diff options
Diffstat (limited to 'net/rds/message.c')
-rw-r--r-- | net/rds/message.c | 118 |
1 files changed, 62 insertions, 56 deletions
diff --git a/net/rds/message.c b/net/rds/message.c index 9a1d67e001ba..84f937f11d47 100644 --- a/net/rds/message.c +++ b/net/rds/message.c | |||
@@ -34,9 +34,6 @@ | |||
34 | #include <linux/slab.h> | 34 | #include <linux/slab.h> |
35 | 35 | ||
36 | #include "rds.h" | 36 | #include "rds.h" |
37 | #include "rdma.h" | ||
38 | |||
39 | static DECLARE_WAIT_QUEUE_HEAD(rds_message_flush_waitq); | ||
40 | 37 | ||
41 | static unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = { | 38 | static unsigned int rds_exthdr_size[__RDS_EXTHDR_MAX] = { |
42 | [RDS_EXTHDR_NONE] = 0, | 39 | [RDS_EXTHDR_NONE] = 0, |
@@ -63,29 +60,31 @@ static void rds_message_purge(struct rds_message *rm) | |||
63 | if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags))) | 60 | if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags))) |
64 | return; | 61 | return; |
65 | 62 | ||
66 | for (i = 0; i < rm->m_nents; i++) { | 63 | for (i = 0; i < rm->data.op_nents; i++) { |
67 | rdsdebug("putting data page %p\n", (void *)sg_page(&rm->m_sg[i])); | 64 | rdsdebug("putting data page %p\n", (void *)sg_page(&rm->data.op_sg[i])); |
68 | /* XXX will have to put_page for page refs */ | 65 | /* XXX will have to put_page for page refs */ |
69 | __free_page(sg_page(&rm->m_sg[i])); | 66 | __free_page(sg_page(&rm->data.op_sg[i])); |
70 | } | 67 | } |
71 | rm->m_nents = 0; | 68 | rm->data.op_nents = 0; |
72 | 69 | ||
73 | if (rm->m_rdma_op) | 70 | if (rm->rdma.op_active) |
74 | rds_rdma_free_op(rm->m_rdma_op); | 71 | rds_rdma_free_op(&rm->rdma); |
75 | if (rm->m_rdma_mr) | 72 | if (rm->rdma.op_rdma_mr) |
76 | rds_mr_put(rm->m_rdma_mr); | 73 | rds_mr_put(rm->rdma.op_rdma_mr); |
77 | } | ||
78 | 74 | ||
79 | void rds_message_inc_purge(struct rds_incoming *inc) | 75 | if (rm->atomic.op_active) |
80 | { | 76 | rds_atomic_free_op(&rm->atomic); |
81 | struct rds_message *rm = container_of(inc, struct rds_message, m_inc); | 77 | if (rm->atomic.op_rdma_mr) |
82 | rds_message_purge(rm); | 78 | rds_mr_put(rm->atomic.op_rdma_mr); |
83 | } | 79 | } |
84 | 80 | ||
85 | void rds_message_put(struct rds_message *rm) | 81 | void rds_message_put(struct rds_message *rm) |
86 | { | 82 | { |
87 | rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount)); | 83 | rdsdebug("put rm %p ref %d\n", rm, atomic_read(&rm->m_refcount)); |
88 | 84 | if (atomic_read(&rm->m_refcount) == 0) { | |
85 | printk(KERN_CRIT "danger refcount zero on %p\n", rm); | ||
86 | WARN_ON(1); | ||
87 | } | ||
89 | if (atomic_dec_and_test(&rm->m_refcount)) { | 88 | if (atomic_dec_and_test(&rm->m_refcount)) { |
90 | BUG_ON(!list_empty(&rm->m_sock_item)); | 89 | BUG_ON(!list_empty(&rm->m_sock_item)); |
91 | BUG_ON(!list_empty(&rm->m_conn_item)); | 90 | BUG_ON(!list_empty(&rm->m_conn_item)); |
@@ -96,12 +95,6 @@ void rds_message_put(struct rds_message *rm) | |||
96 | } | 95 | } |
97 | EXPORT_SYMBOL_GPL(rds_message_put); | 96 | EXPORT_SYMBOL_GPL(rds_message_put); |
98 | 97 | ||
99 | void rds_message_inc_free(struct rds_incoming *inc) | ||
100 | { | ||
101 | struct rds_message *rm = container_of(inc, struct rds_message, m_inc); | ||
102 | rds_message_put(rm); | ||
103 | } | ||
104 | |||
105 | void rds_message_populate_header(struct rds_header *hdr, __be16 sport, | 98 | void rds_message_populate_header(struct rds_header *hdr, __be16 sport, |
106 | __be16 dport, u64 seq) | 99 | __be16 dport, u64 seq) |
107 | { | 100 | { |
@@ -214,41 +207,68 @@ int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 o | |||
214 | } | 207 | } |
215 | EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension); | 208 | EXPORT_SYMBOL_GPL(rds_message_add_rdma_dest_extension); |
216 | 209 | ||
217 | struct rds_message *rds_message_alloc(unsigned int nents, gfp_t gfp) | 210 | /* |
211 | * Each rds_message is allocated with extra space for the scatterlist entries | ||
212 | * rds ops will need. This is to minimize memory allocation count. Then, each rds op | ||
213 | * can grab SGs when initializing its part of the rds_message. | ||
214 | */ | ||
215 | struct rds_message *rds_message_alloc(unsigned int extra_len, gfp_t gfp) | ||
218 | { | 216 | { |
219 | struct rds_message *rm; | 217 | struct rds_message *rm; |
220 | 218 | ||
221 | rm = kzalloc(sizeof(struct rds_message) + | 219 | rm = kzalloc(sizeof(struct rds_message) + extra_len, gfp); |
222 | (nents * sizeof(struct scatterlist)), gfp); | ||
223 | if (!rm) | 220 | if (!rm) |
224 | goto out; | 221 | goto out; |
225 | 222 | ||
226 | if (nents) | 223 | rm->m_used_sgs = 0; |
227 | sg_init_table(rm->m_sg, nents); | 224 | rm->m_total_sgs = extra_len / sizeof(struct scatterlist); |
225 | |||
228 | atomic_set(&rm->m_refcount, 1); | 226 | atomic_set(&rm->m_refcount, 1); |
229 | INIT_LIST_HEAD(&rm->m_sock_item); | 227 | INIT_LIST_HEAD(&rm->m_sock_item); |
230 | INIT_LIST_HEAD(&rm->m_conn_item); | 228 | INIT_LIST_HEAD(&rm->m_conn_item); |
231 | spin_lock_init(&rm->m_rs_lock); | 229 | spin_lock_init(&rm->m_rs_lock); |
230 | init_waitqueue_head(&rm->m_flush_wait); | ||
232 | 231 | ||
233 | out: | 232 | out: |
234 | return rm; | 233 | return rm; |
235 | } | 234 | } |
236 | 235 | ||
236 | /* | ||
237 | * RDS ops use this to grab SG entries from the rm's sg pool. | ||
238 | */ | ||
239 | struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents) | ||
240 | { | ||
241 | struct scatterlist *sg_first = (struct scatterlist *) &rm[1]; | ||
242 | struct scatterlist *sg_ret; | ||
243 | |||
244 | WARN_ON(rm->m_used_sgs + nents > rm->m_total_sgs); | ||
245 | WARN_ON(!nents); | ||
246 | |||
247 | sg_ret = &sg_first[rm->m_used_sgs]; | ||
248 | sg_init_table(sg_ret, nents); | ||
249 | rm->m_used_sgs += nents; | ||
250 | |||
251 | return sg_ret; | ||
252 | } | ||
253 | |||
237 | struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len) | 254 | struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len) |
238 | { | 255 | { |
239 | struct rds_message *rm; | 256 | struct rds_message *rm; |
240 | unsigned int i; | 257 | unsigned int i; |
258 | int num_sgs = ceil(total_len, PAGE_SIZE); | ||
259 | int extra_bytes = num_sgs * sizeof(struct scatterlist); | ||
241 | 260 | ||
242 | rm = rds_message_alloc(ceil(total_len, PAGE_SIZE), GFP_KERNEL); | 261 | rm = rds_message_alloc(extra_bytes, GFP_NOWAIT); |
243 | if (rm == NULL) | 262 | if (!rm) |
244 | return ERR_PTR(-ENOMEM); | 263 | return ERR_PTR(-ENOMEM); |
245 | 264 | ||
246 | set_bit(RDS_MSG_PAGEVEC, &rm->m_flags); | 265 | set_bit(RDS_MSG_PAGEVEC, &rm->m_flags); |
247 | rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); | 266 | rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); |
248 | rm->m_nents = ceil(total_len, PAGE_SIZE); | 267 | rm->data.op_nents = ceil(total_len, PAGE_SIZE); |
268 | rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs); | ||
249 | 269 | ||
250 | for (i = 0; i < rm->m_nents; ++i) { | 270 | for (i = 0; i < rm->data.op_nents; ++i) { |
251 | sg_set_page(&rm->m_sg[i], | 271 | sg_set_page(&rm->data.op_sg[i], |
252 | virt_to_page(page_addrs[i]), | 272 | virt_to_page(page_addrs[i]), |
253 | PAGE_SIZE, 0); | 273 | PAGE_SIZE, 0); |
254 | } | 274 | } |
@@ -256,40 +276,33 @@ struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned in | |||
256 | return rm; | 276 | return rm; |
257 | } | 277 | } |
258 | 278 | ||
259 | struct rds_message *rds_message_copy_from_user(struct iovec *first_iov, | 279 | int rds_message_copy_from_user(struct rds_message *rm, struct iovec *first_iov, |
260 | size_t total_len) | 280 | size_t total_len) |
261 | { | 281 | { |
262 | unsigned long to_copy; | 282 | unsigned long to_copy; |
263 | unsigned long iov_off; | 283 | unsigned long iov_off; |
264 | unsigned long sg_off; | 284 | unsigned long sg_off; |
265 | struct rds_message *rm; | ||
266 | struct iovec *iov; | 285 | struct iovec *iov; |
267 | struct scatterlist *sg; | 286 | struct scatterlist *sg; |
268 | int ret; | 287 | int ret = 0; |
269 | |||
270 | rm = rds_message_alloc(ceil(total_len, PAGE_SIZE), GFP_KERNEL); | ||
271 | if (rm == NULL) { | ||
272 | ret = -ENOMEM; | ||
273 | goto out; | ||
274 | } | ||
275 | 288 | ||
276 | rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); | 289 | rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len); |
277 | 290 | ||
278 | /* | 291 | /* |
279 | * now allocate and copy in the data payload. | 292 | * now allocate and copy in the data payload. |
280 | */ | 293 | */ |
281 | sg = rm->m_sg; | 294 | sg = rm->data.op_sg; |
282 | iov = first_iov; | 295 | iov = first_iov; |
283 | iov_off = 0; | 296 | iov_off = 0; |
284 | sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */ | 297 | sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */ |
285 | 298 | ||
286 | while (total_len) { | 299 | while (total_len) { |
287 | if (sg_page(sg) == NULL) { | 300 | if (!sg_page(sg)) { |
288 | ret = rds_page_remainder_alloc(sg, total_len, | 301 | ret = rds_page_remainder_alloc(sg, total_len, |
289 | GFP_HIGHUSER); | 302 | GFP_HIGHUSER); |
290 | if (ret) | 303 | if (ret) |
291 | goto out; | 304 | goto out; |
292 | rm->m_nents++; | 305 | rm->data.op_nents++; |
293 | sg_off = 0; | 306 | sg_off = 0; |
294 | } | 307 | } |
295 | 308 | ||
@@ -320,14 +333,8 @@ struct rds_message *rds_message_copy_from_user(struct iovec *first_iov, | |||
320 | sg++; | 333 | sg++; |
321 | } | 334 | } |
322 | 335 | ||
323 | ret = 0; | ||
324 | out: | 336 | out: |
325 | if (ret) { | 337 | return ret; |
326 | if (rm) | ||
327 | rds_message_put(rm); | ||
328 | rm = ERR_PTR(ret); | ||
329 | } | ||
330 | return rm; | ||
331 | } | 338 | } |
332 | 339 | ||
333 | int rds_message_inc_copy_to_user(struct rds_incoming *inc, | 340 | int rds_message_inc_copy_to_user(struct rds_incoming *inc, |
@@ -348,7 +355,7 @@ int rds_message_inc_copy_to_user(struct rds_incoming *inc, | |||
348 | 355 | ||
349 | iov = first_iov; | 356 | iov = first_iov; |
350 | iov_off = 0; | 357 | iov_off = 0; |
351 | sg = rm->m_sg; | 358 | sg = rm->data.op_sg; |
352 | vec_off = 0; | 359 | vec_off = 0; |
353 | copied = 0; | 360 | copied = 0; |
354 | 361 | ||
@@ -394,15 +401,14 @@ int rds_message_inc_copy_to_user(struct rds_incoming *inc, | |||
394 | */ | 401 | */ |
395 | void rds_message_wait(struct rds_message *rm) | 402 | void rds_message_wait(struct rds_message *rm) |
396 | { | 403 | { |
397 | wait_event(rds_message_flush_waitq, | 404 | wait_event_interruptible(rm->m_flush_wait, |
398 | !test_bit(RDS_MSG_MAPPED, &rm->m_flags)); | 405 | !test_bit(RDS_MSG_MAPPED, &rm->m_flags)); |
399 | } | 406 | } |
400 | 407 | ||
401 | void rds_message_unmapped(struct rds_message *rm) | 408 | void rds_message_unmapped(struct rds_message *rm) |
402 | { | 409 | { |
403 | clear_bit(RDS_MSG_MAPPED, &rm->m_flags); | 410 | clear_bit(RDS_MSG_MAPPED, &rm->m_flags); |
404 | if (waitqueue_active(&rds_message_flush_waitq)) | 411 | wake_up_interruptible(&rm->m_flush_wait); |
405 | wake_up(&rds_message_flush_waitq); | ||
406 | } | 412 | } |
407 | EXPORT_SYMBOL_GPL(rds_message_unmapped); | 413 | EXPORT_SYMBOL_GPL(rds_message_unmapped); |
408 | 414 | ||