Merge tag 'xfs-5.1-merge-5' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
[sfrench/cifs-2.6.git] / drivers / scsi / libiscsi_tcp.c
1 /*
2  * iSCSI over TCP/IP Data-Path lib
3  *
4  * Copyright (C) 2004 Dmitry Yusupov
5  * Copyright (C) 2004 Alex Aizman
6  * Copyright (C) 2005 - 2006 Mike Christie
7  * Copyright (C) 2006 Red Hat, Inc.  All rights reserved.
8  * maintained by open-iscsi@googlegroups.com
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published
12  * by the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18  * General Public License for more details.
19  *
20  * See the file COPYING included with this distribution for more details.
21  *
22  * Credits:
23  *      Christoph Hellwig
24  *      FUJITA Tomonori
25  *      Arne Redlich
26  *      Zhenyu Wang
27  */
28
29 #include <crypto/hash.h>
30 #include <linux/types.h>
31 #include <linux/list.h>
32 #include <linux/inet.h>
33 #include <linux/slab.h>
34 #include <linux/file.h>
35 #include <linux/blkdev.h>
36 #include <linux/delay.h>
37 #include <linux/kfifo.h>
38 #include <linux/scatterlist.h>
39 #include <linux/module.h>
40 #include <net/tcp.h>
41 #include <scsi/scsi_cmnd.h>
42 #include <scsi/scsi_device.h>
43 #include <scsi/scsi_host.h>
44 #include <scsi/scsi.h>
45 #include <scsi/scsi_transport_iscsi.h>
46 #include <trace/events/iscsi.h>
47
48 #include "iscsi_tcp.h"
49
50 MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
51               "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
52               "Alex Aizman <itn780@yahoo.com>");
53 MODULE_DESCRIPTION("iSCSI/TCP data-path");
54 MODULE_LICENSE("GPL");
55
56 static int iscsi_dbg_libtcp;
57 module_param_named(debug_libiscsi_tcp, iscsi_dbg_libtcp, int,
58                    S_IRUGO | S_IWUSR);
59 MODULE_PARM_DESC(debug_libiscsi_tcp, "Turn on debugging for libiscsi_tcp "
60                  "module. Set to 1 to turn on, and zero to turn off. Default "
61                  "is off.");
62
63 #define ISCSI_DBG_TCP(_conn, dbg_fmt, arg...)                   \
64         do {                                                    \
65                 if (iscsi_dbg_libtcp)                           \
66                         iscsi_conn_printk(KERN_INFO, _conn,     \
67                                              "%s " dbg_fmt,     \
68                                              __func__, ##arg);  \
69                 iscsi_dbg_trace(trace_iscsi_dbg_tcp,            \
70                                 &(_conn)->cls_conn->dev,        \
71                                 "%s " dbg_fmt, __func__, ##arg);\
72         } while (0);
73
74 static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
75                                    struct iscsi_segment *segment);
76
77 /*
78  * Scatterlist handling: inside the iscsi_segment, we
79  * remember an index into the scatterlist, and set data/size
80  * to the current scatterlist entry. For highmem pages, we
81  * kmap as needed.
82  *
83  * Note that the page is unmapped when we return from
84  * TCP's data_ready handler, so we may end up mapping and
85  * unmapping the same page repeatedly. The whole reason
86  * for this is that we shouldn't keep the page mapped
87  * outside the softirq.
88  */
89
90 /**
91  * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
92  * @segment: the buffer object
93  * @sg: scatterlist
94  * @offset: byte offset into that sg entry
95  *
96  * This function sets up the segment so that subsequent
97  * data is copied to the indicated sg entry, at the given
98  * offset.
99  */
100 static inline void
101 iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
102                           struct scatterlist *sg, unsigned int offset)
103 {
104         segment->sg = sg;
105         segment->sg_offset = offset;
106         segment->size = min(sg->length - offset,
107                             segment->total_size - segment->total_copied);
108         segment->data = NULL;
109 }
110
111 /**
112  * iscsi_tcp_segment_map - map the current S/G page
113  * @segment: iscsi_segment
114  * @recv: 1 if called from recv path
115  *
116  * We only need to possibly kmap data if scatter lists are being used,
117  * because the iscsi passthrough and internal IO paths will never use high
118  * mem pages.
119  */
120 static void iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
121 {
122         struct scatterlist *sg;
123
124         if (segment->data != NULL || !segment->sg)
125                 return;
126
127         sg = segment->sg;
128         BUG_ON(segment->sg_mapped);
129         BUG_ON(sg->length == 0);
130
131         /*
132          * If the page count is greater than one it is ok to send
133          * to the network layer's zero copy send path. If not we
134          * have to go the slow sendmsg path. We always map for the
135          * recv path.
136          */
137         if (page_count(sg_page(sg)) >= 1 && !recv)
138                 return;
139
140         if (recv) {
141                 segment->atomic_mapped = true;
142                 segment->sg_mapped = kmap_atomic(sg_page(sg));
143         } else {
144                 segment->atomic_mapped = false;
145                 /* the xmit path can sleep with the page mapped so use kmap */
146                 segment->sg_mapped = kmap(sg_page(sg));
147         }
148
149         segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
150 }
151
152 void iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
153 {
154         if (segment->sg_mapped) {
155                 if (segment->atomic_mapped)
156                         kunmap_atomic(segment->sg_mapped);
157                 else
158                         kunmap(sg_page(segment->sg));
159                 segment->sg_mapped = NULL;
160                 segment->data = NULL;
161         }
162 }
163 EXPORT_SYMBOL_GPL(iscsi_tcp_segment_unmap);
164
165 /*
166  * Splice the digest buffer into the buffer
167  */
168 static inline void
169 iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
170 {
171         segment->data = digest;
172         segment->digest_len = ISCSI_DIGEST_SIZE;
173         segment->total_size += ISCSI_DIGEST_SIZE;
174         segment->size = ISCSI_DIGEST_SIZE;
175         segment->copied = 0;
176         segment->sg = NULL;
177         segment->hash = NULL;
178 }
179
180 /**
181  * iscsi_tcp_segment_done - check whether the segment is complete
182  * @tcp_conn: iscsi tcp connection
183  * @segment: iscsi segment to check
184  * @recv: set to one of this is called from the recv path
185  * @copied: number of bytes copied
186  *
187  * Check if we're done receiving this segment. If the receive
188  * buffer is full but we expect more data, move on to the
189  * next entry in the scatterlist.
190  *
191  * If the amount of data we received isn't a multiple of 4,
192  * we will transparently receive the pad bytes, too.
193  *
194  * This function must be re-entrant.
195  */
196 int iscsi_tcp_segment_done(struct iscsi_tcp_conn *tcp_conn,
197                            struct iscsi_segment *segment, int recv,
198                            unsigned copied)
199 {
200         struct scatterlist sg;
201         unsigned int pad;
202
203         ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copied %u %u size %u %s\n",
204                       segment->copied, copied, segment->size,
205                       recv ? "recv" : "xmit");
206         if (segment->hash && copied) {
207                 /*
208                  * If a segment is kmapd we must unmap it before sending
209                  * to the crypto layer since that will try to kmap it again.
210                  */
211                 iscsi_tcp_segment_unmap(segment);
212
213                 if (!segment->data) {
214                         sg_init_table(&sg, 1);
215                         sg_set_page(&sg, sg_page(segment->sg), copied,
216                                     segment->copied + segment->sg_offset +
217                                                         segment->sg->offset);
218                 } else
219                         sg_init_one(&sg, segment->data + segment->copied,
220                                     copied);
221                 ahash_request_set_crypt(segment->hash, &sg, NULL, copied);
222                 crypto_ahash_update(segment->hash);
223         }
224
225         segment->copied += copied;
226         if (segment->copied < segment->size) {
227                 iscsi_tcp_segment_map(segment, recv);
228                 return 0;
229         }
230
231         segment->total_copied += segment->copied;
232         segment->copied = 0;
233         segment->size = 0;
234
235         /* Unmap the current scatterlist page, if there is one. */
236         iscsi_tcp_segment_unmap(segment);
237
238         /* Do we have more scatterlist entries? */
239         ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "total copied %u total size %u\n",
240                       segment->total_copied, segment->total_size);
241         if (segment->total_copied < segment->total_size) {
242                 /* Proceed to the next entry in the scatterlist. */
243                 iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
244                                           0);
245                 iscsi_tcp_segment_map(segment, recv);
246                 BUG_ON(segment->size == 0);
247                 return 0;
248         }
249
250         /* Do we need to handle padding? */
251         if (!(tcp_conn->iscsi_conn->session->tt->caps & CAP_PADDING_OFFLOAD)) {
252                 pad = iscsi_padding(segment->total_copied);
253                 if (pad != 0) {
254                         ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
255                                       "consume %d pad bytes\n", pad);
256                         segment->total_size += pad;
257                         segment->size = pad;
258                         segment->data = segment->padbuf;
259                         return 0;
260                 }
261         }
262
263         /*
264          * Set us up for transferring the data digest. hdr digest
265          * is completely handled in hdr done function.
266          */
267         if (segment->hash) {
268                 ahash_request_set_crypt(segment->hash, NULL,
269                                         segment->digest, 0);
270                 crypto_ahash_final(segment->hash);
271                 iscsi_tcp_segment_splice_digest(segment,
272                                  recv ? segment->recv_digest : segment->digest);
273                 return 0;
274         }
275
276         return 1;
277 }
278 EXPORT_SYMBOL_GPL(iscsi_tcp_segment_done);
279
280 /**
281  * iscsi_tcp_segment_recv - copy data to segment
282  * @tcp_conn: the iSCSI TCP connection
283  * @segment: the buffer to copy to
284  * @ptr: data pointer
285  * @len: amount of data available
286  *
287  * This function copies up to @len bytes to the
288  * given buffer, and returns the number of bytes
289  * consumed, which can actually be less than @len.
290  *
291  * If hash digest is enabled, the function will update the
292  * hash while copying.
293  * Combining these two operations doesn't buy us a lot (yet),
294  * but in the future we could implement combined copy+crc,
295  * just way we do for network layer checksums.
296  */
297 static int
298 iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
299                        struct iscsi_segment *segment, const void *ptr,
300                        unsigned int len)
301 {
302         unsigned int copy = 0, copied = 0;
303
304         while (!iscsi_tcp_segment_done(tcp_conn, segment, 1, copy)) {
305                 if (copied == len) {
306                         ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
307                                       "copied %d bytes\n", len);
308                         break;
309                 }
310
311                 copy = min(len - copied, segment->size - segment->copied);
312                 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copying %d\n", copy);
313                 memcpy(segment->data + segment->copied, ptr + copied, copy);
314                 copied += copy;
315         }
316         return copied;
317 }
318
319 inline void
320 iscsi_tcp_dgst_header(struct ahash_request *hash, const void *hdr,
321                       size_t hdrlen, unsigned char digest[ISCSI_DIGEST_SIZE])
322 {
323         struct scatterlist sg;
324
325         sg_init_one(&sg, hdr, hdrlen);
326         ahash_request_set_crypt(hash, &sg, digest, hdrlen);
327         crypto_ahash_digest(hash);
328 }
329 EXPORT_SYMBOL_GPL(iscsi_tcp_dgst_header);
330
331 static inline int
332 iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
333                       struct iscsi_segment *segment)
334 {
335         if (!segment->digest_len)
336                 return 1;
337
338         if (memcmp(segment->recv_digest, segment->digest,
339                    segment->digest_len)) {
340                 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "digest mismatch\n");
341                 return 0;
342         }
343
344         return 1;
345 }
346
347 /*
348  * Helper function to set up segment buffer
349  */
350 static inline void
351 __iscsi_segment_init(struct iscsi_segment *segment, size_t size,
352                      iscsi_segment_done_fn_t *done, struct ahash_request *hash)
353 {
354         memset(segment, 0, sizeof(*segment));
355         segment->total_size = size;
356         segment->done = done;
357
358         if (hash) {
359                 segment->hash = hash;
360                 crypto_ahash_init(hash);
361         }
362 }
363
364 inline void
365 iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
366                           size_t size, iscsi_segment_done_fn_t *done,
367                           struct ahash_request *hash)
368 {
369         __iscsi_segment_init(segment, size, done, hash);
370         segment->data = data;
371         segment->size = size;
372 }
373 EXPORT_SYMBOL_GPL(iscsi_segment_init_linear);
374
375 inline int
376 iscsi_segment_seek_sg(struct iscsi_segment *segment,
377                       struct scatterlist *sg_list, unsigned int sg_count,
378                       unsigned int offset, size_t size,
379                       iscsi_segment_done_fn_t *done,
380                       struct ahash_request *hash)
381 {
382         struct scatterlist *sg;
383         unsigned int i;
384
385         __iscsi_segment_init(segment, size, done, hash);
386         for_each_sg(sg_list, sg, sg_count, i) {
387                 if (offset < sg->length) {
388                         iscsi_tcp_segment_init_sg(segment, sg, offset);
389                         return 0;
390                 }
391                 offset -= sg->length;
392         }
393
394         return ISCSI_ERR_DATA_OFFSET;
395 }
396 EXPORT_SYMBOL_GPL(iscsi_segment_seek_sg);
397
398 /**
399  * iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
400  * @tcp_conn: iscsi connection to prep for
401  *
402  * This function always passes NULL for the hash argument, because when this
403  * function is called we do not yet know the final size of the header and want
404  * to delay the digest processing until we know that.
405  */
406 void iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
407 {
408         ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
409                       "(%s)\n", tcp_conn->iscsi_conn->hdrdgst_en ?
410                       "digest enabled" : "digest disabled");
411         iscsi_segment_init_linear(&tcp_conn->in.segment,
412                                 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
413                                 iscsi_tcp_hdr_recv_done, NULL);
414 }
415 EXPORT_SYMBOL_GPL(iscsi_tcp_hdr_recv_prep);
416
417 /*
418  * Handle incoming reply to any other type of command
419  */
420 static int
421 iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
422                          struct iscsi_segment *segment)
423 {
424         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
425         int rc = 0;
426
427         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
428                 return ISCSI_ERR_DATA_DGST;
429
430         rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
431                         conn->data, tcp_conn->in.datalen);
432         if (rc)
433                 return rc;
434
435         iscsi_tcp_hdr_recv_prep(tcp_conn);
436         return 0;
437 }
438
439 static void
440 iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
441 {
442         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
443         struct ahash_request *rx_hash = NULL;
444
445         if (conn->datadgst_en &&
446             !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
447                 rx_hash = tcp_conn->rx_hash;
448
449         iscsi_segment_init_linear(&tcp_conn->in.segment,
450                                 conn->data, tcp_conn->in.datalen,
451                                 iscsi_tcp_data_recv_done, rx_hash);
452 }
453
454 /**
455  * iscsi_tcp_cleanup_task - free tcp_task resources
456  * @task: iscsi task
457  *
458  * must be called with session back_lock
459  */
460 void iscsi_tcp_cleanup_task(struct iscsi_task *task)
461 {
462         struct iscsi_tcp_task *tcp_task = task->dd_data;
463         struct iscsi_r2t_info *r2t;
464
465         /* nothing to do for mgmt */
466         if (!task->sc)
467                 return;
468
469         spin_lock_bh(&tcp_task->queue2pool);
470         /* flush task's r2t queues */
471         while (kfifo_out(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
472                 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
473                             sizeof(void*));
474                 ISCSI_DBG_TCP(task->conn, "pending r2t dropped\n");
475         }
476
477         r2t = tcp_task->r2t;
478         if (r2t != NULL) {
479                 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
480                             sizeof(void*));
481                 tcp_task->r2t = NULL;
482         }
483         spin_unlock_bh(&tcp_task->queue2pool);
484 }
485 EXPORT_SYMBOL_GPL(iscsi_tcp_cleanup_task);
486
487 /**
488  * iscsi_tcp_data_in - SCSI Data-In Response processing
489  * @conn: iscsi connection
490  * @task: scsi command task
491  */
492 static int iscsi_tcp_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
493 {
494         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
495         struct iscsi_tcp_task *tcp_task = task->dd_data;
496         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
497         int datasn = be32_to_cpu(rhdr->datasn);
498         unsigned total_in_length = task->sc->sdb.length;
499
500         /*
501          * lib iscsi will update this in the completion handling if there
502          * is status.
503          */
504         if (!(rhdr->flags & ISCSI_FLAG_DATA_STATUS))
505                 iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
506
507         if (tcp_conn->in.datalen == 0)
508                 return 0;
509
510         if (tcp_task->exp_datasn != datasn) {
511                 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->datasn(%d)"
512                               "\n", tcp_task->exp_datasn, datasn);
513                 return ISCSI_ERR_DATASN;
514         }
515
516         tcp_task->exp_datasn++;
517
518         tcp_task->data_offset = be32_to_cpu(rhdr->offset);
519         if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
520                 ISCSI_DBG_TCP(conn, "data_offset(%d) + data_len(%d) > "
521                               "total_length_in(%d)\n", tcp_task->data_offset,
522                               tcp_conn->in.datalen, total_in_length);
523                 return ISCSI_ERR_DATA_OFFSET;
524         }
525
526         conn->datain_pdus_cnt++;
527         return 0;
528 }
529
530 /**
531  * iscsi_tcp_r2t_rsp - iSCSI R2T Response processing
532  * @conn: iscsi connection
533  * @task: scsi command task
534  */
535 static int iscsi_tcp_r2t_rsp(struct iscsi_conn *conn, struct iscsi_task *task)
536 {
537         struct iscsi_session *session = conn->session;
538         struct iscsi_tcp_task *tcp_task = task->dd_data;
539         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
540         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
541         struct iscsi_r2t_info *r2t;
542         int r2tsn = be32_to_cpu(rhdr->r2tsn);
543         u32 data_length;
544         u32 data_offset;
545         int rc;
546
547         if (tcp_conn->in.datalen) {
548                 iscsi_conn_printk(KERN_ERR, conn,
549                                   "invalid R2t with datalen %d\n",
550                                   tcp_conn->in.datalen);
551                 return ISCSI_ERR_DATALEN;
552         }
553
554         if (tcp_task->exp_datasn != r2tsn){
555                 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
556                               tcp_task->exp_datasn, r2tsn);
557                 return ISCSI_ERR_R2TSN;
558         }
559
560         /* fill-in new R2T associated with the task */
561         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
562
563         if (!task->sc || session->state != ISCSI_STATE_LOGGED_IN) {
564                 iscsi_conn_printk(KERN_INFO, conn,
565                                   "dropping R2T itt %d in recovery.\n",
566                                   task->itt);
567                 return 0;
568         }
569
570         data_length = be32_to_cpu(rhdr->data_length);
571         if (data_length == 0) {
572                 iscsi_conn_printk(KERN_ERR, conn,
573                                   "invalid R2T with zero data len\n");
574                 return ISCSI_ERR_DATALEN;
575         }
576
577         if (data_length > session->max_burst)
578                 ISCSI_DBG_TCP(conn, "invalid R2T with data len %u and max "
579                               "burst %u. Attempting to execute request.\n",
580                               data_length, session->max_burst);
581
582         data_offset = be32_to_cpu(rhdr->data_offset);
583         if (data_offset + data_length > task->sc->sdb.length) {
584                 iscsi_conn_printk(KERN_ERR, conn,
585                                   "invalid R2T with data len %u at offset %u "
586                                   "and total length %d\n", data_length,
587                                   data_offset, task->sc->sdb.length);
588                 return ISCSI_ERR_DATALEN;
589         }
590
591         spin_lock(&tcp_task->pool2queue);
592         rc = kfifo_out(&tcp_task->r2tpool.queue, (void *)&r2t, sizeof(void *));
593         if (!rc) {
594                 iscsi_conn_printk(KERN_ERR, conn, "Could not allocate R2T. "
595                                   "Target has sent more R2Ts than it "
596                                   "negotiated for or driver has leaked.\n");
597                 spin_unlock(&tcp_task->pool2queue);
598                 return ISCSI_ERR_PROTO;
599         }
600
601         r2t->exp_statsn = rhdr->statsn;
602         r2t->data_length = data_length;
603         r2t->data_offset = data_offset;
604
605         r2t->ttt = rhdr->ttt; /* no flip */
606         r2t->datasn = 0;
607         r2t->sent = 0;
608
609         tcp_task->exp_datasn = r2tsn + 1;
610         kfifo_in(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
611         conn->r2t_pdus_cnt++;
612         spin_unlock(&tcp_task->pool2queue);
613
614         iscsi_requeue_task(task);
615         return 0;
616 }
617
618 /*
619  * Handle incoming reply to DataIn command
620  */
621 static int
622 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
623                           struct iscsi_segment *segment)
624 {
625         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
626         struct iscsi_hdr *hdr = tcp_conn->in.hdr;
627         int rc;
628
629         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
630                 return ISCSI_ERR_DATA_DGST;
631
632         /* check for non-exceptional status */
633         if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
634                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
635                 if (rc)
636                         return rc;
637         }
638
639         iscsi_tcp_hdr_recv_prep(tcp_conn);
640         return 0;
641 }
642
643 /**
644  * iscsi_tcp_hdr_dissect - process PDU header
645  * @conn: iSCSI connection
646  * @hdr: PDU header
647  *
648  * This function analyzes the header of the PDU received,
649  * and performs several sanity checks. If the PDU is accompanied
650  * by data, the receive buffer is set up to copy the incoming data
651  * to the correct location.
652  */
653 static int
654 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
655 {
656         int rc = 0, opcode, ahslen;
657         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
658         struct iscsi_task *task;
659
660         /* verify PDU length */
661         tcp_conn->in.datalen = ntoh24(hdr->dlength);
662         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
663                 iscsi_conn_printk(KERN_ERR, conn,
664                                   "iscsi_tcp: datalen %d > %d\n",
665                                   tcp_conn->in.datalen, conn->max_recv_dlength);
666                 return ISCSI_ERR_DATALEN;
667         }
668
669         /* Additional header segments. So far, we don't
670          * process additional headers.
671          */
672         ahslen = hdr->hlength << 2;
673
674         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
675         /* verify itt (itt encoding: age+cid+itt) */
676         rc = iscsi_verify_itt(conn, hdr->itt);
677         if (rc)
678                 return rc;
679
680         ISCSI_DBG_TCP(conn, "opcode 0x%x ahslen %d datalen %d\n",
681                       opcode, ahslen, tcp_conn->in.datalen);
682
683         switch(opcode) {
684         case ISCSI_OP_SCSI_DATA_IN:
685                 spin_lock(&conn->session->back_lock);
686                 task = iscsi_itt_to_ctask(conn, hdr->itt);
687                 if (!task)
688                         rc = ISCSI_ERR_BAD_ITT;
689                 else
690                         rc = iscsi_tcp_data_in(conn, task);
691                 if (rc) {
692                         spin_unlock(&conn->session->back_lock);
693                         break;
694                 }
695
696                 if (tcp_conn->in.datalen) {
697                         struct iscsi_tcp_task *tcp_task = task->dd_data;
698                         struct ahash_request *rx_hash = NULL;
699                         struct scsi_data_buffer *sdb = &task->sc->sdb;
700
701                         /*
702                          * Setup copy of Data-In into the struct scsi_cmnd
703                          * Scatterlist case:
704                          * We set up the iscsi_segment to point to the next
705                          * scatterlist entry to copy to. As we go along,
706                          * we move on to the next scatterlist entry and
707                          * update the digest per-entry.
708                          */
709                         if (conn->datadgst_en &&
710                             !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
711                                 rx_hash = tcp_conn->rx_hash;
712
713                         ISCSI_DBG_TCP(conn, "iscsi_tcp_begin_data_in( "
714                                      "offset=%d, datalen=%d)\n",
715                                       tcp_task->data_offset,
716                                       tcp_conn->in.datalen);
717                         task->last_xfer = jiffies;
718                         rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
719                                                    sdb->table.sgl,
720                                                    sdb->table.nents,
721                                                    tcp_task->data_offset,
722                                                    tcp_conn->in.datalen,
723                                                    iscsi_tcp_process_data_in,
724                                                    rx_hash);
725                         spin_unlock(&conn->session->back_lock);
726                         return rc;
727                 }
728                 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
729                 spin_unlock(&conn->session->back_lock);
730                 break;
731         case ISCSI_OP_SCSI_CMD_RSP:
732                 if (tcp_conn->in.datalen) {
733                         iscsi_tcp_data_recv_prep(tcp_conn);
734                         return 0;
735                 }
736                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
737                 break;
738         case ISCSI_OP_R2T:
739                 spin_lock(&conn->session->back_lock);
740                 task = iscsi_itt_to_ctask(conn, hdr->itt);
741                 spin_unlock(&conn->session->back_lock);
742                 if (!task)
743                         rc = ISCSI_ERR_BAD_ITT;
744                 else if (ahslen)
745                         rc = ISCSI_ERR_AHSLEN;
746                 else if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
747                         task->last_xfer = jiffies;
748                         spin_lock(&conn->session->frwd_lock);
749                         rc = iscsi_tcp_r2t_rsp(conn, task);
750                         spin_unlock(&conn->session->frwd_lock);
751                 } else
752                         rc = ISCSI_ERR_PROTO;
753                 break;
754         case ISCSI_OP_LOGIN_RSP:
755         case ISCSI_OP_TEXT_RSP:
756         case ISCSI_OP_REJECT:
757         case ISCSI_OP_ASYNC_EVENT:
758                 /*
759                  * It is possible that we could get a PDU with a buffer larger
760                  * than 8K, but there are no targets that currently do this.
761                  * For now we fail until we find a vendor that needs it
762                  */
763                 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
764                         iscsi_conn_printk(KERN_ERR, conn,
765                                           "iscsi_tcp: received buffer of "
766                                           "len %u but conn buffer is only %u "
767                                           "(opcode %0x)\n",
768                                           tcp_conn->in.datalen,
769                                           ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
770                         rc = ISCSI_ERR_PROTO;
771                         break;
772                 }
773
774                 /* If there's data coming in with the response,
775                  * receive it to the connection's buffer.
776                  */
777                 if (tcp_conn->in.datalen) {
778                         iscsi_tcp_data_recv_prep(tcp_conn);
779                         return 0;
780                 }
781         /* fall through */
782         case ISCSI_OP_LOGOUT_RSP:
783         case ISCSI_OP_NOOP_IN:
784         case ISCSI_OP_SCSI_TMFUNC_RSP:
785                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
786                 break;
787         default:
788                 rc = ISCSI_ERR_BAD_OPCODE;
789                 break;
790         }
791
792         if (rc == 0) {
793                 /* Anything that comes with data should have
794                  * been handled above. */
795                 if (tcp_conn->in.datalen)
796                         return ISCSI_ERR_PROTO;
797                 iscsi_tcp_hdr_recv_prep(tcp_conn);
798         }
799
800         return rc;
801 }
802
803 /**
804  * iscsi_tcp_hdr_recv_done - process PDU header
805  * @tcp_conn: iSCSI TCP connection
806  * @segment: the buffer segment being processed
807  *
808  * This is the callback invoked when the PDU header has
809  * been received. If the header is followed by additional
810  * header segments, we go back for more data.
811  */
812 static int
813 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
814                         struct iscsi_segment *segment)
815 {
816         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
817         struct iscsi_hdr *hdr;
818
819         /* Check if there are additional header segments
820          * *prior* to computing the digest, because we
821          * may need to go back to the caller for more.
822          */
823         hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
824         if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
825                 /* Bump the header length - the caller will
826                  * just loop around and get the AHS for us, and
827                  * call again. */
828                 unsigned int ahslen = hdr->hlength << 2;
829
830                 /* Make sure we don't overflow */
831                 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
832                         return ISCSI_ERR_AHSLEN;
833
834                 segment->total_size += ahslen;
835                 segment->size += ahslen;
836                 return 0;
837         }
838
839         /* We're done processing the header. See if we're doing
840          * header digests; if so, set up the recv_digest buffer
841          * and go back for more. */
842         if (conn->hdrdgst_en &&
843             !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD)) {
844                 if (segment->digest_len == 0) {
845                         /*
846                          * Even if we offload the digest processing we
847                          * splice it in so we can increment the skb/segment
848                          * counters in preparation for the data segment.
849                          */
850                         iscsi_tcp_segment_splice_digest(segment,
851                                                         segment->recv_digest);
852                         return 0;
853                 }
854
855                 iscsi_tcp_dgst_header(tcp_conn->rx_hash, hdr,
856                                       segment->total_copied - ISCSI_DIGEST_SIZE,
857                                       segment->digest);
858
859                 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
860                         return ISCSI_ERR_HDR_DGST;
861         }
862
863         tcp_conn->in.hdr = hdr;
864         return iscsi_tcp_hdr_dissect(conn, hdr);
865 }
866
867 /**
868  * iscsi_tcp_recv_segment_is_hdr - tests if we are reading in a header
869  * @tcp_conn: iscsi tcp conn
870  *
871  * returns non zero if we are currently processing or setup to process
872  * a header.
873  */
874 inline int iscsi_tcp_recv_segment_is_hdr(struct iscsi_tcp_conn *tcp_conn)
875 {
876         return tcp_conn->in.segment.done == iscsi_tcp_hdr_recv_done;
877 }
878 EXPORT_SYMBOL_GPL(iscsi_tcp_recv_segment_is_hdr);
879
880 /**
881  * iscsi_tcp_recv_skb - Process skb
882  * @conn: iscsi connection
883  * @skb: network buffer with header and/or data segment
884  * @offset: offset in skb
885  * @offloaded: bool indicating if transfer was offloaded
886  * @status: iscsi TCP status result
887  *
888  * Will return status of transfer in @status. And will return
889  * number of bytes copied.
890  */
891 int iscsi_tcp_recv_skb(struct iscsi_conn *conn, struct sk_buff *skb,
892                        unsigned int offset, bool offloaded, int *status)
893 {
894         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
895         struct iscsi_segment *segment = &tcp_conn->in.segment;
896         struct skb_seq_state seq;
897         unsigned int consumed = 0;
898         int rc = 0;
899
900         ISCSI_DBG_TCP(conn, "in %d bytes\n", skb->len - offset);
901         /*
902          * Update for each skb instead of pdu, because over slow networks a
903          * data_in's data could take a while to read in. We also want to
904          * account for r2ts.
905          */
906         conn->last_recv = jiffies;
907
908         if (unlikely(conn->suspend_rx)) {
909                 ISCSI_DBG_TCP(conn, "Rx suspended!\n");
910                 *status = ISCSI_TCP_SUSPENDED;
911                 return 0;
912         }
913
914         if (offloaded) {
915                 segment->total_copied = segment->total_size;
916                 goto segment_done;
917         }
918
919         skb_prepare_seq_read(skb, offset, skb->len, &seq);
920         while (1) {
921                 unsigned int avail;
922                 const u8 *ptr;
923
924                 avail = skb_seq_read(consumed, &ptr, &seq);
925                 if (avail == 0) {
926                         ISCSI_DBG_TCP(conn, "no more data avail. Consumed %d\n",
927                                       consumed);
928                         *status = ISCSI_TCP_SKB_DONE;
929                         goto skb_done;
930                 }
931                 BUG_ON(segment->copied >= segment->size);
932
933                 ISCSI_DBG_TCP(conn, "skb %p ptr=%p avail=%u\n", skb, ptr,
934                               avail);
935                 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
936                 BUG_ON(rc == 0);
937                 consumed += rc;
938
939                 if (segment->total_copied >= segment->total_size) {
940                         skb_abort_seq_read(&seq);
941                         goto segment_done;
942                 }
943         }
944
945 segment_done:
946         *status = ISCSI_TCP_SEGMENT_DONE;
947         ISCSI_DBG_TCP(conn, "segment done\n");
948         rc = segment->done(tcp_conn, segment);
949         if (rc != 0) {
950                 *status = ISCSI_TCP_CONN_ERR;
951                 ISCSI_DBG_TCP(conn, "Error receiving PDU, errno=%d\n", rc);
952                 iscsi_conn_failure(conn, rc);
953                 return 0;
954         }
955         /* The done() functions sets up the next segment. */
956
957 skb_done:
958         conn->rxdata_octets += consumed;
959         return consumed;
960 }
961 EXPORT_SYMBOL_GPL(iscsi_tcp_recv_skb);
962
963 /**
964  * iscsi_tcp_task_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
965  * @task: scsi command task
966  */
967 int iscsi_tcp_task_init(struct iscsi_task *task)
968 {
969         struct iscsi_tcp_task *tcp_task = task->dd_data;
970         struct iscsi_conn *conn = task->conn;
971         struct scsi_cmnd *sc = task->sc;
972         int err;
973
974         if (!sc) {
975                 /*
976                  * mgmt tasks do not have a scatterlist since they come
977                  * in from the iscsi interface.
978                  */
979                 ISCSI_DBG_TCP(conn, "mtask deq [itt 0x%x]\n", task->itt);
980
981                 return conn->session->tt->init_pdu(task, 0, task->data_count);
982         }
983
984         BUG_ON(kfifo_len(&tcp_task->r2tqueue));
985         tcp_task->exp_datasn = 0;
986
987         /* Prepare PDU, optionally w/ immediate data */
988         ISCSI_DBG_TCP(conn, "task deq [itt 0x%x imm %d unsol %d]\n",
989                       task->itt, task->imm_count, task->unsol_r2t.data_length);
990
991         err = conn->session->tt->init_pdu(task, 0, task->imm_count);
992         if (err)
993                 return err;
994         task->imm_count = 0;
995         return 0;
996 }
997 EXPORT_SYMBOL_GPL(iscsi_tcp_task_init);
998
999 static struct iscsi_r2t_info *iscsi_tcp_get_curr_r2t(struct iscsi_task *task)
1000 {
1001         struct iscsi_tcp_task *tcp_task = task->dd_data;
1002         struct iscsi_r2t_info *r2t = NULL;
1003
1004         if (iscsi_task_has_unsol_data(task))
1005                 r2t = &task->unsol_r2t;
1006         else {
1007                 spin_lock_bh(&tcp_task->queue2pool);
1008                 if (tcp_task->r2t) {
1009                         r2t = tcp_task->r2t;
1010                         /* Continue with this R2T? */
1011                         if (r2t->data_length <= r2t->sent) {
1012                                 ISCSI_DBG_TCP(task->conn,
1013                                               "  done with r2t %p\n", r2t);
1014                                 kfifo_in(&tcp_task->r2tpool.queue,
1015                                             (void *)&tcp_task->r2t,
1016                                             sizeof(void *));
1017                                 tcp_task->r2t = r2t = NULL;
1018                         }
1019                 }
1020
1021                 if (r2t == NULL) {
1022                         if (kfifo_out(&tcp_task->r2tqueue,
1023                             (void *)&tcp_task->r2t, sizeof(void *)) !=
1024                             sizeof(void *))
1025                                 r2t = NULL;
1026                         else
1027                                 r2t = tcp_task->r2t;
1028                 }
1029                 spin_unlock_bh(&tcp_task->queue2pool);
1030         }
1031
1032         return r2t;
1033 }
1034
1035 /**
1036  * iscsi_tcp_task_xmit - xmit normal PDU task
1037  * @task: iscsi command task
1038  *
1039  * We're expected to return 0 when everything was transmitted successfully,
1040  * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1041  * of error.
1042  */
1043 int iscsi_tcp_task_xmit(struct iscsi_task *task)
1044 {
1045         struct iscsi_conn *conn = task->conn;
1046         struct iscsi_session *session = conn->session;
1047         struct iscsi_r2t_info *r2t;
1048         int rc = 0;
1049
1050 flush:
1051         /* Flush any pending data first. */
1052         rc = session->tt->xmit_pdu(task);
1053         if (rc < 0)
1054                 return rc;
1055
1056         /* mgmt command */
1057         if (!task->sc) {
1058                 if (task->hdr->itt == RESERVED_ITT)
1059                         iscsi_put_task(task);
1060                 return 0;
1061         }
1062
1063         /* Are we done already? */
1064         if (task->sc->sc_data_direction != DMA_TO_DEVICE)
1065                 return 0;
1066
1067         r2t = iscsi_tcp_get_curr_r2t(task);
1068         if (r2t == NULL) {
1069                 /* Waiting for more R2Ts to arrive. */
1070                 ISCSI_DBG_TCP(conn, "no R2Ts yet\n");
1071                 return 0;
1072         }
1073
1074         rc = conn->session->tt->alloc_pdu(task, ISCSI_OP_SCSI_DATA_OUT);
1075         if (rc)
1076                 return rc;
1077         iscsi_prep_data_out_pdu(task, r2t, (struct iscsi_data *) task->hdr);
1078
1079         ISCSI_DBG_TCP(conn, "sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1080                       r2t, r2t->datasn - 1, task->hdr->itt,
1081                       r2t->data_offset + r2t->sent, r2t->data_count);
1082
1083         rc = conn->session->tt->init_pdu(task, r2t->data_offset + r2t->sent,
1084                                          r2t->data_count);
1085         if (rc) {
1086                 iscsi_conn_failure(conn, ISCSI_ERR_XMIT_FAILED);
1087                 return rc;
1088         }
1089
1090         r2t->sent += r2t->data_count;
1091         goto flush;
1092 }
1093 EXPORT_SYMBOL_GPL(iscsi_tcp_task_xmit);
1094
1095 struct iscsi_cls_conn *
1096 iscsi_tcp_conn_setup(struct iscsi_cls_session *cls_session, int dd_data_size,
1097                       uint32_t conn_idx)
1098
1099 {
1100         struct iscsi_conn *conn;
1101         struct iscsi_cls_conn *cls_conn;
1102         struct iscsi_tcp_conn *tcp_conn;
1103
1104         cls_conn = iscsi_conn_setup(cls_session,
1105                                     sizeof(*tcp_conn) + dd_data_size, conn_idx);
1106         if (!cls_conn)
1107                 return NULL;
1108         conn = cls_conn->dd_data;
1109         /*
1110          * due to strange issues with iser these are not set
1111          * in iscsi_conn_setup
1112          */
1113         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1114
1115         tcp_conn = conn->dd_data;
1116         tcp_conn->iscsi_conn = conn;
1117         tcp_conn->dd_data = conn->dd_data + sizeof(*tcp_conn);
1118         return cls_conn;
1119 }
1120 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_setup);
1121
1122 void iscsi_tcp_conn_teardown(struct iscsi_cls_conn *cls_conn)
1123 {
1124         iscsi_conn_teardown(cls_conn);
1125 }
1126 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_teardown);
1127
1128 int iscsi_tcp_r2tpool_alloc(struct iscsi_session *session)
1129 {
1130         int i;
1131         int cmd_i;
1132
1133         /*
1134          * initialize per-task: R2T pool and xmit queue
1135          */
1136         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1137                 struct iscsi_task *task = session->cmds[cmd_i];
1138                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1139
1140                 /*
1141                  * pre-allocated x2 as much r2ts to handle race when
1142                  * target acks DataOut faster than we data_xmit() queues
1143                  * could replenish r2tqueue.
1144                  */
1145
1146                 /* R2T pool */
1147                 if (iscsi_pool_init(&tcp_task->r2tpool,
1148                                     session->max_r2t * 2, NULL,
1149                                     sizeof(struct iscsi_r2t_info))) {
1150                         goto r2t_alloc_fail;
1151                 }
1152
1153                 /* R2T xmit queue */
1154                 if (kfifo_alloc(&tcp_task->r2tqueue,
1155                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL)) {
1156                         iscsi_pool_free(&tcp_task->r2tpool);
1157                         goto r2t_alloc_fail;
1158                 }
1159                 spin_lock_init(&tcp_task->pool2queue);
1160                 spin_lock_init(&tcp_task->queue2pool);
1161         }
1162
1163         return 0;
1164
1165 r2t_alloc_fail:
1166         for (i = 0; i < cmd_i; i++) {
1167                 struct iscsi_task *task = session->cmds[i];
1168                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1169
1170                 kfifo_free(&tcp_task->r2tqueue);
1171                 iscsi_pool_free(&tcp_task->r2tpool);
1172         }
1173         return -ENOMEM;
1174 }
1175 EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_alloc);
1176
1177 void iscsi_tcp_r2tpool_free(struct iscsi_session *session)
1178 {
1179         int i;
1180
1181         for (i = 0; i < session->cmds_max; i++) {
1182                 struct iscsi_task *task = session->cmds[i];
1183                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1184
1185                 kfifo_free(&tcp_task->r2tqueue);
1186                 iscsi_pool_free(&tcp_task->r2tpool);
1187         }
1188 }
1189 EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_free);
1190
1191 int iscsi_tcp_set_max_r2t(struct iscsi_conn *conn, char *buf)
1192 {
1193         struct iscsi_session *session = conn->session;
1194         unsigned short r2ts = 0;
1195
1196         sscanf(buf, "%hu", &r2ts);
1197         if (session->max_r2t == r2ts)
1198                 return 0;
1199
1200         if (!r2ts || !is_power_of_2(r2ts))
1201                 return -EINVAL;
1202
1203         session->max_r2t = r2ts;
1204         iscsi_tcp_r2tpool_free(session);
1205         return iscsi_tcp_r2tpool_alloc(session);
1206 }
1207 EXPORT_SYMBOL_GPL(iscsi_tcp_set_max_r2t);
1208
1209 void iscsi_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
1210                               struct iscsi_stats *stats)
1211 {
1212         struct iscsi_conn *conn = cls_conn->dd_data;
1213
1214         stats->txdata_octets = conn->txdata_octets;
1215         stats->rxdata_octets = conn->rxdata_octets;
1216         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1217         stats->dataout_pdus = conn->dataout_pdus_cnt;
1218         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1219         stats->datain_pdus = conn->datain_pdus_cnt;
1220         stats->r2t_pdus = conn->r2t_pdus_cnt;
1221         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1222         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1223 }
1224 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_get_stats);