sc->sdb.length;
493
494         /*
495          * lib iscsi will update this in the completion handling if there
496          * is status.
497          */
498         if (!(rhdr->flags & ISCSI_FLAG_DATA_STATUS))
499                 iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
500
501         if (tcp_conn->in.datalen == 0)
502                 return 0;
503
504         if (tcp_task->exp_datasn != datasn) {
505                 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->datasn(%d)"
506                               "\n", tcp_task->exp_datasn, datasn);
507                 return ISCSI_ERR_DATASN;
508         }
509
510         tcp_task->exp_datasn++;
511
512         tcp_task->data_offset = be32_to_cpu(rhdr->offset);
513         if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
514                 ISCSI_DBG_TCP(conn, "data_offset(%d) + data_len(%d) > "
515                               "total_length_in(%d)\n", tcp_task->data_offset,
516                               tcp_conn->in.datalen, total_in_length);
517                 return ISCSI_ERR_DATA_OFFSET;
518         }
519
520         conn->datain_pdus_cnt++;
521         return 0;
522 }
523
524 /**
525  * iscsi_tcp_r2t_rsp - iSCSI R2T Response processing
526  * @conn: iscsi connection
527  * @task: scsi command task
528  */
529 static int iscsi_tcp_r2t_rsp(struct iscsi_conn *conn, struct iscsi_task *task)
530 {
531         struct iscsi_session *session = conn->session;
532         struct iscsi_tcp_task *tcp_task = task->dd_data;
533         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
534         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
535         struct iscsi_r2t_info *r2t;
536         int r2tsn = be32_to_cpu(rhdr->r2tsn);
537         u32 data_length;
538         u32 data_offset;
539         int rc;
540
541         if (tcp_conn->in.datalen) {
542                 iscsi_conn_printk(KERN_ERR, conn,
543                                   "invalid R2t with datalen %d\n",
544                                   tcp_conn->in.datalen);
545                 return ISCSI_ERR_DATALEN;
546         }
547
548         if (tcp_task->exp_datasn != r2tsn){
549                 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
550                               tcp_task->exp_datasn, r2tsn);
551                 return ISCSI_ERR_R2TSN;
552         }
553
554         /* fill-in new R2T associated with the task */
555         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
556
557         if (!task->sc || session->state != ISCSI_STATE_LOGGED_IN) {
558                 iscsi_conn_printk(KERN_INFO, conn,
559                                   "dropping R2T itt %d in recovery.\n",
560                                   task->itt);
561                 return 0;
562         }
563
564         data_length = be32_to_cpu(rhdr->data_length);
565         if (data_length == 0) {
566                 iscsi_conn_printk(KERN_ERR, conn,
567                                   "invalid R2T with zero data len\n");
568                 return ISCSI_ERR_DATALEN;
569         }
570
571         if (data_length > session->max_burst)
572                 ISCSI_DBG_TCP(conn, "invalid R2T with data len %u and max "
573                               "burst %u. Attempting to execute request.\n",
574                               data_length, session->max_burst);
575
576         data_offset = be32_to_cpu(rhdr->data_offset);
577         if (data_offset + data_length > task->sc->sdb.length) {
578                 iscsi_conn_printk(KERN_ERR, conn,
579                                   "invalid R2T with data len %u at offset %u "
580                                   "and total length %d\n", data_length,
581                                   data_offset, task->sc->sdb.length);
582                 return ISCSI_ERR_DATALEN;
583         }
584
585         spin_lock(&tcp_task->pool2queue);
586         rc = kfifo_out(&tcp_task->r2tpool.queue, (void *)&r2t, sizeof(void *));
587         if (!rc) {
588                 iscsi_conn_printk(KERN_ERR, conn, "Could not allocate R2T. "
589                                   "Target has sent more R2Ts than it "
590                                   "negotiated for or driver has leaked.\n");
591                 spin_unlock(&tcp_task->pool2queue);
592                 return ISCSI_ERR_PROTO;
593         }
594
595         r2t->exp_statsn = rhdr->statsn;
596         r2t->data_length = data_length;
597         r2t->data_offset = data_offset;
598
599         r2t->ttt = rhdr->ttt; /* no flip */
600         r2t->datasn = 0;
601         r2t->sent = 0;
602
603         tcp_task->exp_datasn = r2tsn + 1;
604         kfifo_in(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
605         conn->r2t_pdus_cnt++;
606         spin_unlock(&tcp_task->pool2queue);
607
608         iscsi_requeue_task(task);
609         return 0;
610 }
611
612 /*
613  * Handle incoming reply to DataIn command
614  */
615 static int
616 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
617                           struct iscsi_segment *segment)
618 {
619         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
620         struct iscsi_hdr *hdr = tcp_conn->in.hdr;
621         int rc;
622
623         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
624                 return ISCSI_ERR_DATA_DGST;
625
626         /* check for non-exceptional status */
627         if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
628                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
629                 if (rc)
630                         return rc;
631         }
632
633         iscsi_tcp_hdr_recv_prep(tcp_conn);
634         return 0;
635 }
636
637 /**
638  * iscsi_tcp_hdr_dissect - process PDU header
639  * @conn: iSCSI connection
640  * @hdr: PDU header
641  *
642  * This function analyzes the header of the PDU received,
643  * and performs several sanity checks. If the PDU is accompanied
644  * by data, the receive buffer is set up to copy the incoming data
645  * to the correct location.
646  */
647 static int
648 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
649 {
650         int rc = 0, opcode, ahslen;
651         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
652         struct iscsi_task *task;
653
654         /* verify PDU length */
655         tcp_conn->in.datalen = ntoh24(hdr->dlength);
656         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
657                 iscsi_conn_printk(KERN_ERR, conn,
658                                   "iscsi_tcp: datalen %d > %d\n",
659                                   tcp_conn->in.datalen, conn->max_recv_dlength);
660                 return ISCSI_ERR_DATALEN;
661         }
662
663         /* Additional header segments. So far, we don't
664          * process additional headers.
665          */
666         ahslen = hdr->hlength << 2;
667
668         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
669         /* verify itt (itt encoding: age+cid+itt) */
670         rc = iscsi_verify_itt(conn, hdr->itt);
671         if (rc)
672                 return rc;
673
674         ISCSI_DBG_TCP(conn, "opcode 0x%x ahslen %d datalen %d\n",
675                       opcode, ahslen, tcp_conn->in.datalen);
676
677         switch(opcode) {
678         case ISCSI_OP_SCSI_DATA_IN:
679                 spin_lock(&conn->session->back_lock);
680                 task = iscsi_itt_to_ctask(conn, hdr->itt);
681                 if (!task)
682                         rc = ISCSI_ERR_BAD_ITT;
683                 else
684                         rc = iscsi_tcp_data_in(conn, task);
685                 if (rc) {
686                         spin_unlock(&conn->session->back_lock);
687                         break;
688                 }
689
690                 if (tcp_conn->in.datalen) {
691                         struct iscsi_tcp_task *tcp_task = task->dd_data;
692                         struct ahash_request *rx_hash = NULL;
693                         struct scsi_data_buffer *sdb = &task->sc->sdb;
694
695                         /*
696                          * Setup copy of Data-In into the struct scsi_cmnd
697                          * Scatterlist case:
698                          * We set up the iscsi_segment to point to the next
699                          * scatterlist entry to copy to. As we go along,
700                          * we move on to the next scatterlist entry and
701                          * update the digest per-entry.
702                          */
703                         if (conn->datadgst_en &&
704                             !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
705                                 rx_hash = tcp_conn->rx_hash;
706
707                         ISCSI_DBG_TCP(conn, "iscsi_tcp_begin_data_in( "
708                                      "offset=%d, datalen=%d)\n",
709                                       tcp_task->data_offset,
710                                       tcp_conn->in.datalen);
711                         task->last_xfer = jiffies;
712                         rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
713                                                    sdb->table.sgl,
714                                                    sdb->table.nents,
715                                                    tcp_task->data_offset,
716                                                    tcp_conn->in.datalen,
717                                                    iscsi_tcp_process_data_in,
718                                                    rx_hash);
719                         spin_unlock(&conn->session->back_lock);
720                         return rc;
721                 }
722                 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
723                 spin_unlock(&conn->session->back_lock);
724                 break;
725         case ISCSI_OP_SCSI_CMD_RSP:
726                 if (tcp_conn->in.datalen) {
727                         iscsi_tcp_data_recv_prep(tcp_conn);
728                         return 0;
729                 }
730                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
731                 break;
732         case ISCSI_OP_R2T:
733                 spin_lock(&conn->session->back_lock);
734                 task = iscsi_itt_to_ctask(conn, hdr->itt);
735                 spin_unlock(&conn->session->back_lock);
736                 if (!task)
737                         rc = ISCSI_ERR_BAD_ITT;
738                 else if (ahslen)
739                         rc = ISCSI_ERR_AHSLEN;
740                 else if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
741                         task->last_xfer = jiffies;
742                         spin_lock(&conn->session->frwd_lock);
743                         rc = iscsi_tcp_r2t_rsp(conn, task);
744                         spin_unlock(&conn->session->frwd_lock);
745                 } else
746                         rc = ISCSI_ERR_PROTO;
747                 break;
748         case ISCSI_OP_LOGIN_RSP:
749         case ISCSI_OP_TEXT_RSP:
750         case ISCSI_OP_REJECT:
751         case ISCSI_OP_ASYNC_EVENT:
752                 /*
753                  * It is possible that we could get a PDU with a buffer larger
754                  * than 8K, but there are no targets that currently do this.
755                  * For now we fail until we find a vendor that needs it
756                  */
757                 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
758                         iscsi_conn_printk(KERN_ERR, conn,
759                                           "iscsi_tcp: received buffer of "
760                                           "len %u but conn buffer is only %u "
761                                           "(opcode %0x)\n",
762                                           tcp_conn->in.datalen,
763                                           ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
764                         rc = ISCSI_ERR_PROTO;
765                         break;
766                 }
767
768                 /* If there's data coming in with the response,
769                  * receive it to the connection's buffer.
770                  */
771                 if (tcp_conn->in.datalen) {
772                         iscsi_tcp_data_recv_prep(tcp_conn);
773                         return 0;
774                 }
775         /* fall through */
776         case ISCSI_OP_LOGOUT_RSP:
777         case ISCSI_OP_NOOP_IN:
778         case ISCSI_OP_SCSI_TMFUNC_RSP:
779                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
780                 break;
781         default:
782                 rc = ISCSI_ERR_BAD_OPCODE;
783                 break;
784         }
785
786         if (rc == 0) {
787                 /* Anything that comes with data should have
788                  * been handled above. */
789                 if (tcp_conn->in.datalen)
790                         return ISCSI_ERR_PROTO;
791                 iscsi_tcp_hdr_recv_prep(tcp_conn);
792         }
793
794         return rc;
795 }
796
797 /**
798  * iscsi_tcp_hdr_recv_done - process PDU header
799  * @tcp_conn: iSCSI TCP connection
800  * @segment: the buffer segment being processed
801  *
802  * This is the callback invoked when the PDU header has
803  * been received. If the header is followed by additional
804  * header segments, we go back for more data.
805  */
806 static int
807 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
808                         struct iscsi_segment *segment)
809 {
810         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
811         struct iscsi_hdr *hdr;
812
813         /* Check if there are additional header segments
814          * *prior* to computing the digest, because we
815          * may need to go back to the caller for more.
816          */
817         hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
818         if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
819                 /* Bump the header length - the caller will
820                  * just loop around and get the AHS for us, and
821                  * call again. */
822                 unsigned int ahslen = hdr->hlength << 2;
823
824                 /* Make sure we don't overflow */
825                 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
826                         return ISCSI_ERR_AHSLEN;
827
828                 segment->total_size += ahslen;
829                 segment->size += ahslen;
830                 return 0;
831         }
832
833         /* We're done processing the header. See if we're doing
834          * header digests; if so, set up the recv_digest buffer
835          * and go back for more. */
836         if (conn->hdrdgst_en &&
837             !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD)) {
838                 if (segment->digest_len == 0) {
839                         /*
840                          * Even if we offload the digest processing we
841                          * splice it in so we can increment the skb/segment
842                          * counters in preparation for the data segment.
843                          */
844                         iscsi_tcp_segment_splice_digest(segment,
845                                                         segment->recv_digest);
846                         return 0;
847                 }
848
849                 iscsi_tcp_dgst_header(tcp_conn->rx_hash, hdr,
850                                       segment->total_copied - ISCSI_DIGEST_SIZE,
851                                       segment->digest);
852
853                 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
854                         return ISCSI_ERR_HDR_DGST;
855         }
856
857         tcp_conn->in.hdr = hdr;
858         return iscsi_tcp_hdr_dissect(conn, hdr);
859 }
860
861 /**
862  * iscsi_tcp_recv_segment_is_hdr - tests if we are reading in a header
863  * @tcp_conn: iscsi tcp conn
864  *
865  * returns non zero if we are currently processing or setup to process
866  * a header.
867  */
868 inline int iscsi_tcp_recv_segment_is_hdr(struct iscsi_tcp_conn *tcp_conn)
869 {
870         return tcp_conn->in.segment.done == iscsi_tcp_hdr_recv_done;
871 }
872 EXPORT_SYMBOL_GPL(iscsi_tcp_recv_segment_is_hdr);
873
874 /**
875  * iscsi_tcp_recv_skb - Process skb
876  * @conn: iscsi connection
877  * @skb: network buffer with header and/or data segment
878  * @offset: offset in skb
879  * @offloaded: bool indicating if transfer was offloaded
880  * @status: iscsi TCP status result
881  *
882  * Will return status of transfer in @status. And will return
883  * number of bytes copied.
884  */
885 int iscsi_tcp_recv_skb(struct iscsi_conn *conn, struct sk_buff *skb,
886                        unsigned int offset, bool offloaded, int *status)
887 {
888         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
889         struct iscsi_segment *segment = &tcp_conn->in.segment;
890         struct skb_seq_state seq;
891         unsigned int consumed = 0;
892         int rc = 0;
893
894         ISCSI_DBG_TCP(conn, "in %d bytes\n", skb->len - offset);
895         /*
896          * Update for each skb instead of pdu, because over slow networks a
897          * data_in's data could take a while to read in. We also want to
898          * account for r2ts.
899          */
900         conn->last_recv = jiffies;
901
902         if (unlikely(conn->suspend_rx)) {
903                 ISCSI_DBG_TCP(conn, "Rx suspended!\n");
904                 *status = ISCSI_TCP_SUSPENDED;
905                 return 0;
906         }
907
908         if (offloaded) {
909                 segment->total_copied = segment->total_size;
910                 goto segment_done;
911         }
912
913         skb_prepare_seq_read(skb, offset, skb->len, &seq);
914         while (1) {
915                 unsigned int avail;
916                 const u8 *ptr;
917
918                 avail = skb_seq_read(consumed, &ptr, &seq);
919                 if (avail == 0) {
920                         ISCSI_DBG_TCP(conn, "no more data avail. Consumed %d\n",
921                                       consumed);
922                         *status = ISCSI_TCP_SKB_DONE;
923                         goto skb_done;
924                 }
925                 BUG_ON(segment->copied >= segment->size);
926
927                 ISCSI_DBG_TCP(conn, "skb %p ptr=%p avail=%u\n", skb, ptr,
928                               avail);
929                 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
930                 BUG_ON(rc == 0);
931                 consumed += rc;
932
933                 if (segment->total_copied >= segment->total_size) {
934                         skb_abort_seq_read(&seq);
935                         goto segment_done;
936                 }
937         }
938
939 segment_done:
940         *status = ISCSI_TCP_SEGMENT_DONE;
941         ISCSI_DBG_TCP(conn, "segment done\n");
942         rc = segment->done(tcp_conn, segment);
943         if (rc != 0) {
944                 *status = ISCSI_TCP_CONN_ERR;
945                 ISCSI_DBG_TCP(conn, "Error receiving PDU, errno=%d\n", rc);
946                 iscsi_conn_failure(conn, rc);
947                 return 0;
948         }
949         /* The done() functions sets up the next segment. */
950
951 skb_done:
952         conn->rxdata_octets += consumed;
953         return consumed;
954 }
955 EXPORT_SYMBOL_GPL(iscsi_tcp_recv_skb);
956
957 /**
958  * iscsi_tcp_task_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
959  * @task: scsi command task
960  */
961 int iscsi_tcp_task_init(struct iscsi_task *task)
962 {
963         struct iscsi_tcp_task *tcp_task = task->dd_data;
964         struct iscsi_conn *conn = task->conn;
965         struct scsi_cmnd *sc = task->sc;
966         int err;
967
968         if (!sc) {
969                 /*
970                  * mgmt tasks do not have a scatterlist since they come
971                  * in from the iscsi interface.
972                  */
973                 ISCSI_DBG_TCP(conn, "mtask deq [itt 0x%x]\n", task->itt);
974
975                 return conn->session->tt->init_pdu(task, 0, task->data_count);
976         }
977
978         BUG_ON(kfifo_len(&tcp_task->r2tqueue));
979         tcp_task->exp_datasn = 0;
980
981         /* Prepare PDU, optionally w/ immediate data */
982         ISCSI_DBG_TCP(conn, "task deq [itt 0x%x imm %d unsol %d]\n",
983                       task->itt, task->imm_count, task->unsol_r2t.data_length);
984
985         err = conn->session->tt->init_pdu(task, 0, task->imm_count);
986         if (err)
987                 return err;
988         task->imm_count = 0;
989         return 0;
990 }
991 EXPORT_SYMBOL_GPL(iscsi_tcp_task_init);
992
993 static struct iscsi_r2t_info *iscsi_tcp_get_curr_r2t(struct iscsi_task *task)
994 {
995         struct iscsi_tcp_task *tcp_task = task->dd_data;
996         struct iscsi_r2t_info *r2t = NULL;
997
998         if (iscsi_task_has_unsol_data(task))
999                 r2t = &task->unsol_r2t;
1000         else {
1001                 spin_lock_bh(&tcp_task->queue2pool);
1002                 if (tcp_task->r2t) {
1003                         r2t = tcp_task->r2t;
1004                         /* Continue with this R2T? */
1005                         if (r2t->data_length <= r2t->sent) {
1006                                 ISCSI_DBG_TCP(task->conn,
1007                                               "  done with r2t %p\n", r2t);
1008                                 kfifo_in(&tcp_task->r2tpool.queue,
1009                                             (void *)&tcp_task->r2t,
1010                                             sizeof(void *));
1011                                 tcp_task->r2t = r2t = NULL;
1012                         }
1013                 }
1014
1015                 if (r2t == NULL) {
1016                         if (kfifo_out(&tcp_task->r2tqueue,
1017                             (void *)&tcp_task->r2t, sizeof(void *)) !=
1018                             sizeof(void *))
1019                                 r2t = NULL;
1020                         else
1021                                 r2t = tcp_task->r2t;
1022                 }
1023                 spin_unlock_bh(&tcp_task->queue2pool);
1024         }
1025
1026         return r2t;
1027 }
1028
1029 /**
1030  * iscsi_tcp_task_xmit - xmit normal PDU task
1031  * @task: iscsi command task
1032  *
1033  * We're expected to return 0 when everything was transmitted successfully,
1034  * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1035  * of error.
1036  */
1037 int iscsi_tcp_task_xmit(struct iscsi_task *task)
1038 {
1039         struct iscsi_conn *conn = task->conn;
1040         struct iscsi_session *session = conn->session;
1041         struct iscsi_r2t_info *r2t;
1042         int rc = 0;
1043
1044 flush:
1045         /* Flush any pending data first. */
1046         rc = session->tt->xmit_pdu(task);
1047         if (rc < 0)
1048                 return rc;
1049
1050         /* mgmt command */
1051         if (!task->sc) {
1052                 if (task->hdr->itt == RESERVED_ITT)
1053                         iscsi_put_task(task);
1054                 return 0;
1055         }
1056
1057         /* Are we done already? */
1058         if (task->sc->sc_data_direction != DMA_TO_DEVICE)
1059                 return 0;
1060
1061         r2t = iscsi_tcp_get_curr_r2t(task);
1062         if (r2t == NULL) {
1063                 /* Waiting for more R2Ts to arrive. */
1064                 ISCSI_DBG_TCP(conn, "no R2Ts yet\n");
1065                 return 0;
1066         }
1067
1068         rc = conn->session->tt->alloc_pdu(task, ISCSI_OP_SCSI_DATA_OUT);
1069         if (rc)
1070                 return rc;
1071         iscsi_prep_data_out_pdu(task, r2t, (struct iscsi_data *) task->hdr);
1072
1073         ISCSI_DBG_TCP(conn, "sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1074                       r2t, r2t->datasn - 1, task->hdr->itt,
1075                       r2t->data_offset + r2t->sent, r2t->data_count);
1076
1077         rc = conn->session->tt->init_pdu(task, r2t->data_offset + r2t->sent,
1078                                          r2t->data_count);
1079         if (rc) {
1080                 iscsi_conn_failure(conn, ISCSI_ERR_XMIT_FAILED);
1081                 return rc;
1082         }
1083
1084         r2t->sent += r2t->data_count;
1085         goto flush;
1086 }
1087 EXPORT_SYMBOL_GPL(iscsi_tcp_task_xmit);
1088
1089 struct iscsi_cls_conn *
1090 iscsi_tcp_conn_setup(struct iscsi_cls_session *cls_session, int dd_data_size,
1091                       uint32_t conn_idx)
1092
1093 {
1094         struct iscsi_conn *conn;
1095         struct iscsi_cls_conn *cls_conn;
1096         struct iscsi_tcp_conn *tcp_conn;
1097
1098         cls_conn = iscsi_conn_setup(cls_session,
1099                                     sizeof(*tcp_conn) + dd_data_size, conn_idx);
1100         if (!cls_conn)
1101                 return NULL;
1102         conn = cls_conn->dd_data;
1103         /*
1104          * due to strange issues with iser these are not set
1105          * in iscsi_conn_setup
1106          */
1107         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1108
1109         tcp_conn = conn->dd_data;
1110         tcp_conn->iscsi_conn = conn;
1111         tcp_conn->dd_data = conn->dd_data + sizeof(*tcp_conn);
1112         return cls_conn;
1113 }
1114 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_setup);
1115
1116 void iscsi_tcp_conn_teardown(struct iscsi_cls_conn *cls_conn)
1117 {
1118         iscsi_conn_teardown(cls_conn);
1119 }
1120 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_teardown);
1121
1122 int iscsi_tcp_r2tpool_alloc(struct iscsi_session *session)
1123 {
1124         int i;
1125         int cmd_i;
1126
1127         /*
1128          * initialize per-task: R2T pool and xmit queue
1129          */
1130         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1131                 struct iscsi_task *task = session->cmds[cmd_i];
1132                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1133
1134                 /*
1135                  * pre-allocated x2 as much r2ts to handle race when
1136                  * target acks DataOut faster than we data_xmit() queues
1137                  * could replenish r2tqueue.
1138                  */
1139
1140                 /* R2T pool */
1141                 if (iscsi_pool_init(&tcp_task->r2tpool,
1142                                     session->max_r2t * 2, NULL,
1143                                     sizeof(struct iscsi_r2t_info))) {
1144                         goto r2t_alloc_fail;
1145                 }
1146
1147                 /* R2T xmit queue */
1148                 if (kfifo_alloc(&tcp_task->r2tqueue,
1149                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL)) {
1150                         iscsi_pool_free(&tcp_task->r2tpool);
1151                         goto r2t_alloc_fail;
1152                 }
1153                 spin_lock_init(&tcp_task->pool2queue);
1154                 spin_lock_init(&tcp_task->queue2pool);
1155         }
1156
1157         return 0;
1158
1159 r2t_alloc_fail:
1160         for (i = 0; i < cmd_i; i++) {
1161                 struct iscsi_task *task = session->cmds[i];
1162                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1163
1164                 kfifo_free(&tcp_task->r2tqueue);
1165                 iscsi_pool_free(&tcp_task->r2tpool);
1166         }
1167         return -ENOMEM;
1168 }
1169 EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_alloc);
1170
1171 void iscsi_tcp_r2tpool_free(struct iscsi_session *session)
1172 {
1173         int i;
1174
1175         for (i = 0; i < session->cmds_max; i++) {
1176                 struct iscsi_task *task = session->cmds[i];
1177                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1178
1179                 kfifo_free(&tcp_task->r2tqueue);
1180                 iscsi_pool_free(&tcp_task->r2tpool);
1181         }
1182 }
1183 EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_free);
1184
1185 int iscsi_tcp_set_max_r2t(struct iscsi_conn *conn, char *buf)
1186 {
1187         struct iscsi_session *session = conn->session;
1188         unsigned short r2ts = 0;
1189
1190         sscanf(buf, "%hu", &r2ts);
1191         if (session->max_r2t == r2ts)
1192                 return 0;
1193
1194         if (!r2ts || !is_power_of_2(r2ts))
1195                 return -EINVAL;
1196
1197         session->max_r2t = r2ts;
1198         iscsi_tcp_r2tpool_free(session);
1199         return iscsi_tcp_r2tpool_alloc(session);
1200 }
1201 EXPORT_SYMBOL_GPL(iscsi_tcp_set_max_r2t);
1202
1203 void iscsi_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
1204                               struct iscsi_stats *stats)
1205 {
1206         struct iscsi_conn *conn = cls_conn->dd_data;
1207
1208         stats->txdata_octets = conn->txdata_octets;
1209         stats->rxdata_octets = conn->rxdata_octets;
1210         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1211         stats->dataout_pdus = conn->dataout_pdus_cnt;
1212         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1213         stats->datain_pdus = conn->datain_pdus_cnt;
1214         stats->r2t_pdus = conn->r2t_pdus_cnt;
1215         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1216         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1217 }
1218 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_get_stats);