~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/ceph/osd_client.c

Version: ~ [ linux-5.16 ] ~ [ linux-5.15.13 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.90 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.170 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.224 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.261 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.296 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.298 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 
  2 #include <linux/ceph/ceph_debug.h>
  3 
  4 #include <linux/module.h>
  5 #include <linux/err.h>
  6 #include <linux/highmem.h>
  7 #include <linux/mm.h>
  8 #include <linux/pagemap.h>
  9 #include <linux/slab.h>
 10 #include <linux/uaccess.h>
 11 #ifdef CONFIG_BLOCK
 12 #include <linux/bio.h>
 13 #endif
 14 
 15 #include <linux/ceph/libceph.h>
 16 #include <linux/ceph/osd_client.h>
 17 #include <linux/ceph/messenger.h>
 18 #include <linux/ceph/decode.h>
 19 #include <linux/ceph/auth.h>
 20 #include <linux/ceph/pagelist.h>
 21 
 22 #define OSD_OP_FRONT_LEN        4096
 23 #define OSD_OPREPLY_FRONT_LEN   512
 24 
 25 static struct kmem_cache        *ceph_osd_request_cache;
 26 
 27 static const struct ceph_connection_operations osd_con_ops;
 28 
 29 static void __send_queued(struct ceph_osd_client *osdc);
 30 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd);
 31 static void __register_request(struct ceph_osd_client *osdc,
 32                                struct ceph_osd_request *req);
 33 static void __unregister_linger_request(struct ceph_osd_client *osdc,
 34                                         struct ceph_osd_request *req);
 35 static void __send_request(struct ceph_osd_client *osdc,
 36                            struct ceph_osd_request *req);
 37 
 38 /*
 39  * Implement client access to distributed object storage cluster.
 40  *
 41  * All data objects are stored within a cluster/cloud of OSDs, or
 42  * "object storage devices."  (Note that Ceph OSDs have _nothing_ to
 43  * do with the T10 OSD extensions to SCSI.)  Ceph OSDs are simply
 44  * remote daemons serving up and coordinating consistent and safe
 45  * access to storage.
 46  *
 47  * Cluster membership and the mapping of data objects onto storage devices
 48  * are described by the osd map.
 49  *
 50  * We keep track of pending OSD requests (read, write), resubmit
 51  * requests to different OSDs when the cluster topology/data layout
 52  * change, or retry the affected requests when the communications
 53  * channel with an OSD is reset.
 54  */
 55 
 56 /*
 57  * calculate the mapping of a file extent onto an object, and fill out the
 58  * request accordingly.  shorten extent as necessary if it crosses an
 59  * object boundary.
 60  *
 61  * fill osd op in request message.
 62  */
 63 static int calc_layout(struct ceph_file_layout *layout, u64 off, u64 *plen,
 64                         u64 *objnum, u64 *objoff, u64 *objlen)
 65 {
 66         u64 orig_len = *plen;
 67         int r;
 68 
 69         /* object extent? */
 70         r = ceph_calc_file_object_mapping(layout, off, orig_len, objnum,
 71                                           objoff, objlen);
 72         if (r < 0)
 73                 return r;
 74         if (*objlen < orig_len) {
 75                 *plen = *objlen;
 76                 dout(" skipping last %llu, final file extent %llu~%llu\n",
 77                      orig_len - *plen, off, *plen);
 78         }
 79 
 80         dout("calc_layout objnum=%llx %llu~%llu\n", *objnum, *objoff, *objlen);
 81 
 82         return 0;
 83 }
 84 
 85 static void ceph_osd_data_init(struct ceph_osd_data *osd_data)
 86 {
 87         memset(osd_data, 0, sizeof (*osd_data));
 88         osd_data->type = CEPH_OSD_DATA_TYPE_NONE;
 89 }
 90 
 91 static void ceph_osd_data_pages_init(struct ceph_osd_data *osd_data,
 92                         struct page **pages, u64 length, u32 alignment,
 93                         bool pages_from_pool, bool own_pages)
 94 {
 95         osd_data->type = CEPH_OSD_DATA_TYPE_PAGES;
 96         osd_data->pages = pages;
 97         osd_data->length = length;
 98         osd_data->alignment = alignment;
 99         osd_data->pages_from_pool = pages_from_pool;
100         osd_data->own_pages = own_pages;
101 }
102 
103 static void ceph_osd_data_pagelist_init(struct ceph_osd_data *osd_data,
104                         struct ceph_pagelist *pagelist)
105 {
106         osd_data->type = CEPH_OSD_DATA_TYPE_PAGELIST;
107         osd_data->pagelist = pagelist;
108 }
109 
110 #ifdef CONFIG_BLOCK
111 static void ceph_osd_data_bio_init(struct ceph_osd_data *osd_data,
112                         struct bio *bio, size_t bio_length)
113 {
114         osd_data->type = CEPH_OSD_DATA_TYPE_BIO;
115         osd_data->bio = bio;
116         osd_data->bio_length = bio_length;
117 }
118 #endif /* CONFIG_BLOCK */
119 
120 #define osd_req_op_data(oreq, whch, typ, fld)   \
121         ({                                              \
122                 BUG_ON(whch >= (oreq)->r_num_ops);      \
123                 &(oreq)->r_ops[whch].typ.fld;           \
124         })
125 
126 static struct ceph_osd_data *
127 osd_req_op_raw_data_in(struct ceph_osd_request *osd_req, unsigned int which)
128 {
129         BUG_ON(which >= osd_req->r_num_ops);
130 
131         return &osd_req->r_ops[which].raw_data_in;
132 }
133 
134 struct ceph_osd_data *
135 osd_req_op_extent_osd_data(struct ceph_osd_request *osd_req,
136                         unsigned int which)
137 {
138         return osd_req_op_data(osd_req, which, extent, osd_data);
139 }
140 EXPORT_SYMBOL(osd_req_op_extent_osd_data);
141 
142 struct ceph_osd_data *
143 osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
144                         unsigned int which)
145 {
146         return osd_req_op_data(osd_req, which, cls, response_data);
147 }
148 EXPORT_SYMBOL(osd_req_op_cls_response_data);    /* ??? */
149 
150 void osd_req_op_raw_data_in_pages(struct ceph_osd_request *osd_req,
151                         unsigned int which, struct page **pages,
152                         u64 length, u32 alignment,
153                         bool pages_from_pool, bool own_pages)
154 {
155         struct ceph_osd_data *osd_data;
156 
157         osd_data = osd_req_op_raw_data_in(osd_req, which);
158         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
159                                 pages_from_pool, own_pages);
160 }
161 EXPORT_SYMBOL(osd_req_op_raw_data_in_pages);
162 
163 void osd_req_op_extent_osd_data_pages(struct ceph_osd_request *osd_req,
164                         unsigned int which, struct page **pages,
165                         u64 length, u32 alignment,
166                         bool pages_from_pool, bool own_pages)
167 {
168         struct ceph_osd_data *osd_data;
169 
170         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
171         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
172                                 pages_from_pool, own_pages);
173 }
174 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pages);
175 
176 void osd_req_op_extent_osd_data_pagelist(struct ceph_osd_request *osd_req,
177                         unsigned int which, struct ceph_pagelist *pagelist)
178 {
179         struct ceph_osd_data *osd_data;
180 
181         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
182         ceph_osd_data_pagelist_init(osd_data, pagelist);
183 }
184 EXPORT_SYMBOL(osd_req_op_extent_osd_data_pagelist);
185 
186 #ifdef CONFIG_BLOCK
187 void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *osd_req,
188                         unsigned int which, struct bio *bio, size_t bio_length)
189 {
190         struct ceph_osd_data *osd_data;
191 
192         osd_data = osd_req_op_data(osd_req, which, extent, osd_data);
193         ceph_osd_data_bio_init(osd_data, bio, bio_length);
194 }
195 EXPORT_SYMBOL(osd_req_op_extent_osd_data_bio);
196 #endif /* CONFIG_BLOCK */
197 
198 static void osd_req_op_cls_request_info_pagelist(
199                         struct ceph_osd_request *osd_req,
200                         unsigned int which, struct ceph_pagelist *pagelist)
201 {
202         struct ceph_osd_data *osd_data;
203 
204         osd_data = osd_req_op_data(osd_req, which, cls, request_info);
205         ceph_osd_data_pagelist_init(osd_data, pagelist);
206 }
207 
208 void osd_req_op_cls_request_data_pagelist(
209                         struct ceph_osd_request *osd_req,
210                         unsigned int which, struct ceph_pagelist *pagelist)
211 {
212         struct ceph_osd_data *osd_data;
213 
214         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
215         ceph_osd_data_pagelist_init(osd_data, pagelist);
216 }
217 EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
218 
219 void osd_req_op_cls_request_data_pages(struct ceph_osd_request *osd_req,
220                         unsigned int which, struct page **pages, u64 length,
221                         u32 alignment, bool pages_from_pool, bool own_pages)
222 {
223         struct ceph_osd_data *osd_data;
224 
225         osd_data = osd_req_op_data(osd_req, which, cls, request_data);
226         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
227                                 pages_from_pool, own_pages);
228 }
229 EXPORT_SYMBOL(osd_req_op_cls_request_data_pages);
230 
231 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
232                         unsigned int which, struct page **pages, u64 length,
233                         u32 alignment, bool pages_from_pool, bool own_pages)
234 {
235         struct ceph_osd_data *osd_data;
236 
237         osd_data = osd_req_op_data(osd_req, which, cls, response_data);
238         ceph_osd_data_pages_init(osd_data, pages, length, alignment,
239                                 pages_from_pool, own_pages);
240 }
241 EXPORT_SYMBOL(osd_req_op_cls_response_data_pages);
242 
243 static u64 ceph_osd_data_length(struct ceph_osd_data *osd_data)
244 {
245         switch (osd_data->type) {
246         case CEPH_OSD_DATA_TYPE_NONE:
247                 return 0;
248         case CEPH_OSD_DATA_TYPE_PAGES:
249                 return osd_data->length;
250         case CEPH_OSD_DATA_TYPE_PAGELIST:
251                 return (u64)osd_data->pagelist->length;
252 #ifdef CONFIG_BLOCK
253         case CEPH_OSD_DATA_TYPE_BIO:
254                 return (u64)osd_data->bio_length;
255 #endif /* CONFIG_BLOCK */
256         default:
257                 WARN(true, "unrecognized data type %d\n", (int)osd_data->type);
258                 return 0;
259         }
260 }
261 
262 static void ceph_osd_data_release(struct ceph_osd_data *osd_data)
263 {
264         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES && osd_data->own_pages) {
265                 int num_pages;
266 
267                 num_pages = calc_pages_for((u64)osd_data->alignment,
268                                                 (u64)osd_data->length);
269                 ceph_release_page_vector(osd_data->pages, num_pages);
270         }
271         ceph_osd_data_init(osd_data);
272 }
273 
274 static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
275                         unsigned int which)
276 {
277         struct ceph_osd_req_op *op;
278 
279         BUG_ON(which >= osd_req->r_num_ops);
280         op = &osd_req->r_ops[which];
281 
282         switch (op->op) {
283         case CEPH_OSD_OP_READ:
284         case CEPH_OSD_OP_WRITE:
285                 ceph_osd_data_release(&op->extent.osd_data);
286                 break;
287         case CEPH_OSD_OP_CALL:
288                 ceph_osd_data_release(&op->cls.request_info);
289                 ceph_osd_data_release(&op->cls.request_data);
290                 ceph_osd_data_release(&op->cls.response_data);
291                 break;
292         default:
293                 break;
294         }
295 }
296 
297 /*
298  * requests
299  */
300 void ceph_osdc_release_request(struct kref *kref)
301 {
302         struct ceph_osd_request *req;
303         unsigned int which;
304 
305         req = container_of(kref, struct ceph_osd_request, r_kref);
306         if (req->r_request)
307                 ceph_msg_put(req->r_request);
308         if (req->r_reply) {
309                 ceph_msg_revoke_incoming(req->r_reply);
310                 ceph_msg_put(req->r_reply);
311         }
312 
313         for (which = 0; which < req->r_num_ops; which++)
314                 osd_req_op_data_release(req, which);
315 
316         ceph_put_snap_context(req->r_snapc);
317         if (req->r_mempool)
318                 mempool_free(req, req->r_osdc->req_mempool);
319         else
320                 kmem_cache_free(ceph_osd_request_cache, req);
321 
322 }
323 EXPORT_SYMBOL(ceph_osdc_release_request);
324 
325 struct ceph_osd_request *ceph_osdc_alloc_request(struct ceph_osd_client *osdc,
326                                                struct ceph_snap_context *snapc,
327                                                unsigned int num_ops,
328                                                bool use_mempool,
329                                                gfp_t gfp_flags)
330 {
331         struct ceph_osd_request *req;
332         struct ceph_msg *msg;
333         size_t msg_size;
334 
335         BUILD_BUG_ON(CEPH_OSD_MAX_OP > U16_MAX);
336         BUG_ON(num_ops > CEPH_OSD_MAX_OP);
337 
338         msg_size = 4 + 4 + 8 + 8 + 4+8;
339         msg_size += 2 + 4 + 8 + 4 + 4; /* oloc */
340         msg_size += 1 + 8 + 4 + 4;     /* pg_t */
341         msg_size += 4 + MAX_OBJ_NAME_SIZE;
342         msg_size += 2 + num_ops*sizeof(struct ceph_osd_op);
343         msg_size += 8;  /* snapid */
344         msg_size += 8;  /* snap_seq */
345         msg_size += 8 * (snapc ? snapc->num_snaps : 0);  /* snaps */
346         msg_size += 4;
347 
348         if (use_mempool) {
349                 req = mempool_alloc(osdc->req_mempool, gfp_flags);
350                 memset(req, 0, sizeof(*req));
351         } else {
352                 req = kmem_cache_zalloc(ceph_osd_request_cache, gfp_flags);
353         }
354         if (req == NULL)
355                 return NULL;
356 
357         req->r_osdc = osdc;
358         req->r_mempool = use_mempool;
359         req->r_num_ops = num_ops;
360 
361         kref_init(&req->r_kref);
362         init_completion(&req->r_completion);
363         init_completion(&req->r_safe_completion);
364         RB_CLEAR_NODE(&req->r_node);
365         INIT_LIST_HEAD(&req->r_unsafe_item);
366         INIT_LIST_HEAD(&req->r_linger_item);
367         INIT_LIST_HEAD(&req->r_linger_osd);
368         INIT_LIST_HEAD(&req->r_req_lru_item);
369         INIT_LIST_HEAD(&req->r_osd_item);
370 
371         /* create reply message */
372         if (use_mempool)
373                 msg = ceph_msgpool_get(&osdc->msgpool_op_reply, 0);
374         else
375                 msg = ceph_msg_new(CEPH_MSG_OSD_OPREPLY,
376                                    OSD_OPREPLY_FRONT_LEN, gfp_flags, true);
377         if (!msg) {
378                 ceph_osdc_put_request(req);
379                 return NULL;
380         }
381         req->r_reply = msg;
382 
383         /* create request message; allow space for oid */
384         if (use_mempool)
385                 msg = ceph_msgpool_get(&osdc->msgpool_op, 0);
386         else
387                 msg = ceph_msg_new(CEPH_MSG_OSD_OP, msg_size, gfp_flags, true);
388         if (!msg) {
389                 ceph_osdc_put_request(req);
390                 return NULL;
391         }
392 
393         memset(msg->front.iov_base, 0, msg->front.iov_len);
394 
395         req->r_request = msg;
396 
397         return req;
398 }
399 EXPORT_SYMBOL(ceph_osdc_alloc_request);
400 
401 static bool osd_req_opcode_valid(u16 opcode)
402 {
403         switch (opcode) {
404         case CEPH_OSD_OP_READ:
405         case CEPH_OSD_OP_STAT:
406         case CEPH_OSD_OP_MAPEXT:
407         case CEPH_OSD_OP_MASKTRUNC:
408         case CEPH_OSD_OP_SPARSE_READ:
409         case CEPH_OSD_OP_NOTIFY:
410         case CEPH_OSD_OP_NOTIFY_ACK:
411         case CEPH_OSD_OP_ASSERT_VER:
412         case CEPH_OSD_OP_WRITE:
413         case CEPH_OSD_OP_WRITEFULL:
414         case CEPH_OSD_OP_TRUNCATE:
415         case CEPH_OSD_OP_ZERO:
416         case CEPH_OSD_OP_DELETE:
417         case CEPH_OSD_OP_APPEND:
418         case CEPH_OSD_OP_STARTSYNC:
419         case CEPH_OSD_OP_SETTRUNC:
420         case CEPH_OSD_OP_TRIMTRUNC:
421         case CEPH_OSD_OP_TMAPUP:
422         case CEPH_OSD_OP_TMAPPUT:
423         case CEPH_OSD_OP_TMAPGET:
424         case CEPH_OSD_OP_CREATE:
425         case CEPH_OSD_OP_ROLLBACK:
426         case CEPH_OSD_OP_WATCH:
427         case CEPH_OSD_OP_OMAPGETKEYS:
428         case CEPH_OSD_OP_OMAPGETVALS:
429         case CEPH_OSD_OP_OMAPGETHEADER:
430         case CEPH_OSD_OP_OMAPGETVALSBYKEYS:
431         case CEPH_OSD_OP_OMAPSETVALS:
432         case CEPH_OSD_OP_OMAPSETHEADER:
433         case CEPH_OSD_OP_OMAPCLEAR:
434         case CEPH_OSD_OP_OMAPRMKEYS:
435         case CEPH_OSD_OP_OMAP_CMP:
436         case CEPH_OSD_OP_CLONERANGE:
437         case CEPH_OSD_OP_ASSERT_SRC_VERSION:
438         case CEPH_OSD_OP_SRC_CMPXATTR:
439         case CEPH_OSD_OP_GETXATTR:
440         case CEPH_OSD_OP_GETXATTRS:
441         case CEPH_OSD_OP_CMPXATTR:
442         case CEPH_OSD_OP_SETXATTR:
443         case CEPH_OSD_OP_SETXATTRS:
444         case CEPH_OSD_OP_RESETXATTRS:
445         case CEPH_OSD_OP_RMXATTR:
446         case CEPH_OSD_OP_PULL:
447         case CEPH_OSD_OP_PUSH:
448         case CEPH_OSD_OP_BALANCEREADS:
449         case CEPH_OSD_OP_UNBALANCEREADS:
450         case CEPH_OSD_OP_SCRUB:
451         case CEPH_OSD_OP_SCRUB_RESERVE:
452         case CEPH_OSD_OP_SCRUB_UNRESERVE:
453         case CEPH_OSD_OP_SCRUB_STOP:
454         case CEPH_OSD_OP_SCRUB_MAP:
455         case CEPH_OSD_OP_WRLOCK:
456         case CEPH_OSD_OP_WRUNLOCK:
457         case CEPH_OSD_OP_RDLOCK:
458         case CEPH_OSD_OP_RDUNLOCK:
459         case CEPH_OSD_OP_UPLOCK:
460         case CEPH_OSD_OP_DNLOCK:
461         case CEPH_OSD_OP_CALL:
462         case CEPH_OSD_OP_PGLS:
463         case CEPH_OSD_OP_PGLS_FILTER:
464                 return true;
465         default:
466                 return false;
467         }
468 }
469 
470 /*
471  * This is an osd op init function for opcodes that have no data or
472  * other information associated with them.  It also serves as a
473  * common init routine for all the other init functions, below.
474  */
475 static struct ceph_osd_req_op *
476 _osd_req_op_init(struct ceph_osd_request *osd_req, unsigned int which,
477                                 u16 opcode)
478 {
479         struct ceph_osd_req_op *op;
480 
481         BUG_ON(which >= osd_req->r_num_ops);
482         BUG_ON(!osd_req_opcode_valid(opcode));
483 
484         op = &osd_req->r_ops[which];
485         memset(op, 0, sizeof (*op));
486         op->op = opcode;
487 
488         return op;
489 }
490 
491 void osd_req_op_init(struct ceph_osd_request *osd_req,
492                                 unsigned int which, u16 opcode)
493 {
494         (void)_osd_req_op_init(osd_req, which, opcode);
495 }
496 EXPORT_SYMBOL(osd_req_op_init);
497 
498 void osd_req_op_extent_init(struct ceph_osd_request *osd_req,
499                                 unsigned int which, u16 opcode,
500                                 u64 offset, u64 length,
501                                 u64 truncate_size, u32 truncate_seq)
502 {
503         struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
504         size_t payload_len = 0;
505 
506         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
507                opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
508                opcode != CEPH_OSD_OP_TRUNCATE);
509 
510         op->extent.offset = offset;
511         op->extent.length = length;
512         op->extent.truncate_size = truncate_size;
513         op->extent.truncate_seq = truncate_seq;
514         if (opcode == CEPH_OSD_OP_WRITE)
515                 payload_len += length;
516 
517         op->payload_len = payload_len;
518 }
519 EXPORT_SYMBOL(osd_req_op_extent_init);
520 
521 void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
522                                 unsigned int which, u64 length)
523 {
524         struct ceph_osd_req_op *op;
525         u64 previous;
526 
527         BUG_ON(which >= osd_req->r_num_ops);
528         op = &osd_req->r_ops[which];
529         previous = op->extent.length;
530 
531         if (length == previous)
532                 return;         /* Nothing to do */
533         BUG_ON(length > previous);
534 
535         op->extent.length = length;
536         op->payload_len -= previous - length;
537 }
538 EXPORT_SYMBOL(osd_req_op_extent_update);
539 
540 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
541                         u16 opcode, const char *class, const char *method)
542 {
543         struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
544         struct ceph_pagelist *pagelist;
545         size_t payload_len = 0;
546         size_t size;
547 
548         BUG_ON(opcode != CEPH_OSD_OP_CALL);
549 
550         pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
551         BUG_ON(!pagelist);
552         ceph_pagelist_init(pagelist);
553 
554         op->cls.class_name = class;
555         size = strlen(class);
556         BUG_ON(size > (size_t) U8_MAX);
557         op->cls.class_len = size;
558         ceph_pagelist_append(pagelist, class, size);
559         payload_len += size;
560 
561         op->cls.method_name = method;
562         size = strlen(method);
563         BUG_ON(size > (size_t) U8_MAX);
564         op->cls.method_len = size;
565         ceph_pagelist_append(pagelist, method, size);
566         payload_len += size;
567 
568         osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
569 
570         op->cls.argc = 0;       /* currently unused */
571 
572         op->payload_len = payload_len;
573 }
574 EXPORT_SYMBOL(osd_req_op_cls_init);
575 
576 void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
577                                 unsigned int which, u16 opcode,
578                                 u64 cookie, u64 version, int flag)
579 {
580         struct ceph_osd_req_op *op = _osd_req_op_init(osd_req, which, opcode);
581 
582         BUG_ON(opcode != CEPH_OSD_OP_NOTIFY_ACK && opcode != CEPH_OSD_OP_WATCH);
583 
584         op->watch.cookie = cookie;
585         op->watch.ver = version;
586         if (opcode == CEPH_OSD_OP_WATCH && flag)
587                 op->watch.flag = (u8)1;
588 }
589 EXPORT_SYMBOL(osd_req_op_watch_init);
590 
591 static void ceph_osdc_msg_data_add(struct ceph_msg *msg,
592                                 struct ceph_osd_data *osd_data)
593 {
594         u64 length = ceph_osd_data_length(osd_data);
595 
596         if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
597                 BUG_ON(length > (u64) SIZE_MAX);
598                 if (length)
599                         ceph_msg_data_add_pages(msg, osd_data->pages,
600                                         length, osd_data->alignment);
601         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGELIST) {
602                 BUG_ON(!length);
603                 ceph_msg_data_add_pagelist(msg, osd_data->pagelist);
604 #ifdef CONFIG_BLOCK
605         } else if (osd_data->type == CEPH_OSD_DATA_TYPE_BIO) {
606                 ceph_msg_data_add_bio(msg, osd_data->bio, length);
607 #endif
608         } else {
609                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_NONE);
610         }
611 }
612 
613 static u64 osd_req_encode_op(struct ceph_osd_request *req,
614                               struct ceph_osd_op *dst, unsigned int which)
615 {
616         struct ceph_osd_req_op *src;
617         struct ceph_osd_data *osd_data;
618         u64 request_data_len = 0;
619         u64 data_length;
620 
621         BUG_ON(which >= req->r_num_ops);
622         src = &req->r_ops[which];
623         if (WARN_ON(!osd_req_opcode_valid(src->op))) {
624                 pr_err("unrecognized osd opcode %d\n", src->op);
625 
626                 return 0;
627         }
628 
629         switch (src->op) {
630         case CEPH_OSD_OP_STAT:
631                 osd_data = &src->raw_data_in;
632                 ceph_osdc_msg_data_add(req->r_reply, osd_data);
633                 break;
634         case CEPH_OSD_OP_READ:
635         case CEPH_OSD_OP_WRITE:
636         case CEPH_OSD_OP_ZERO:
637         case CEPH_OSD_OP_DELETE:
638         case CEPH_OSD_OP_TRUNCATE:
639                 if (src->op == CEPH_OSD_OP_WRITE)
640                         request_data_len = src->extent.length;
641                 dst->extent.offset = cpu_to_le64(src->extent.offset);
642                 dst->extent.length = cpu_to_le64(src->extent.length);
643                 dst->extent.truncate_size =
644                         cpu_to_le64(src->extent.truncate_size);
645                 dst->extent.truncate_seq =
646                         cpu_to_le32(src->extent.truncate_seq);
647                 osd_data = &src->extent.osd_data;
648                 if (src->op == CEPH_OSD_OP_WRITE)
649                         ceph_osdc_msg_data_add(req->r_request, osd_data);
650                 else
651                         ceph_osdc_msg_data_add(req->r_reply, osd_data);
652                 break;
653         case CEPH_OSD_OP_CALL:
654                 dst->cls.class_len = src->cls.class_len;
655                 dst->cls.method_len = src->cls.method_len;
656                 osd_data = &src->cls.request_info;
657                 ceph_osdc_msg_data_add(req->r_request, osd_data);
658                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
659                 request_data_len = osd_data->pagelist->length;
660 
661                 osd_data = &src->cls.request_data;
662                 data_length = ceph_osd_data_length(osd_data);
663                 if (data_length) {
664                         BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
665                         dst->cls.indata_len = cpu_to_le32(data_length);
666                         ceph_osdc_msg_data_add(req->r_request, osd_data);
667                         src->payload_len += data_length;
668                         request_data_len += data_length;
669                 }
670                 osd_data = &src->cls.response_data;
671                 ceph_osdc_msg_data_add(req->r_reply, osd_data);
672                 break;
673         case CEPH_OSD_OP_STARTSYNC:
674                 break;
675         case CEPH_OSD_OP_NOTIFY_ACK:
676         case CEPH_OSD_OP_WATCH:
677                 dst->watch.cookie = cpu_to_le64(src->watch.cookie);
678                 dst->watch.ver = cpu_to_le64(src->watch.ver);
679                 dst->watch.flag = src->watch.flag;
680                 break;
681         default:
682                 pr_err("unsupported osd opcode %s\n",
683                         ceph_osd_op_name(src->op));
684                 WARN_ON(1);
685 
686                 return 0;
687         }
688         dst->op = cpu_to_le16(src->op);
689         dst->payload_len = cpu_to_le32(src->payload_len);
690 
691         return request_data_len;
692 }
693 
694 /*
695  * build new request AND message, calculate layout, and adjust file
696  * extent as needed.
697  *
698  * if the file was recently truncated, we include information about its
699  * old and new size so that the object can be updated appropriately.  (we
700  * avoid synchronously deleting truncated objects because it's slow.)
701  *
702  * if @do_sync, include a 'startsync' command so that the osd will flush
703  * data quickly.
704  */
705 struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
706                                                struct ceph_file_layout *layout,
707                                                struct ceph_vino vino,
708                                                u64 off, u64 *plen, int num_ops,
709                                                int opcode, int flags,
710                                                struct ceph_snap_context *snapc,
711                                                u32 truncate_seq,
712                                                u64 truncate_size,
713                                                bool use_mempool)
714 {
715         struct ceph_osd_request *req;
716         u64 objnum = 0;
717         u64 objoff = 0;
718         u64 objlen = 0;
719         u32 object_size;
720         u64 object_base;
721         int r;
722 
723         BUG_ON(opcode != CEPH_OSD_OP_READ && opcode != CEPH_OSD_OP_WRITE &&
724                opcode != CEPH_OSD_OP_DELETE && opcode != CEPH_OSD_OP_ZERO &&
725                opcode != CEPH_OSD_OP_TRUNCATE);
726 
727         req = ceph_osdc_alloc_request(osdc, snapc, num_ops, use_mempool,
728                                         GFP_NOFS);
729         if (!req)
730                 return ERR_PTR(-ENOMEM);
731 
732         req->r_flags = flags;
733 
734         /* calculate max write size */
735         r = calc_layout(layout, off, plen, &objnum, &objoff, &objlen);
736         if (r < 0) {
737                 ceph_osdc_put_request(req);
738                 return ERR_PTR(r);
739         }
740 
741         object_size = le32_to_cpu(layout->fl_object_size);
742         object_base = off - objoff;
743         if (!(truncate_seq == 1 && truncate_size == -1ULL)) {
744                 if (truncate_size <= object_base) {
745                         truncate_size = 0;
746                 } else {
747                         truncate_size -= object_base;
748                         if (truncate_size > object_size)
749                                 truncate_size = object_size;
750                 }
751         }
752 
753         osd_req_op_extent_init(req, 0, opcode, objoff, objlen,
754                                 truncate_size, truncate_seq);
755 
756         /*
757          * A second op in the ops array means the caller wants to
758          * also issue a include a 'startsync' command so that the
759          * osd will flush data quickly.
760          */
761         if (num_ops > 1)
762                 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
763 
764         req->r_file_layout = *layout;  /* keep a copy */
765 
766         snprintf(req->r_oid, sizeof(req->r_oid), "%llx.%08llx",
767                 vino.ino, objnum);
768         req->r_oid_len = strlen(req->r_oid);
769 
770         return req;
771 }
772 EXPORT_SYMBOL(ceph_osdc_new_request);
773 
774 /*
775  * We keep osd requests in an rbtree, sorted by ->r_tid.
776  */
777 static void __insert_request(struct ceph_osd_client *osdc,
778                              struct ceph_osd_request *new)
779 {
780         struct rb_node **p = &osdc->requests.rb_node;
781         struct rb_node *parent = NULL;
782         struct ceph_osd_request *req = NULL;
783 
784         while (*p) {
785                 parent = *p;
786                 req = rb_entry(parent, struct ceph_osd_request, r_node);
787                 if (new->r_tid < req->r_tid)
788                         p = &(*p)->rb_left;
789                 else if (new->r_tid > req->r_tid)
790                         p = &(*p)->rb_right;
791                 else
792                         BUG();
793         }
794 
795         rb_link_node(&new->r_node, parent, p);
796         rb_insert_color(&new->r_node, &osdc->requests);
797 }
798 
799 static struct ceph_osd_request *__lookup_request(struct ceph_osd_client *osdc,
800                                                  u64 tid)
801 {
802         struct ceph_osd_request *req;
803         struct rb_node *n = osdc->requests.rb_node;
804 
805         while (n) {
806                 req = rb_entry(n, struct ceph_osd_request, r_node);
807                 if (tid < req->r_tid)
808                         n = n->rb_left;
809                 else if (tid > req->r_tid)
810                         n = n->rb_right;
811                 else
812                         return req;
813         }
814         return NULL;
815 }
816 
817 static struct ceph_osd_request *
818 __lookup_request_ge(struct ceph_osd_client *osdc,
819                     u64 tid)
820 {
821         struct ceph_osd_request *req;
822         struct rb_node *n = osdc->requests.rb_node;
823 
824         while (n) {
825                 req = rb_entry(n, struct ceph_osd_request, r_node);
826                 if (tid < req->r_tid) {
827                         if (!n->rb_left)
828                                 return req;
829                         n = n->rb_left;
830                 } else if (tid > req->r_tid) {
831                         n = n->rb_right;
832                 } else {
833                         return req;
834                 }
835         }
836         return NULL;
837 }
838 
839 /*
840  * Resubmit requests pending on the given osd.
841  */
842 static void __kick_osd_requests(struct ceph_osd_client *osdc,
843                                 struct ceph_osd *osd)
844 {
845         struct ceph_osd_request *req, *nreq;
846         LIST_HEAD(resend);
847         int err;
848 
849         dout("__kick_osd_requests osd%d\n", osd->o_osd);
850         err = __reset_osd(osdc, osd);
851         if (err)
852                 return;
853         /*
854          * Build up a list of requests to resend by traversing the
855          * osd's list of requests.  Requests for a given object are
856          * sent in tid order, and that is also the order they're
857          * kept on this list.  Therefore all requests that are in
858          * flight will be found first, followed by all requests that
859          * have not yet been sent.  And to resend requests while
860          * preserving this order we will want to put any sent
861          * requests back on the front of the osd client's unsent
862          * list.
863          *
864          * So we build a separate ordered list of already-sent
865          * requests for the affected osd and splice it onto the
866          * front of the osd client's unsent list.  Once we've seen a
867          * request that has not yet been sent we're done.  Those
868          * requests are already sitting right where they belong.
869          */
870         list_for_each_entry(req, &osd->o_requests, r_osd_item) {
871                 if (!req->r_sent)
872                         break;
873                 list_move_tail(&req->r_req_lru_item, &resend);
874                 dout("requeueing %p tid %llu osd%d\n", req, req->r_tid,
875                      osd->o_osd);
876                 if (!req->r_linger)
877                         req->r_flags |= CEPH_OSD_FLAG_RETRY;
878         }
879         list_splice(&resend, &osdc->req_unsent);
880 
881         /*
882          * Linger requests are re-registered before sending, which
883          * sets up a new tid for each.  We add them to the unsent
884          * list at the end to keep things in tid order.
885          */
886         list_for_each_entry_safe(req, nreq, &osd->o_linger_requests,
887                                  r_linger_osd) {
888                 /*
889                  * reregister request prior to unregistering linger so
890                  * that r_osd is preserved.
891                  */
892                 BUG_ON(!list_empty(&req->r_req_lru_item));
893                 __register_request(osdc, req);
894                 list_add_tail(&req->r_req_lru_item, &osdc->req_unsent);
895                 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
896                 __unregister_linger_request(osdc, req);
897                 dout("requeued lingering %p tid %llu osd%d\n", req, req->r_tid,
898                      osd->o_osd);
899         }
900 }
901 
902 /*
903  * If the osd connection drops, we need to resubmit all requests.
904  */
905 static void osd_reset(struct ceph_connection *con)
906 {
907         struct ceph_osd *osd = con->private;
908         struct ceph_osd_client *osdc;
909 
910         if (!osd)
911                 return;
912         dout("osd_reset osd%d\n", osd->o_osd);
913         osdc = osd->o_osdc;
914         down_read(&osdc->map_sem);
915         mutex_lock(&osdc->request_mutex);
916         __kick_osd_requests(osdc, osd);
917         __send_queued(osdc);
918         mutex_unlock(&osdc->request_mutex);
919         up_read(&osdc->map_sem);
920 }
921 
922 /*
923  * Track open sessions with osds.
924  */
925 static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum)
926 {
927         struct ceph_osd *osd;
928 
929         osd = kzalloc(sizeof(*osd), GFP_NOFS);
930         if (!osd)
931                 return NULL;
932 
933         atomic_set(&osd->o_ref, 1);
934         osd->o_osdc = osdc;
935         osd->o_osd = onum;
936         RB_CLEAR_NODE(&osd->o_node);
937         INIT_LIST_HEAD(&osd->o_requests);
938         INIT_LIST_HEAD(&osd->o_linger_requests);
939         INIT_LIST_HEAD(&osd->o_osd_lru);
940         osd->o_incarnation = 1;
941 
942         ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr);
943 
944         INIT_LIST_HEAD(&osd->o_keepalive_item);
945         return osd;
946 }
947 
948 static struct ceph_osd *get_osd(struct ceph_osd *osd)
949 {
950         if (atomic_inc_not_zero(&osd->o_ref)) {
951                 dout("get_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref)-1,
952                      atomic_read(&osd->o_ref));
953                 return osd;
954         } else {
955                 dout("get_osd %p FAIL\n", osd);
956                 return NULL;
957         }
958 }
959 
960 static void put_osd(struct ceph_osd *osd)
961 {
962         dout("put_osd %p %d -> %d\n", osd, atomic_read(&osd->o_ref),
963              atomic_read(&osd->o_ref) - 1);
964         if (atomic_dec_and_test(&osd->o_ref) && osd->o_auth.authorizer) {
965                 struct ceph_auth_client *ac = osd->o_osdc->client->monc.auth;
966 
967                 ceph_auth_destroy_authorizer(ac, osd->o_auth.authorizer);
968                 kfree(osd);
969         }
970 }
971 
972 /*
973  * remove an osd from our map
974  */
975 static void __remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
976 {
977         dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
978         WARN_ON(!list_empty(&osd->o_requests));
979         WARN_ON(!list_empty(&osd->o_linger_requests));
980 
981         list_del_init(&osd->o_osd_lru);
982         rb_erase(&osd->o_node, &osdc->osds);
983         RB_CLEAR_NODE(&osd->o_node);
984 }
985 
986 static void remove_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
987 {
988         dout("%s %p osd%d\n", __func__, osd, osd->o_osd);
989 
990         if (!RB_EMPTY_NODE(&osd->o_node)) {
991                 ceph_con_close(&osd->o_con);
992                 __remove_osd(osdc, osd);
993                 put_osd(osd);
994         }
995 }
996 
997 static void remove_all_osds(struct ceph_osd_client *osdc)
998 {
999         dout("%s %p\n", __func__, osdc);
1000         mutex_lock(&osdc->request_mutex);
1001         while (!RB_EMPTY_ROOT(&osdc->osds)) {
1002                 struct ceph_osd *osd = rb_entry(rb_first(&osdc->osds),
1003                                                 struct ceph_osd, o_node);
1004                 remove_osd(osdc, osd);
1005         }
1006         mutex_unlock(&osdc->request_mutex);
1007 }
1008 
1009 static void __move_osd_to_lru(struct ceph_osd_client *osdc,
1010                               struct ceph_osd *osd)
1011 {
1012         dout("__move_osd_to_lru %p\n", osd);
1013         BUG_ON(!list_empty(&osd->o_osd_lru));
1014         list_add_tail(&osd->o_osd_lru, &osdc->osd_lru);
1015         osd->lru_ttl = jiffies + osdc->client->options->osd_idle_ttl * HZ;
1016 }
1017 
1018 static void __remove_osd_from_lru(struct ceph_osd *osd)
1019 {
1020         dout("__remove_osd_from_lru %p\n", osd);
1021         if (!list_empty(&osd->o_osd_lru))
1022                 list_del_init(&osd->o_osd_lru);
1023 }
1024 
1025 static void remove_old_osds(struct ceph_osd_client *osdc)
1026 {
1027         struct ceph_osd *osd, *nosd;
1028 
1029         dout("__remove_old_osds %p\n", osdc);
1030         mutex_lock(&osdc->request_mutex);
1031         list_for_each_entry_safe(osd, nosd, &osdc->osd_lru, o_osd_lru) {
1032                 if (time_before(jiffies, osd->lru_ttl))
1033                         break;
1034                 remove_osd(osdc, osd);
1035         }
1036         mutex_unlock(&osdc->request_mutex);
1037 }
1038 
1039 /*
1040  * reset osd connect
1041  */
1042 static int __reset_osd(struct ceph_osd_client *osdc, struct ceph_osd *osd)
1043 {
1044         struct ceph_entity_addr *peer_addr;
1045 
1046         dout("__reset_osd %p osd%d\n", osd, osd->o_osd);
1047         if (list_empty(&osd->o_requests) &&
1048             list_empty(&osd->o_linger_requests)) {
1049                 remove_osd(osdc, osd);
1050                 return -ENODEV;
1051         }
1052 
1053         peer_addr = &osdc->osdmap->osd_addr[osd->o_osd];
1054         if (!memcmp(peer_addr, &osd->o_con.peer_addr, sizeof (*peer_addr)) &&
1055                         !ceph_con_opened(&osd->o_con)) {
1056                 struct ceph_osd_request *req;
1057 
1058                 dout(" osd addr hasn't changed and connection never opened,"
1059                      " letting msgr retry");
1060                 /* touch each r_stamp for handle_timeout()'s benfit */
1061                 list_for_each_entry(req, &osd->o_requests, r_osd_item)
1062                         req->r_stamp = jiffies;
1063 
1064                 return -EAGAIN;
1065         }
1066 
1067         ceph_con_close(&osd->o_con);
1068         ceph_con_open(&osd->o_con, CEPH_ENTITY_TYPE_OSD, osd->o_osd, peer_addr);
1069         osd->o_incarnation++;
1070 
1071         return 0;
1072 }
1073 
1074 static void __insert_osd(struct ceph_osd_client *osdc, struct ceph_osd *new)
1075 {
1076         struct rb_node **p = &osdc->osds.rb_node;
1077         struct rb_node *parent = NULL;
1078         struct ceph_osd *osd = NULL;
1079 
1080         dout("__insert_osd %p osd%d\n", new, new->o_osd);
1081         while (*p) {
1082                 parent = *p;
1083                 osd = rb_entry(parent, struct ceph_osd, o_node);
1084                 if (new->o_osd < osd->o_osd)
1085                         p = &(*p)->rb_left;
1086                 else if (new->o_osd > osd->o_osd)
1087                         p = &(*p)->rb_right;
1088                 else
1089                         BUG();
1090         }
1091 
1092         rb_link_node(&new->o_node, parent, p);
1093         rb_insert_color(&new->o_node, &osdc->osds);
1094 }
1095 
1096 static struct ceph_osd *__lookup_osd(struct ceph_osd_client *osdc, int o)
1097 {
1098         struct ceph_osd *osd;
1099         struct rb_node *n = osdc->osds.rb_node;
1100 
1101         while (n) {
1102                 osd = rb_entry(n, struct ceph_osd, o_node);
1103                 if (o < osd->o_osd)
1104                         n = n->rb_left;
1105                 else if (o > osd->o_osd)
1106                         n = n->rb_right;
1107                 else
1108                         return osd;
1109         }
1110         return NULL;
1111 }
1112 
1113 static void __schedule_osd_timeout(struct ceph_osd_client *osdc)
1114 {
1115         schedule_delayed_work(&osdc->timeout_work,
1116                         osdc->client->options->osd_keepalive_timeout * HZ);
1117 }
1118 
1119 static void __cancel_osd_timeout(struct ceph_osd_client *osdc)
1120 {
1121         cancel_delayed_work(&osdc->timeout_work);
1122 }
1123 
1124 /*
1125  * Register request, assign tid.  If this is the first request, set up
1126  * the timeout event.
1127  */
1128 static void __register_request(struct ceph_osd_client *osdc,
1129                                struct ceph_osd_request *req)
1130 {
1131         req->r_tid = ++osdc->last_tid;
1132         req->r_request->hdr.tid = cpu_to_le64(req->r_tid);
1133         dout("__register_request %p tid %lld\n", req, req->r_tid);
1134         __insert_request(osdc, req);
1135         ceph_osdc_get_request(req);
1136         osdc->num_requests++;
1137         if (osdc->num_requests == 1) {
1138                 dout(" first request, scheduling timeout\n");
1139                 __schedule_osd_timeout(osdc);
1140         }
1141 }
1142 
1143 /*
1144  * called under osdc->request_mutex
1145  */
1146 static void __unregister_request(struct ceph_osd_client *osdc,
1147                                  struct ceph_osd_request *req)
1148 {
1149         if (RB_EMPTY_NODE(&req->r_node)) {
1150                 dout("__unregister_request %p tid %lld not registered\n",
1151                         req, req->r_tid);
1152                 return;
1153         }
1154 
1155         dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1156         rb_erase(&req->r_node, &osdc->requests);
1157         osdc->num_requests--;
1158 
1159         if (req->r_osd) {
1160                 /* make sure the original request isn't in flight. */
1161                 ceph_msg_revoke(req->r_request);
1162 
1163                 list_del_init(&req->r_osd_item);
1164                 if (list_empty(&req->r_osd->o_requests) &&
1165                     list_empty(&req->r_osd->o_linger_requests)) {
1166                         dout("moving osd to %p lru\n", req->r_osd);
1167                         __move_osd_to_lru(osdc, req->r_osd);
1168                 }
1169                 if (list_empty(&req->r_linger_item))
1170                         req->r_osd = NULL;
1171         }
1172 
1173         list_del_init(&req->r_req_lru_item);
1174         ceph_osdc_put_request(req);
1175 
1176         if (osdc->num_requests == 0) {
1177                 dout(" no requests, canceling timeout\n");
1178                 __cancel_osd_timeout(osdc);
1179         }
1180 }
1181 
1182 /*
1183  * Cancel a previously queued request message
1184  */
1185 static void __cancel_request(struct ceph_osd_request *req)
1186 {
1187         if (req->r_sent && req->r_osd) {
1188                 ceph_msg_revoke(req->r_request);
1189                 req->r_sent = 0;
1190         }
1191 }
1192 
1193 static void __register_linger_request(struct ceph_osd_client *osdc,
1194                                     struct ceph_osd_request *req)
1195 {
1196         dout("__register_linger_request %p\n", req);
1197         ceph_osdc_get_request(req);
1198         list_add_tail(&req->r_linger_item, &osdc->req_linger);
1199         if (req->r_osd)
1200                 list_add_tail(&req->r_linger_osd,
1201                               &req->r_osd->o_linger_requests);
1202 }
1203 
1204 static void __unregister_linger_request(struct ceph_osd_client *osdc,
1205                                         struct ceph_osd_request *req)
1206 {
1207         dout("__unregister_linger_request %p\n", req);
1208         list_del_init(&req->r_linger_item);
1209         if (req->r_osd) {
1210                 list_del_init(&req->r_linger_osd);
1211 
1212                 if (list_empty(&req->r_osd->o_requests) &&
1213                     list_empty(&req->r_osd->o_linger_requests)) {
1214                         dout("moving osd to %p lru\n", req->r_osd);
1215                         __move_osd_to_lru(osdc, req->r_osd);
1216                 }
1217                 if (list_empty(&req->r_osd_item))
1218                         req->r_osd = NULL;
1219         }
1220         ceph_osdc_put_request(req);
1221 }
1222 
1223 void ceph_osdc_unregister_linger_request(struct ceph_osd_client *osdc,
1224                                          struct ceph_osd_request *req)
1225 {
1226         mutex_lock(&osdc->request_mutex);
1227         if (req->r_linger) {
1228                 req->r_linger = 0;
1229                 __unregister_linger_request(osdc, req);
1230         }
1231         mutex_unlock(&osdc->request_mutex);
1232 }
1233 EXPORT_SYMBOL(ceph_osdc_unregister_linger_request);
1234 
1235 void ceph_osdc_set_request_linger(struct ceph_osd_client *osdc,
1236                                   struct ceph_osd_request *req)
1237 {
1238         if (!req->r_linger) {
1239                 dout("set_request_linger %p\n", req);
1240                 req->r_linger = 1;
1241         }
1242 }
1243 EXPORT_SYMBOL(ceph_osdc_set_request_linger);
1244 
1245 /*
1246  * Returns whether a request should be blocked from being sent
1247  * based on the current osdmap and osd_client settings.
1248  *
1249  * Caller should hold map_sem for read.
1250  */
1251 static bool __req_should_be_paused(struct ceph_osd_client *osdc,
1252                                    struct ceph_osd_request *req)
1253 {
1254         bool pauserd = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD);
1255         bool pausewr = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR) ||
1256                 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1257         return (req->r_flags & CEPH_OSD_FLAG_READ && pauserd) ||
1258                 (req->r_flags & CEPH_OSD_FLAG_WRITE && pausewr);
1259 }
1260 
1261 /*
1262  * Pick an osd (the first 'up' osd in the pg), allocate the osd struct
1263  * (as needed), and set the request r_osd appropriately.  If there is
1264  * no up osd, set r_osd to NULL.  Move the request to the appropriate list
1265  * (unsent, homeless) or leave on in-flight lru.
1266  *
1267  * Return 0 if unchanged, 1 if changed, or negative on error.
1268  *
1269  * Caller should hold map_sem for read and request_mutex.
1270  */
1271 static int __map_request(struct ceph_osd_client *osdc,
1272                          struct ceph_osd_request *req, int force_resend)
1273 {
1274         struct ceph_pg pgid;
1275         int acting[CEPH_PG_MAX_SIZE];
1276         int o = -1, num = 0;
1277         int err;
1278         bool was_paused;
1279 
1280         dout("map_request %p tid %lld\n", req, req->r_tid);
1281         err = ceph_calc_ceph_pg(&pgid, req->r_oid, osdc->osdmap,
1282                                 ceph_file_layout_pg_pool(req->r_file_layout));
1283         if (err) {
1284                 list_move(&req->r_req_lru_item, &osdc->req_notarget);
1285                 return err;
1286         }
1287         req->r_pgid = pgid;
1288 
1289         err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
1290         if (err > 0) {
1291                 o = acting[0];
1292                 num = err;
1293         }
1294 
1295         was_paused = req->r_paused;
1296         req->r_paused = __req_should_be_paused(osdc, req);
1297         if (was_paused && !req->r_paused)
1298                 force_resend = 1;
1299 
1300         if ((!force_resend &&
1301              req->r_osd && req->r_osd->o_osd == o &&
1302              req->r_sent >= req->r_osd->o_incarnation &&
1303              req->r_num_pg_osds == num &&
1304              memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
1305             (req->r_osd == NULL && o == -1) ||
1306             req->r_paused)
1307                 return 0;  /* no change */
1308 
1309         dout("map_request tid %llu pgid %lld.%x osd%d (was osd%d)\n",
1310              req->r_tid, pgid.pool, pgid.seed, o,
1311              req->r_osd ? req->r_osd->o_osd : -1);
1312 
1313         /* record full pg acting set */
1314         memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
1315         req->r_num_pg_osds = num;
1316 
1317         if (req->r_osd) {
1318                 __cancel_request(req);
1319                 list_del_init(&req->r_osd_item);
1320                 req->r_osd = NULL;
1321         }
1322 
1323         req->r_osd = __lookup_osd(osdc, o);
1324         if (!req->r_osd && o >= 0) {
1325                 err = -ENOMEM;
1326                 req->r_osd = create_osd(osdc, o);
1327                 if (!req->r_osd) {
1328                         list_move(&req->r_req_lru_item, &osdc->req_notarget);
1329                         goto out;
1330                 }
1331 
1332                 dout("map_request osd %p is osd%d\n", req->r_osd, o);
1333                 __insert_osd(osdc, req->r_osd);
1334 
1335                 ceph_con_open(&req->r_osd->o_con,
1336                               CEPH_ENTITY_TYPE_OSD, o,
1337                               &osdc->osdmap->osd_addr[o]);
1338         }
1339 
1340         if (req->r_osd) {
1341                 __remove_osd_from_lru(req->r_osd);
1342                 list_add_tail(&req->r_osd_item, &req->r_osd->o_requests);
1343                 list_move_tail(&req->r_req_lru_item, &osdc->req_unsent);
1344         } else {
1345                 list_move_tail(&req->r_req_lru_item, &osdc->req_notarget);
1346         }
1347         err = 1;   /* osd or pg changed */
1348 
1349 out:
1350         return err;
1351 }
1352 
1353 /*
1354  * caller should hold map_sem (for read) and request_mutex
1355  */
1356 static void __send_request(struct ceph_osd_client *osdc,
1357                            struct ceph_osd_request *req)
1358 {
1359         void *p;
1360 
1361         dout("send_request %p tid %llu to osd%d flags %d pg %lld.%x\n",
1362              req, req->r_tid, req->r_osd->o_osd, req->r_flags,
1363              (unsigned long long)req->r_pgid.pool, req->r_pgid.seed);
1364 
1365         /* fill in message content that changes each time we send it */
1366         put_unaligned_le32(osdc->osdmap->epoch, req->r_request_osdmap_epoch);
1367         put_unaligned_le32(req->r_flags, req->r_request_flags);
1368         put_unaligned_le64(req->r_pgid.pool, req->r_request_pool);
1369         p = req->r_request_pgid;
1370         ceph_encode_64(&p, req->r_pgid.pool);
1371         ceph_encode_32(&p, req->r_pgid.seed);
1372         put_unaligned_le64(1, req->r_request_attempts);  /* FIXME */
1373         memcpy(req->r_request_reassert_version, &req->r_reassert_version,
1374                sizeof(req->r_reassert_version));
1375 
1376         req->r_stamp = jiffies;
1377         list_move_tail(&req->r_req_lru_item, &osdc->req_lru);
1378 
1379         ceph_msg_get(req->r_request); /* send consumes a ref */
1380 
1381         req->r_sent = req->r_osd->o_incarnation;
1382 
1383         ceph_con_send(&req->r_osd->o_con, req->r_request);
1384 }
1385 
1386 /*
1387  * Send any requests in the queue (req_unsent).
1388  */
1389 static void __send_queued(struct ceph_osd_client *osdc)
1390 {
1391         struct ceph_osd_request *req, *tmp;
1392 
1393         dout("__send_queued\n");
1394         list_for_each_entry_safe(req, tmp, &osdc->req_unsent, r_req_lru_item)
1395                 __send_request(osdc, req);
1396 }
1397 
1398 /*
1399  * Timeout callback, called every N seconds when 1 or more osd
1400  * requests has been active for more than N seconds.  When this
1401  * happens, we ping all OSDs with requests who have timed out to
1402  * ensure any communications channel reset is detected.  Reset the
1403  * request timeouts another N seconds in the future as we go.
1404  * Reschedule the timeout event another N seconds in future (unless
1405  * there are no open requests).
1406  */
1407 static void handle_timeout(struct work_struct *work)
1408 {
1409         struct ceph_osd_client *osdc =
1410                 container_of(work, struct ceph_osd_client, timeout_work.work);
1411         struct ceph_osd_request *req;
1412         struct ceph_osd *osd;
1413         unsigned long keepalive =
1414                 osdc->client->options->osd_keepalive_timeout * HZ;
1415         struct list_head slow_osds;
1416         dout("timeout\n");
1417         down_read(&osdc->map_sem);
1418 
1419         ceph_monc_request_next_osdmap(&osdc->client->monc);
1420 
1421         mutex_lock(&osdc->request_mutex);
1422 
1423         /*
1424          * ping osds that are a bit slow.  this ensures that if there
1425          * is a break in the TCP connection we will notice, and reopen
1426          * a connection with that osd (from the fault callback).
1427          */
1428         INIT_LIST_HEAD(&slow_osds);
1429         list_for_each_entry(req, &osdc->req_lru, r_req_lru_item) {
1430                 if (time_before(jiffies, req->r_stamp + keepalive))
1431                         break;
1432 
1433                 osd = req->r_osd;
1434                 BUG_ON(!osd);
1435                 dout(" tid %llu is slow, will send keepalive on osd%d\n",
1436                      req->r_tid, osd->o_osd);
1437                 list_move_tail(&osd->o_keepalive_item, &slow_osds);
1438         }
1439         while (!list_empty(&slow_osds)) {
1440                 osd = list_entry(slow_osds.next, struct ceph_osd,
1441                                  o_keepalive_item);
1442                 list_del_init(&osd->o_keepalive_item);
1443                 ceph_con_keepalive(&osd->o_con);
1444         }
1445 
1446         __schedule_osd_timeout(osdc);
1447         __send_queued(osdc);
1448         mutex_unlock(&osdc->request_mutex);
1449         up_read(&osdc->map_sem);
1450 }
1451 
1452 static void handle_osds_timeout(struct work_struct *work)
1453 {
1454         struct ceph_osd_client *osdc =
1455                 container_of(work, struct ceph_osd_client,
1456                              osds_timeout_work.work);
1457         unsigned long delay =
1458                 osdc->client->options->osd_idle_ttl * HZ >> 2;
1459 
1460         dout("osds timeout\n");
1461         down_read(&osdc->map_sem);
1462         remove_old_osds(osdc);
1463         up_read(&osdc->map_sem);
1464 
1465         schedule_delayed_work(&osdc->osds_timeout_work,
1466                               round_jiffies_relative(delay));
1467 }
1468 
1469 static void complete_request(struct ceph_osd_request *req)
1470 {
1471         complete_all(&req->r_safe_completion);  /* fsync waiter */
1472 }
1473 
1474 /*
1475  * handle osd op reply.  either call the callback if it is specified,
1476  * or do the completion to wake up the waiting thread.
1477  */
1478 static void handle_reply(struct ceph_osd_client *osdc, struct ceph_msg *msg,
1479                          struct ceph_connection *con)
1480 {
1481         void *p, *end;
1482         struct ceph_osd_request *req;
1483         u64 tid;
1484         int object_len;
1485         unsigned int numops;
1486         int payload_len, flags;
1487         s32 result;
1488         s32 retry_attempt;
1489         struct ceph_pg pg;
1490         int err;
1491         u32 reassert_epoch;
1492         u64 reassert_version;
1493         u32 osdmap_epoch;
1494         int already_completed;
1495         u32 bytes;
1496         unsigned int i;
1497 
1498         tid = le64_to_cpu(msg->hdr.tid);
1499         dout("handle_reply %p tid %llu\n", msg, tid);
1500 
1501         p = msg->front.iov_base;
1502         end = p + msg->front.iov_len;
1503 
1504         ceph_decode_need(&p, end, 4, bad);
1505         object_len = ceph_decode_32(&p);
1506         ceph_decode_need(&p, end, object_len, bad);
1507         p += object_len;
1508 
1509         err = ceph_decode_pgid(&p, end, &pg);
1510         if (err)
1511                 goto bad;
1512 
1513         ceph_decode_need(&p, end, 8 + 4 + 4 + 8 + 4, bad);
1514         flags = ceph_decode_64(&p);
1515         result = ceph_decode_32(&p);
1516         reassert_epoch = ceph_decode_32(&p);
1517         reassert_version = ceph_decode_64(&p);
1518         osdmap_epoch = ceph_decode_32(&p);
1519 
1520         /* lookup */
1521         mutex_lock(&osdc->request_mutex);
1522         req = __lookup_request(osdc, tid);
1523         if (req == NULL) {
1524                 dout("handle_reply tid %llu dne\n", tid);
1525                 goto bad_mutex;
1526         }
1527         ceph_osdc_get_request(req);
1528 
1529         dout("handle_reply %p tid %llu req %p result %d\n", msg, tid,
1530              req, result);
1531 
1532         ceph_decode_need(&p, end, 4, bad_put);
1533         numops = ceph_decode_32(&p);
1534         if (numops > CEPH_OSD_MAX_OP)
1535                 goto bad_put;
1536         if (numops != req->r_num_ops)
1537                 goto bad_put;
1538         payload_len = 0;
1539         ceph_decode_need(&p, end, numops * sizeof(struct ceph_osd_op), bad_put);
1540         for (i = 0; i < numops; i++) {
1541                 struct ceph_osd_op *op = p;
1542                 int len;
1543 
1544                 len = le32_to_cpu(op->payload_len);
1545                 req->r_reply_op_len[i] = len;
1546                 dout(" op %d has %d bytes\n", i, len);
1547                 payload_len += len;
1548                 p += sizeof(*op);
1549         }
1550         bytes = le32_to_cpu(msg->hdr.data_len);
1551         if (payload_len != bytes) {
1552                 pr_warning("sum of op payload lens %d != data_len %d",
1553                            payload_len, bytes);
1554                 goto bad_put;
1555         }
1556 
1557         ceph_decode_need(&p, end, 4 + numops * 4, bad_put);
1558         retry_attempt = ceph_decode_32(&p);
1559         for (i = 0; i < numops; i++)
1560                 req->r_reply_op_result[i] = ceph_decode_32(&p);
1561 
1562         already_completed = req->r_got_reply;
1563 
1564         if (!req->r_got_reply) {
1565 
1566                 req->r_result = result;
1567                 dout("handle_reply result %d bytes %d\n", req->r_result,
1568                      bytes);
1569                 if (req->r_result == 0)
1570                         req->r_result = bytes;
1571 
1572                 /* in case this is a write and we need to replay, */
1573                 req->r_reassert_version.epoch = cpu_to_le32(reassert_epoch);
1574                 req->r_reassert_version.version = cpu_to_le64(reassert_version);
1575 
1576                 req->r_got_reply = 1;
1577         } else if ((flags & CEPH_OSD_FLAG_ONDISK) == 0) {
1578                 dout("handle_reply tid %llu dup ack\n", tid);
1579                 mutex_unlock(&osdc->request_mutex);
1580                 goto done;
1581         }
1582 
1583         dout("handle_reply tid %llu flags %d\n", tid, flags);
1584 
1585         if (req->r_linger && (flags & CEPH_OSD_FLAG_ONDISK))
1586                 __register_linger_request(osdc, req);
1587 
1588         /* either this is a read, or we got the safe response */
1589         if (result < 0 ||
1590             (flags & CEPH_OSD_FLAG_ONDISK) ||
1591             ((flags & CEPH_OSD_FLAG_WRITE) == 0))
1592                 __unregister_request(osdc, req);
1593 
1594         mutex_unlock(&osdc->request_mutex);
1595 
1596         if (!already_completed) {
1597                 if (req->r_unsafe_callback &&
1598                     result >= 0 && !(flags & CEPH_OSD_FLAG_ONDISK))
1599                         req->r_unsafe_callback(req, true);
1600                 if (req->r_callback)
1601                         req->r_callback(req, msg);
1602                 else
1603                         complete_all(&req->r_completion);
1604         }
1605 
1606         if (flags & CEPH_OSD_FLAG_ONDISK) {
1607                 if (req->r_unsafe_callback && already_completed)
1608                         req->r_unsafe_callback(req, false);
1609                 complete_request(req);
1610         }
1611 
1612 done:
1613         dout("req=%p req->r_linger=%d\n", req, req->r_linger);
1614         ceph_osdc_put_request(req);
1615         return;
1616 
1617 bad_put:
1618         ceph_osdc_put_request(req);
1619 bad_mutex:
1620         mutex_unlock(&osdc->request_mutex);
1621 bad:
1622         pr_err("corrupt osd_op_reply got %d %d\n",
1623                (int)msg->front.iov_len, le32_to_cpu(msg->hdr.front_len));
1624         ceph_msg_dump(msg);
1625 }
1626 
1627 static void reset_changed_osds(struct ceph_osd_client *osdc)
1628 {
1629         struct rb_node *p, *n;
1630 
1631         dout("%s %p\n", __func__, osdc);
1632         for (p = rb_first(&osdc->osds); p; p = n) {
1633                 struct ceph_osd *osd = rb_entry(p, struct ceph_osd, o_node);
1634 
1635                 n = rb_next(p);
1636                 if (!ceph_osd_is_up(osdc->osdmap, osd->o_osd) ||
1637                     memcmp(&osd->o_con.peer_addr,
1638                            ceph_osd_addr(osdc->osdmap,
1639                                          osd->o_osd),
1640                            sizeof(struct ceph_entity_addr)) != 0)
1641                         __reset_osd(osdc, osd);
1642         }
1643 }
1644 
1645 /*
1646  * Requeue requests whose mapping to an OSD has changed.  If requests map to
1647  * no osd, request a new map.
1648  *
1649  * Caller should hold map_sem for read.
1650  */
1651 static void kick_requests(struct ceph_osd_client *osdc, bool force_resend,
1652                           bool force_resend_writes)
1653 {
1654         struct ceph_osd_request *req, *nreq;
1655         struct rb_node *p;
1656         int needmap = 0;
1657         int err;
1658         bool force_resend_req;
1659 
1660         dout("kick_requests %s %s\n", force_resend ? " (force resend)" : "",
1661                 force_resend_writes ? " (force resend writes)" : "");
1662         mutex_lock(&osdc->request_mutex);
1663         for (p = rb_first(&osdc->requests); p; ) {
1664                 req = rb_entry(p, struct ceph_osd_request, r_node);
1665                 p = rb_next(p);
1666 
1667                 /*
1668                  * For linger requests that have not yet been
1669                  * registered, move them to the linger list; they'll
1670                  * be sent to the osd in the loop below.  Unregister
1671                  * the request before re-registering it as a linger
1672                  * request to ensure the __map_request() below
1673                  * will decide it needs to be sent.
1674                  */
1675                 if (req->r_linger && list_empty(&req->r_linger_item)) {
1676                         dout("%p tid %llu restart on osd%d\n",
1677                              req, req->r_tid,
1678                              req->r_osd ? req->r_osd->o_osd : -1);
1679                         ceph_osdc_get_request(req);
1680                         __unregister_request(osdc, req);
1681                         __register_linger_request(osdc, req);
1682                         ceph_osdc_put_request(req);
1683                         continue;
1684                 }
1685 
1686                 force_resend_req = force_resend ||
1687                         (force_resend_writes &&
1688                                 req->r_flags & CEPH_OSD_FLAG_WRITE);
1689                 err = __map_request(osdc, req, force_resend_req);
1690                 if (err < 0)
1691                         continue;  /* error */
1692                 if (req->r_osd == NULL) {
1693                         dout("%p tid %llu maps to no osd\n", req, req->r_tid);
1694                         needmap++;  /* request a newer map */
1695                 } else if (err > 0) {
1696                         if (!req->r_linger) {
1697                                 dout("%p tid %llu requeued on osd%d\n", req,
1698                                      req->r_tid,
1699                                      req->r_osd ? req->r_osd->o_osd : -1);
1700                                 req->r_flags |= CEPH_OSD_FLAG_RETRY;
1701                         }
1702                 }
1703         }
1704 
1705         list_for_each_entry_safe(req, nreq, &osdc->req_linger,
1706                                  r_linger_item) {
1707                 dout("linger req=%p req->r_osd=%p\n", req, req->r_osd);
1708 
1709                 err = __map_request(osdc, req,
1710                                     force_resend || force_resend_writes);
1711                 dout("__map_request returned %d\n", err);
1712                 if (err < 0)
1713                         continue;  /* hrm! */
1714                 if (req->r_osd == NULL || err > 0) {
1715                         if (req->r_osd == NULL) {
1716                                 dout("lingering %p tid %llu maps to no osd\n",
1717                                      req, req->r_tid);
1718                                 /*
1719                                  * A homeless lingering request makes
1720                                  * no sense, as it's job is to keep
1721                                  * a particular OSD connection open.
1722                                  * Request a newer map and kick the
1723                                  * request, knowing that it won't be
1724                                  * resent until we actually get a map
1725                                  * that can tell us where to send it.
1726                                  */
1727                                 needmap++;
1728                         }
1729 
1730                         dout("kicking lingering %p tid %llu osd%d\n", req,
1731                              req->r_tid, req->r_osd ? req->r_osd->o_osd : -1);
1732                         __register_request(osdc, req);
1733                         __unregister_linger_request(osdc, req);
1734                 }
1735         }
1736         reset_changed_osds(osdc);
1737         mutex_unlock(&osdc->request_mutex);
1738 
1739         if (needmap) {
1740                 dout("%d requests for down osds, need new map\n", needmap);
1741                 ceph_monc_request_next_osdmap(&osdc->client->monc);
1742         }
1743 }
1744 
1745 
1746 /*
1747  * Process updated osd map.
1748  *
1749  * The message contains any number of incremental and full maps, normally
1750  * indicating some sort of topology change in the cluster.  Kick requests
1751  * off to different OSDs as needed.
1752  */
1753 void ceph_osdc_handle_map(struct ceph_osd_client *osdc, struct ceph_msg *msg)
1754 {
1755         void *p, *end, *next;
1756         u32 nr_maps, maplen;
1757         u32 epoch;
1758         struct ceph_osdmap *newmap = NULL, *oldmap;
1759         int err;
1760         struct ceph_fsid fsid;
1761         bool was_full;
1762 
1763         dout("handle_map have %u\n", osdc->osdmap ? osdc->osdmap->epoch : 0);
1764         p = msg->front.iov_base;
1765         end = p + msg->front.iov_len;
1766 
1767         /* verify fsid */
1768         ceph_decode_need(&p, end, sizeof(fsid), bad);
1769         ceph_decode_copy(&p, &fsid, sizeof(fsid));
1770         if (ceph_check_fsid(osdc->client, &fsid) < 0)
1771                 return;
1772 
1773         down_write(&osdc->map_sem);
1774 
1775         was_full = ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL);
1776 
1777         /* incremental maps */
1778         ceph_decode_32_safe(&p, end, nr_maps, bad);
1779         dout(" %d inc maps\n", nr_maps);
1780         while (nr_maps > 0) {
1781                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1782                 epoch = ceph_decode_32(&p);
1783                 maplen = ceph_decode_32(&p);
1784                 ceph_decode_need(&p, end, maplen, bad);
1785                 next = p + maplen;
1786                 if (osdc->osdmap && osdc->osdmap->epoch+1 == epoch) {
1787                         dout("applying incremental map %u len %d\n",
1788                              epoch, maplen);
1789                         newmap = osdmap_apply_incremental(&p, next,
1790                                                           osdc->osdmap,
1791                                                           &osdc->client->msgr);
1792                         if (IS_ERR(newmap)) {
1793                                 err = PTR_ERR(newmap);
1794                                 goto bad;
1795                         }
1796                         BUG_ON(!newmap);
1797                         if (newmap != osdc->osdmap) {
1798                                 ceph_osdmap_destroy(osdc->osdmap);
1799                                 osdc->osdmap = newmap;
1800                         }
1801                         was_full = was_full ||
1802                                 ceph_osdmap_flag(osdc->osdmap,
1803                                                  CEPH_OSDMAP_FULL);
1804                         kick_requests(osdc, 0, was_full);
1805                 } else {
1806                         dout("ignoring incremental map %u len %d\n",
1807                              epoch, maplen);
1808                 }
1809                 p = next;
1810                 nr_maps--;
1811         }
1812         if (newmap)
1813                 goto done;
1814 
1815         /* full maps */
1816         ceph_decode_32_safe(&p, end, nr_maps, bad);
1817         dout(" %d full maps\n", nr_maps);
1818         while (nr_maps) {
1819                 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
1820                 epoch = ceph_decode_32(&p);
1821                 maplen = ceph_decode_32(&p);
1822                 ceph_decode_need(&p, end, maplen, bad);
1823                 if (nr_maps > 1) {
1824                         dout("skipping non-latest full map %u len %d\n",
1825                              epoch, maplen);
1826                 } else if (osdc->osdmap && osdc->osdmap->epoch >= epoch) {
1827                         dout("skipping full map %u len %d, "
1828                              "older than our %u\n", epoch, maplen,
1829                              osdc->osdmap->epoch);
1830                 } else {
1831                         int skipped_map = 0;
1832 
1833                         dout("taking full map %u len %d\n", epoch, maplen);
1834                         newmap = osdmap_decode(&p, p+maplen);
1835                         if (IS_ERR(newmap)) {
1836                                 err = PTR_ERR(newmap);
1837                                 goto bad;
1838                         }
1839                         BUG_ON(!newmap);
1840                         oldmap = osdc->osdmap;
1841                         osdc->osdmap = newmap;
1842                         if (oldmap) {
1843                                 if (oldmap->epoch + 1 < newmap->epoch)
1844                                         skipped_map = 1;
1845                                 ceph_osdmap_destroy(oldmap);
1846                         }
1847                         was_full = was_full ||
1848                                 ceph_osdmap_flag(osdc->osdmap,
1849                                                  CEPH_OSDMAP_FULL);
1850                         kick_requests(osdc, skipped_map, was_full);
1851                 }
1852                 p += maplen;
1853                 nr_maps--;
1854         }
1855 
1856         if (!osdc->osdmap)
1857                 goto bad;
1858 done:
1859         downgrade_write(&osdc->map_sem);
1860         ceph_monc_got_osdmap(&osdc->client->monc, osdc->osdmap->epoch);
1861 
1862         /*
1863          * subscribe to subsequent osdmap updates if full to ensure
1864          * we find out when we are no longer full and stop returning
1865          * ENOSPC.
1866          */
1867         if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) ||
1868                 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSERD) ||
1869                 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_PAUSEWR))
1870                 ceph_monc_request_next_osdmap(&osdc->client->monc);
1871 
1872         mutex_lock(&osdc->request_mutex);
1873         __send_queued(osdc);
1874         mutex_unlock(&osdc->request_mutex);
1875         up_read(&osdc->map_sem);
1876         wake_up_all(&osdc->client->auth_wq);
1877         return;
1878 
1879 bad:
1880         pr_err("osdc handle_map corrupt msg\n");
1881         ceph_msg_dump(msg);
1882         up_write(&osdc->map_sem);
1883         return;
1884 }
1885 
1886 /*
1887  * watch/notify callback event infrastructure
1888  *
1889  * These callbacks are used both for watch and notify operations.
1890  */
1891 static void __release_event(struct kref *kref)
1892 {
1893         struct ceph_osd_event *event =
1894                 container_of(kref, struct ceph_osd_event, kref);
1895 
1896         dout("__release_event %p\n", event);
1897         kfree(event);
1898 }
1899 
1900 static void get_event(struct ceph_osd_event *event)
1901 {
1902         kref_get(&event->kref);
1903 }
1904 
1905 void ceph_osdc_put_event(struct ceph_osd_event *event)
1906 {
1907         kref_put(&event->kref, __release_event);
1908 }
1909 EXPORT_SYMBOL(ceph_osdc_put_event);
1910 
1911 static void __insert_event(struct ceph_osd_client *osdc,
1912                              struct ceph_osd_event *new)
1913 {
1914         struct rb_node **p = &osdc->event_tree.rb_node;
1915         struct rb_node *parent = NULL;
1916         struct ceph_osd_event *event = NULL;
1917 
1918         while (*p) {
1919                 parent = *p;
1920                 event = rb_entry(parent, struct ceph_osd_event, node);
1921                 if (new->cookie < event->cookie)
1922                         p = &(*p)->rb_left;
1923                 else if (new->cookie > event->cookie)
1924                         p = &(*p)->rb_right;
1925                 else
1926                         BUG();
1927         }
1928 
1929         rb_link_node(&new->node, parent, p);
1930         rb_insert_color(&new->node, &osdc->event_tree);
1931 }
1932 
1933 static struct ceph_osd_event *__find_event(struct ceph_osd_client *osdc,
1934                                                 u64 cookie)
1935 {
1936         struct rb_node **p = &osdc->event_tree.rb_node;
1937         struct rb_node *parent = NULL;
1938         struct ceph_osd_event *event = NULL;
1939 
1940         while (*p) {
1941                 parent = *p;
1942                 event = rb_entry(parent, struct ceph_osd_event, node);
1943                 if (cookie < event->cookie)
1944                         p = &(*p)->rb_left;
1945                 else if (cookie > event->cookie)
1946                         p = &(*p)->rb_right;
1947                 else
1948                         return event;
1949         }
1950         return NULL;
1951 }
1952 
1953 static void __remove_event(struct ceph_osd_event *event)
1954 {
1955         struct ceph_osd_client *osdc = event->osdc;
1956 
1957         if (!RB_EMPTY_NODE(&event->node)) {
1958                 dout("__remove_event removed %p\n", event);
1959                 rb_erase(&event->node, &osdc->event_tree);
1960                 ceph_osdc_put_event(event);
1961         } else {
1962                 dout("__remove_event didn't remove %p\n", event);
1963         }
1964 }
1965 
1966 int ceph_osdc_create_event(struct ceph_osd_client *osdc,
1967                            void (*event_cb)(u64, u64, u8, void *),
1968                            void *data, struct ceph_osd_event **pevent)
1969 {
1970         struct ceph_osd_event *event;
1971 
1972         event = kmalloc(sizeof(*event), GFP_NOIO);
1973         if (!event)
1974                 return -ENOMEM;
1975 
1976         dout("create_event %p\n", event);
1977         event->cb = event_cb;
1978         event->one_shot = 0;
1979         event->data = data;
1980         event->osdc = osdc;
1981         INIT_LIST_HEAD(&event->osd_node);
1982         RB_CLEAR_NODE(&event->node);
1983         kref_init(&event->kref);   /* one ref for us */
1984         kref_get(&event->kref);    /* one ref for the caller */
1985 
1986         spin_lock(&osdc->event_lock);
1987         event->cookie = ++osdc->event_count;
1988         __insert_event(osdc, event);
1989         spin_unlock(&osdc->event_lock);
1990 
1991         *pevent = event;
1992         return 0;
1993 }
1994 EXPORT_SYMBOL(ceph_osdc_create_event);
1995 
1996 void ceph_osdc_cancel_event(struct ceph_osd_event *event)
1997 {
1998         struct ceph_osd_client *osdc = event->osdc;
1999 
2000         dout("cancel_event %p\n", event);
2001         spin_lock(&osdc->event_lock);
2002         __remove_event(event);
2003         spin_unlock(&osdc->event_lock);
2004         ceph_osdc_put_event(event); /* caller's */
2005 }
2006 EXPORT_SYMBOL(ceph_osdc_cancel_event);
2007 
2008 
2009 static void do_event_work(struct work_struct *work)
2010 {
2011         struct ceph_osd_event_work *event_work =
2012                 container_of(work, struct ceph_osd_event_work, work);
2013         struct ceph_osd_event *event = event_work->event;
2014         u64 ver = event_work->ver;
2015         u64 notify_id = event_work->notify_id;
2016         u8 opcode = event_work->opcode;
2017 
2018         dout("do_event_work completing %p\n", event);
2019         event->cb(ver, notify_id, opcode, event->data);
2020         dout("do_event_work completed %p\n", event);
2021         ceph_osdc_put_event(event);
2022         kfree(event_work);
2023 }
2024 
2025 
2026 /*
2027  * Process osd watch notifications
2028  */
2029 static void handle_watch_notify(struct ceph_osd_client *osdc,
2030                                 struct ceph_msg *msg)
2031 {
2032         void *p, *end;
2033         u8 proto_ver;
2034         u64 cookie, ver, notify_id;
2035         u8 opcode;
2036         struct ceph_osd_event *event;
2037         struct ceph_osd_event_work *event_work;
2038 
2039         p = msg->front.iov_base;
2040         end = p + msg->front.iov_len;
2041 
2042         ceph_decode_8_safe(&p, end, proto_ver, bad);
2043         ceph_decode_8_safe(&p, end, opcode, bad);
2044         ceph_decode_64_safe(&p, end, cookie, bad);
2045         ceph_decode_64_safe(&p, end, ver, bad);
2046         ceph_decode_64_safe(&p, end, notify_id, bad);
2047 
2048         spin_lock(&osdc->event_lock);
2049         event = __find_event(osdc, cookie);
2050         if (event) {
2051                 BUG_ON(event->one_shot);
2052                 get_event(event);
2053         }
2054         spin_unlock(&osdc->event_lock);
2055         dout("handle_watch_notify cookie %lld ver %lld event %p\n",
2056              cookie, ver, event);
2057         if (event) {
2058                 event_work = kmalloc(sizeof(*event_work), GFP_NOIO);
2059                 if (!event_work) {
2060                         dout("ERROR: could not allocate event_work\n");
2061                         goto done_err;
2062                 }
2063                 INIT_WORK(&event_work->work, do_event_work);
2064                 event_work->event = event;
2065                 event_work->ver = ver;
2066                 event_work->notify_id = notify_id;
2067                 event_work->opcode = opcode;
2068                 if (!queue_work(osdc->notify_wq, &event_work->work)) {
2069                         dout("WARNING: failed to queue notify event work\n");
2070                         goto done_err;
2071                 }
2072         }
2073 
2074         return;
2075 
2076 done_err:
2077         ceph_osdc_put_event(event);
2078         return;
2079 
2080 bad:
2081         pr_err("osdc handle_watch_notify corrupt msg\n");
2082         return;
2083 }
2084 
2085 /*
2086  * build new request AND message
2087  *
2088  */
2089 void ceph_osdc_build_request(struct ceph_osd_request *req, u64 off,
2090                                 struct ceph_snap_context *snapc, u64 snap_id,
2091                                 struct timespec *mtime)
2092 {
2093         struct ceph_msg *msg = req->r_request;
2094         void *p;
2095         size_t msg_size;
2096         int flags = req->r_flags;
2097         u64 data_len;
2098         unsigned int i;
2099 
2100         req->r_snapid = snap_id;
2101         req->r_snapc = ceph_get_snap_context(snapc);
2102 
2103         /* encode request */
2104         msg->hdr.version = cpu_to_le16(4);
2105 
2106         p = msg->front.iov_base;
2107         ceph_encode_32(&p, 1);   /* client_inc  is always 1 */
2108         req->r_request_osdmap_epoch = p;
2109         p += 4;
2110         req->r_request_flags = p;
2111         p += 4;
2112         if (req->r_flags & CEPH_OSD_FLAG_WRITE)
2113                 ceph_encode_timespec(p, mtime);
2114         p += sizeof(struct ceph_timespec);
2115         req->r_request_reassert_version = p;
2116         p += sizeof(struct ceph_eversion); /* will get filled in */
2117 
2118         /* oloc */
2119         ceph_encode_8(&p, 4);
2120         ceph_encode_8(&p, 4);
2121         ceph_encode_32(&p, 8 + 4 + 4);
2122         req->r_request_pool = p;
2123         p += 8;
2124         ceph_encode_32(&p, -1);  /* preferred */
2125         ceph_encode_32(&p, 0);   /* key len */
2126 
2127         ceph_encode_8(&p, 1);
2128         req->r_request_pgid = p;
2129         p += 8 + 4;
2130         ceph_encode_32(&p, -1);  /* preferred */
2131 
2132         /* oid */
2133         ceph_encode_32(&p, req->r_oid_len);
2134         memcpy(p, req->r_oid, req->r_oid_len);
2135         dout("oid '%.*s' len %d\n", req->r_oid_len, req->r_oid, req->r_oid_len);
2136         p += req->r_oid_len;
2137 
2138         /* ops--can imply data */
2139         ceph_encode_16(&p, (u16)req->r_num_ops);
2140         data_len = 0;
2141         for (i = 0; i < req->r_num_ops; i++) {
2142                 data_len += osd_req_encode_op(req, p, i);
2143                 p += sizeof(struct ceph_osd_op);
2144         }
2145 
2146         /* snaps */
2147         ceph_encode_64(&p, req->r_snapid);
2148         ceph_encode_64(&p, req->r_snapc ? req->r_snapc->seq : 0);
2149         ceph_encode_32(&p, req->r_snapc ? req->r_snapc->num_snaps : 0);
2150         if (req->r_snapc) {
2151                 for (i = 0; i < snapc->num_snaps; i++) {
2152                         ceph_encode_64(&p, req->r_snapc->snaps[i]);
2153                 }
2154         }
2155 
2156         req->r_request_attempts = p;
2157         p += 4;
2158 
2159         /* data */
2160         if (flags & CEPH_OSD_FLAG_WRITE) {
2161                 u16 data_off;
2162 
2163                 /*
2164                  * The header "data_off" is a hint to the receiver
2165                  * allowing it to align received data into its
2166                  * buffers such that there's no need to re-copy
2167                  * it before writing it to disk (direct I/O).
2168                  */
2169                 data_off = (u16) (off & 0xffff);
2170                 req->r_request->hdr.data_off = cpu_to_le16(data_off);
2171         }
2172         req->r_request->hdr.data_len = cpu_to_le32(data_len);
2173 
2174         BUG_ON(p > msg->front.iov_base + msg->front.iov_len);
2175         msg_size = p - msg->front.iov_base;
2176         msg->front.iov_len = msg_size;
2177         msg->hdr.front_len = cpu_to_le32(msg_size);
2178 
2179         dout("build_request msg_size was %d\n", (int)msg_size);
2180 }
2181 EXPORT_SYMBOL(ceph_osdc_build_request);
2182 
2183 /*
2184  * Register request, send initial attempt.
2185  */
2186 int ceph_osdc_start_request(struct ceph_osd_client *osdc,
2187                             struct ceph_osd_request *req,
2188                             bool nofail)
2189 {
2190         int rc = 0;
2191 
2192         down_read(&osdc->map_sem);
2193         mutex_lock(&osdc->request_mutex);
2194         __register_request(osdc, req);
2195         req->r_sent = 0;
2196         req->r_got_reply = 0;
2197         rc = __map_request(osdc, req, 0);
2198         if (rc < 0) {
2199                 if (nofail) {
2200                         dout("osdc_start_request failed map, "
2201                                 " will retry %lld\n", req->r_tid);
2202                         rc = 0;
2203                 } else {
2204                         __unregister_request(osdc, req);
2205                 }
2206                 goto out_unlock;
2207         }
2208         if (req->r_osd == NULL) {
2209                 dout("send_request %p no up osds in pg\n", req);
2210                 ceph_monc_request_next_osdmap(&osdc->client->monc);
2211         } else {
2212                 __send_queued(osdc);
2213         }
2214         rc = 0;
2215 out_unlock:
2216         mutex_unlock(&osdc->request_mutex);
2217         up_read(&osdc->map_sem);
2218         return rc;
2219 }
2220 EXPORT_SYMBOL(ceph_osdc_start_request);
2221 
2222 /*
2223  * wait for a request to complete
2224  */
2225 int ceph_osdc_wait_request(struct ceph_osd_client *osdc,
2226                            struct ceph_osd_request *req)
2227 {
2228         int rc;
2229 
2230         rc = wait_for_completion_interruptible(&req->r_completion);
2231         if (rc < 0) {
2232                 mutex_lock(&osdc->request_mutex);
2233                 __cancel_request(req);
2234                 __unregister_request(osdc, req);
2235                 mutex_unlock(&osdc->request_mutex);
2236                 complete_request(req);
2237                 dout("wait_request tid %llu canceled/timed out\n", req->r_tid);
2238                 return rc;
2239         }
2240 
2241         dout("wait_request tid %llu result %d\n", req->r_tid, req->r_result);
2242         return req->r_result;
2243 }
2244 EXPORT_SYMBOL(ceph_osdc_wait_request);
2245 
2246 /*
2247  * sync - wait for all in-flight requests to flush.  avoid starvation.
2248  */
2249 void ceph_osdc_sync(struct ceph_osd_client *osdc)
2250 {
2251         struct ceph_osd_request *req;
2252         u64 last_tid, next_tid = 0;
2253 
2254         mutex_lock(&osdc->request_mutex);
2255         last_tid = osdc->last_tid;
2256         while (1) {
2257                 req = __lookup_request_ge(osdc, next_tid);
2258                 if (!req)
2259                         break;
2260                 if (req->r_tid > last_tid)
2261                         break;
2262 
2263                 next_tid = req->r_tid + 1;
2264                 if ((req->r_flags & CEPH_OSD_FLAG_WRITE) == 0)
2265                         continue;
2266 
2267                 ceph_osdc_get_request(req);
2268                 mutex_unlock(&osdc->request_mutex);
2269                 dout("sync waiting on tid %llu (last is %llu)\n",
2270                      req->r_tid, last_tid);
2271                 wait_for_completion(&req->r_safe_completion);
2272                 mutex_lock(&osdc->request_mutex);
2273                 ceph_osdc_put_request(req);
2274         }
2275         mutex_unlock(&osdc->request_mutex);
2276         dout("sync done (thru tid %llu)\n", last_tid);
2277 }
2278 EXPORT_SYMBOL(ceph_osdc_sync);
2279 
2280 /*
2281  * Call all pending notify callbacks - for use after a watch is
2282  * unregistered, to make sure no more callbacks for it will be invoked
2283  */
2284 extern void ceph_osdc_flush_notifies(struct ceph_osd_client *osdc)
2285 {
2286         flush_workqueue(osdc->notify_wq);
2287 }
2288 EXPORT_SYMBOL(ceph_osdc_flush_notifies);
2289 
2290 
2291 /*
2292  * init, shutdown
2293  */
2294 int ceph_osdc_init(struct ceph_osd_client *osdc, struct ceph_client *client)
2295 {
2296         int err;
2297 
2298         dout("init\n");
2299         osdc->client = client;
2300         osdc->osdmap = NULL;
2301         init_rwsem(&osdc->map_sem);
2302         init_completion(&osdc->map_waiters);
2303         osdc->last_requested_map = 0;
2304         mutex_init(&osdc->request_mutex);
2305         osdc->last_tid = 0;
2306         osdc->osds = RB_ROOT;
2307         INIT_LIST_HEAD(&osdc->osd_lru);
2308         osdc->requests = RB_ROOT;
2309         INIT_LIST_HEAD(&osdc->req_lru);
2310         INIT_LIST_HEAD(&osdc->req_unsent);
2311         INIT_LIST_HEAD(&osdc->req_notarget);
2312         INIT_LIST_HEAD(&osdc->req_linger);
2313         osdc->num_requests = 0;
2314         INIT_DELAYED_WORK(&osdc->timeout_work, handle_timeout);
2315         INIT_DELAYED_WORK(&osdc->osds_timeout_work, handle_osds_timeout);
2316         spin_lock_init(&osdc->event_lock);
2317         osdc->event_tree = RB_ROOT;
2318         osdc->event_count = 0;
2319 
2320         schedule_delayed_work(&osdc->osds_timeout_work,
2321            round_jiffies_relative(osdc->client->options->osd_idle_ttl * HZ));
2322 
2323         err = -ENOMEM;
2324         osdc->req_mempool = mempool_create_kmalloc_pool(10,
2325                                         sizeof(struct ceph_osd_request));
2326         if (!osdc->req_mempool)
2327                 goto out;
2328 
2329         err = ceph_msgpool_init(&osdc->msgpool_op, CEPH_MSG_OSD_OP,
2330                                 OSD_OP_FRONT_LEN, 10, true,
2331                                 "osd_op");
2332         if (err < 0)
2333                 goto out_mempool;
2334         err = ceph_msgpool_init(&osdc->msgpool_op_reply, CEPH_MSG_OSD_OPREPLY,
2335                                 OSD_OPREPLY_FRONT_LEN, 10, true,
2336                                 "osd_op_reply");
2337         if (err < 0)
2338                 goto out_msgpool;
2339 
2340         err = -ENOMEM;
2341         osdc->notify_wq = create_singlethread_workqueue("ceph-watch-notify");
2342         if (!osdc->notify_wq)
2343                 goto out_msgpool;
2344         return 0;
2345 
2346 out_msgpool:
2347         ceph_msgpool_destroy(&osdc->msgpool_op);
2348 out_mempool:
2349         mempool_destroy(osdc->req_mempool);
2350 out:
2351         return err;
2352 }
2353 
2354 void ceph_osdc_stop(struct ceph_osd_client *osdc)
2355 {
2356         flush_workqueue(osdc->notify_wq);
2357         destroy_workqueue(osdc->notify_wq);
2358         cancel_delayed_work_sync(&osdc->timeout_work);
2359         cancel_delayed_work_sync(&osdc->osds_timeout_work);
2360         if (osdc->osdmap) {
2361                 ceph_osdmap_destroy(osdc->osdmap);
2362                 osdc->osdmap = NULL;
2363         }
2364         remove_all_osds(osdc);
2365         mempool_destroy(osdc->req_mempool);
2366         ceph_msgpool_destroy(&osdc->msgpool_op);
2367         ceph_msgpool_destroy(&osdc->msgpool_op_reply);
2368 }
2369 
2370 /*
2371  * Read some contiguous pages.  If we cross a stripe boundary, shorten
2372  * *plen.  Return number of bytes read, or error.
2373  */
2374 int ceph_osdc_readpages(struct ceph_osd_client *osdc,
2375                         struct ceph_vino vino, struct ceph_file_layout *layout,
2376                         u64 off, u64 *plen,
2377                         u32 truncate_seq, u64 truncate_size,
2378                         struct page **pages, int num_pages, int page_align)
2379 {
2380         struct ceph_osd_request *req;
2381         int rc = 0;
2382 
2383         dout("readpages on ino %llx.%llx on %llu~%llu\n", vino.ino,
2384              vino.snap, off, *plen);
2385         req = ceph_osdc_new_request(osdc, layout, vino, off, plen, 1,
2386                                     CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
2387                                     NULL, truncate_seq, truncate_size,
2388                                     false);
2389         if (IS_ERR(req))
2390                 return PTR_ERR(req);
2391 
2392         /* it may be a short read due to an object boundary */
2393 
2394         osd_req_op_extent_osd_data_pages(req, 0,
2395                                 pages, *plen, page_align, false, false);
2396 
2397         dout("readpages  final extent is %llu~%llu (%llu bytes align %d)\n",
2398              off, *plen, *plen, page_align);
2399 
2400         ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
2401 
2402         rc = ceph_osdc_start_request(osdc, req, false);
2403         if (!rc)
2404                 rc = ceph_osdc_wait_request(osdc, req);
2405 
2406         ceph_osdc_put_request(req);
2407         dout("readpages result %d\n", rc);
2408         return rc;
2409 }
2410 EXPORT_SYMBOL(ceph_osdc_readpages);
2411 
2412 /*
2413  * do a synchronous write on N pages
2414  */
2415 int ceph_osdc_writepages(struct ceph_osd_client *osdc, struct ceph_vino vino,
2416                          struct ceph_file_layout *layout,
2417                          struct ceph_snap_context *snapc,
2418                          u64 off, u64 len,
2419                          u32 truncate_seq, u64 truncate_size,
2420                          struct timespec *mtime,
2421                          struct page **pages, int num_pages)
2422 {
2423         struct ceph_osd_request *req;
2424         int rc = 0;
2425         int page_align = off & ~PAGE_MASK;
2426 
2427         BUG_ON(vino.snap != CEPH_NOSNAP);       /* snapshots aren't writeable */
2428         req = ceph_osdc_new_request(osdc, layout, vino, off, &len, 1,
2429                                     CEPH_OSD_OP_WRITE,
2430                                     CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
2431                                     snapc, truncate_seq, truncate_size,
2432                                     true);
2433         if (IS_ERR(req))
2434                 return PTR_ERR(req);
2435 
2436         /* it may be a short write due to an object boundary */
2437         osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_align,
2438                                 false, false);
2439         dout("writepages %llu~%llu (%llu bytes)\n", off, len, len);
2440 
2441         ceph_osdc_build_request(req, off, snapc, CEPH_NOSNAP, mtime);
2442 
2443         rc = ceph_osdc_start_request(osdc, req, true);
2444         if (!rc)
2445                 rc = ceph_osdc_wait_request(osdc, req);
2446 
2447         ceph_osdc_put_request(req);
2448         if (rc == 0)
2449                 rc = len;
2450         dout("writepages result %d\n", rc);
2451         return rc;
2452 }
2453 EXPORT_SYMBOL(ceph_osdc_writepages);
2454 
2455 int ceph_osdc_setup(void)
2456 {
2457         BUG_ON(ceph_osd_request_cache);
2458         ceph_osd_request_cache = kmem_cache_create("ceph_osd_request",
2459                                         sizeof (struct ceph_osd_request),
2460                                         __alignof__(struct ceph_osd_request),
2461                                         0, NULL);
2462 
2463         return ceph_osd_request_cache ? 0 : -ENOMEM;
2464 }
2465 EXPORT_SYMBOL(ceph_osdc_setup);
2466 
2467 void ceph_osdc_cleanup(void)
2468 {
2469         BUG_ON(!ceph_osd_request_cache);
2470         kmem_cache_destroy(ceph_osd_request_cache);
2471         ceph_osd_request_cache = NULL;
2472 }
2473 EXPORT_SYMBOL(ceph_osdc_cleanup);
2474 
2475 /*
2476  * handle incoming message
2477  */
2478 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2479 {
2480         struct ceph_osd *osd = con->private;
2481         struct ceph_osd_client *osdc;
2482         int type = le16_to_cpu(msg->hdr.type);
2483 
2484         if (!osd)
2485                 goto out;
2486         osdc = osd->o_osdc;
2487 
2488         switch (type) {
2489         case CEPH_MSG_OSD_MAP:
2490                 ceph_osdc_handle_map(osdc, msg);
2491                 break;
2492         case CEPH_MSG_OSD_OPREPLY:
2493                 handle_reply(osdc, msg, con);
2494                 break;
2495         case CEPH_MSG_WATCH_NOTIFY:
2496                 handle_watch_notify(osdc, msg);
2497                 break;
2498 
2499         default:
2500                 pr_err("received unknown message type %d %s\n", type,
2501                        ceph_msg_type_name(type));
2502         }
2503 out:
2504         ceph_msg_put(msg);
2505 }
2506 
2507 /*
2508  * lookup and return message for incoming reply.  set up reply message
2509  * pages.
2510  */
2511 static struct ceph_msg *get_reply(struct ceph_connection *con,
2512                                   struct ceph_msg_header *hdr,
2513                                   int *skip)
2514 {
2515         struct ceph_osd *osd = con->private;
2516         struct ceph_osd_client *osdc = osd->o_osdc;
2517         struct ceph_msg *m;
2518         struct ceph_osd_request *req;
2519         int front_len = le32_to_cpu(hdr->front_len);
2520         int data_len = le32_to_cpu(hdr->data_len);
2521         u64 tid;
2522 
2523         tid = le64_to_cpu(hdr->tid);
2524         mutex_lock(&osdc->request_mutex);
2525         req = __lookup_request(osdc, tid);
2526         if (!req) {
2527                 *skip = 1;
2528                 m = NULL;
2529                 dout("get_reply unknown tid %llu from osd%d\n", tid,
2530                      osd->o_osd);
2531                 goto out;
2532         }
2533 
2534         if (req->r_reply->con)
2535                 dout("%s revoking msg %p from old con %p\n", __func__,
2536                      req->r_reply, req->r_reply->con);
2537         ceph_msg_revoke_incoming(req->r_reply);
2538 
2539         if (front_len > req->r_reply->front_alloc_len) {
2540                 pr_warning("get_reply front %d > preallocated %d (%u#%llu)\n",
2541                            front_len, req->r_reply->front_alloc_len,
2542                            (unsigned int)con->peer_name.type,
2543                            le64_to_cpu(con->peer_name.num));
2544                 m = ceph_msg_new(CEPH_MSG_OSD_OPREPLY, front_len, GFP_NOFS,
2545                                  false);
2546                 if (!m)
2547                         goto out;
2548                 ceph_msg_put(req->r_reply);
2549                 req->r_reply = m;
2550         }
2551         m = ceph_msg_get(req->r_reply);
2552 
2553         if (data_len > 0) {
2554                 struct ceph_osd_data *osd_data;
2555 
2556                 /*
2557                  * XXX This is assuming there is only one op containing
2558                  * XXX page data.  Probably OK for reads, but this
2559                  * XXX ought to be done more generally.
2560                  */
2561                 osd_data = osd_req_op_extent_osd_data(req, 0);
2562                 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
2563                         if (osd_data->pages &&
2564                                 unlikely(osd_data->length < data_len)) {
2565 
2566                                 pr_warning("tid %lld reply has %d bytes "
2567                                         "we had only %llu bytes ready\n",
2568                                         tid, data_len, osd_data->length);
2569                                 *skip = 1;
2570                                 ceph_msg_put(m);
2571                                 m = NULL;
2572                                 goto out;
2573                         }
2574                 }
2575         }
2576         *skip = 0;
2577         dout("get_reply tid %lld %p\n", tid, m);
2578 
2579 out:
2580         mutex_unlock(&osdc->request_mutex);
2581         return m;
2582 
2583 }
2584 
2585 static struct ceph_msg *alloc_msg(struct ceph_connection *con,
2586                                   struct ceph_msg_header *hdr,
2587                                   int *skip)
2588 {
2589         struct ceph_osd *osd = con->private;
2590         int type = le16_to_cpu(hdr->type);
2591         int front = le32_to_cpu(hdr->front_len);
2592 
2593         *skip = 0;
2594         switch (type) {
2595         case CEPH_MSG_OSD_MAP:
2596         case CEPH_MSG_WATCH_NOTIFY:
2597                 return ceph_msg_new(type, front, GFP_NOFS, false);
2598         case CEPH_MSG_OSD_OPREPLY:
2599                 return get_reply(con, hdr, skip);
2600         default:
2601                 pr_info("alloc_msg unexpected msg type %d from osd%d\n", type,
2602                         osd->o_osd);
2603                 *skip = 1;
2604                 return NULL;
2605         }
2606 }
2607 
2608 /*
2609  * Wrappers to refcount containing ceph_osd struct
2610  */
2611 static struct ceph_connection *get_osd_con(struct ceph_connection *con)
2612 {
2613         struct ceph_osd *osd = con->private;
2614         if (get_osd(osd))
2615                 return con;
2616         return NULL;
2617 }
2618 
2619 static void put_osd_con(struct ceph_connection *con)
2620 {
2621         struct ceph_osd *osd = con->private;
2622         put_osd(osd);
2623 }
2624 
2625 /*
2626  * authentication
2627  */
2628 /*
2629  * Note: returned pointer is the address of a structure that's
2630  * managed separately.  Caller must *not* attempt to free it.
2631  */
2632 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
2633                                         int *proto, int force_new)
2634 {
2635         struct ceph_osd *o = con->private;
2636         struct ceph_osd_client *osdc = o->o_osdc;
2637         struct ceph_auth_client *ac = osdc->client->monc.auth;
2638         struct ceph_auth_handshake *auth = &o->o_auth;
2639 
2640         if (force_new && auth->authorizer) {
2641                 ceph_auth_destroy_authorizer(ac, auth->authorizer);
2642                 auth->authorizer = NULL;
2643         }
2644         if (!auth->authorizer) {
2645                 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2646                                                       auth);
2647                 if (ret)
2648                         return ERR_PTR(ret);
2649         } else {
2650                 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_OSD,
2651                                                      auth);
2652                 if (ret)
2653                         return ERR_PTR(ret);
2654         }
2655         *proto = ac->protocol;
2656 
2657         return auth;
2658 }
2659 
2660 
2661 static int verify_authorizer_reply(struct ceph_connection *con, int len)
2662 {
2663         struct ceph_osd *o = con->private;
2664         struct ceph_osd_client *osdc = o->o_osdc;
2665         struct ceph_auth_client *ac = osdc->client->monc.auth;
2666 
2667         return ceph_auth_verify_authorizer_reply(ac, o->o_auth.authorizer, len);
2668 }
2669 
2670 static int invalidate_authorizer(struct ceph_connection *con)
2671 {
2672         struct ceph_osd *o = con->private;
2673         struct ceph_osd_client *osdc = o->o_osdc;
2674         struct ceph_auth_client *ac = osdc->client->monc.auth;
2675 
2676         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_OSD);
2677         return ceph_monc_validate_auth(&osdc->client->monc);
2678 }
2679 
2680 static const struct ceph_connection_operations osd_con_ops = {
2681         .get = get_osd_con,
2682         .put = put_osd_con,
2683         .dispatch = dispatch,
2684         .get_authorizer = get_authorizer,
2685         .verify_authorizer_reply = verify_authorizer_reply,
2686         .invalidate_authorizer = invalidate_authorizer,
2687         .alloc_msg = alloc_msg,
2688         .fault = osd_reset,
2689 };
2690 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp