1 /* 2 * blkfront.c 3 * 4 * XenLinux virtual block device driver. 5 * 6 * Copyright (c) 2003-2004, Keir Fraser & Steve Hand 7 * Modifications by Mark A. Williamson are (c) Intel Research Cambridge 8 * Copyright (c) 2004, Christian Limpach 9 * Copyright (c) 2004, Andrew Warfield 10 * Copyright (c) 2005, Christopher Clark 11 * Copyright (c) 2005, XenSource Ltd 12 * 13 * This program is free software; you can redistribute it and/or 14 * modify it under the terms of the GNU General Public License version 2 15 * as published by the Free Software Foundation; or, when distributed 16 * separately from the Linux kernel or incorporated into other 17 * software packages, subject to the following license: 18 * 19 * Permission is hereby granted, free of charge, to any person obtaining a copy 20 * of this source file (the "Software"), to deal in the Software without 21 * restriction, including without limitation the rights to use, copy, modify, 22 * merge, publish, distribute, sublicense, and/or sell copies of the Software, 23 * and to permit persons to whom the Software is furnished to do so, subject to 24 * the following conditions: 25 * 26 * The above copyright notice and this permission notice shall be included in 27 * all copies or substantial portions of the Software. 28 * 29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 30 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 31 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 32 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 33 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 34 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 35 * IN THE SOFTWARE. 36 */ 37 38 #include <linux/interrupt.h> 39 #include <linux/blkdev.h> 40 #include <linux/blk-mq.h> 41 #include <linux/hdreg.h> 42 #include <linux/cdrom.h> 43 #include <linux/module.h> 44 #include <linux/slab.h> 45 #include <linux/major.h> 46 #include <linux/mutex.h> 47 #include <linux/scatterlist.h> 48 #include <linux/bitmap.h> 49 #include <linux/list.h> 50 #include <linux/workqueue.h> 51 #include <linux/sched/mm.h> 52 53 #include <xen/xen.h> 54 #include <xen/xenbus.h> 55 #include <xen/grant_table.h> 56 #include <xen/events.h> 57 #include <xen/page.h> 58 #include <xen/platform_pci.h> 59 60 #include <xen/interface/grant_table.h> 61 #include <xen/interface/io/blkif.h> 62 #include <xen/interface/io/protocols.h> 63 64 #include <asm/xen/hypervisor.h> 65 66 /* 67 * The minimal size of segment supported by the block framework is PAGE_SIZE. 68 * When Linux is using a different page size than Xen, it may not be possible 69 * to put all the data in a single segment. 70 * This can happen when the backend doesn't support indirect descriptor and 71 * therefore the maximum amount of data that a request can carry is 72 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB 73 * 74 * Note that we only support one extra request. So the Linux page size 75 * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) = 76 * 88KB. 77 */ 78 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE) 79 80 enum blkif_state { 81 BLKIF_STATE_DISCONNECTED, 82 BLKIF_STATE_CONNECTED, 83 BLKIF_STATE_SUSPENDED, 84 BLKIF_STATE_ERROR, 85 }; 86 87 struct grant { 88 grant_ref_t gref; 89 struct page *page; 90 struct list_head node; 91 }; 92 93 enum blk_req_status { 94 REQ_PROCESSING, 95 REQ_WAITING, 96 REQ_DONE, 97 REQ_ERROR, 98 REQ_EOPNOTSUPP, 99 }; 100 101 struct blk_shadow { 102 struct blkif_request req; 103 struct request *request; 104 struct grant **grants_used; 105 struct grant **indirect_grants; 106 struct scatterlist *sg; 107 unsigned int num_sg; 108 enum blk_req_status status; 109 110 #define NO_ASSOCIATED_ID ~0UL 111 /* 112 * Id of the sibling if we ever need 2 requests when handling a 113 * block I/O request 114 */ 115 unsigned long associated_id; 116 }; 117 118 struct blkif_req { 119 blk_status_t error; 120 }; 121 122 static inline struct blkif_req *blkif_req(struct request *rq) 123 { 124 return blk_mq_rq_to_pdu(rq); 125 } 126 127 static DEFINE_MUTEX(blkfront_mutex); 128 static const struct block_device_operations xlvbd_block_fops; 129 static struct delayed_work blkfront_work; 130 static LIST_HEAD(info_list); 131 132 /* 133 * Maximum number of segments in indirect requests, the actual value used by 134 * the frontend driver is the minimum of this value and the value provided 135 * by the backend driver. 136 */ 137 138 static unsigned int xen_blkif_max_segments = 32; 139 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444); 140 MODULE_PARM_DESC(max_indirect_segments, 141 "Maximum amount of segments in indirect requests (default is 32)"); 142 143 static unsigned int xen_blkif_max_queues = 4; 144 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444); 145 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk"); 146 147 /* 148 * Maximum order of pages to be used for the shared ring between front and 149 * backend, 4KB page granularity is used. 150 */ 151 static unsigned int xen_blkif_max_ring_order; 152 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444); 153 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring"); 154 155 static bool __read_mostly xen_blkif_trusted = true; 156 module_param_named(trusted, xen_blkif_trusted, bool, 0644); 157 MODULE_PARM_DESC(trusted, "Is the backend trusted"); 158 159 #define BLK_RING_SIZE(info) \ 160 __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages) 161 162 /* 163 * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19 164 * characters are enough. Define to 20 to keep consistent with backend. 165 */ 166 #define RINGREF_NAME_LEN (20) 167 /* 168 * queue-%u would take 7 + 10(UINT_MAX) = 17 characters. 169 */ 170 #define QUEUE_NAME_LEN (17) 171 172 /* 173 * Per-ring info. 174 * Every blkfront device can associate with one or more blkfront_ring_info, 175 * depending on how many hardware queues/rings to be used. 176 */ 177 struct blkfront_ring_info { 178 /* Lock to protect data in every ring buffer. */ 179 spinlock_t ring_lock; 180 struct blkif_front_ring ring; 181 unsigned int ring_ref[XENBUS_MAX_RING_GRANTS]; 182 unsigned int evtchn, irq; 183 struct work_struct work; 184 struct gnttab_free_callback callback; 185 struct list_head indirect_pages; 186 struct list_head grants; 187 unsigned int persistent_gnts_c; 188 unsigned long shadow_free; 189 struct blkfront_info *dev_info; 190 struct blk_shadow shadow[]; 191 }; 192 193 /* 194 * We have one of these per vbd, whether ide, scsi or 'other'. They 195 * hang in private_data off the gendisk structure. We may end up 196 * putting all kinds of interesting stuff here :-) 197 */ 198 struct blkfront_info 199 { 200 struct mutex mutex; 201 struct xenbus_device *xbdev; 202 struct gendisk *gd; 203 u16 sector_size; 204 unsigned int physical_sector_size; 205 unsigned long vdisk_info; 206 int vdevice; 207 blkif_vdev_t handle; 208 enum blkif_state connected; 209 /* Number of pages per ring buffer. */ 210 unsigned int nr_ring_pages; 211 struct request_queue *rq; 212 unsigned int feature_flush:1; 213 unsigned int feature_fua:1; 214 unsigned int feature_discard:1; 215 unsigned int feature_secdiscard:1; 216 /* Connect-time cached feature_persistent parameter */ 217 unsigned int feature_persistent_parm:1; 218 /* Persistent grants feature negotiation result */ 219 unsigned int feature_persistent:1; 220 unsigned int bounce:1; 221 unsigned int discard_granularity; 222 unsigned int discard_alignment; 223 /* Number of 4KB segments handled */ 224 unsigned int max_indirect_segments; 225 int is_ready; 226 struct blk_mq_tag_set tag_set; 227 struct blkfront_ring_info *rinfo; 228 unsigned int nr_rings; 229 unsigned int rinfo_size; 230 /* Save uncomplete reqs and bios for migration. */ 231 struct list_head requests; 232 struct bio_list bio_list; 233 struct list_head info_list; 234 }; 235 236 static unsigned int nr_minors; 237 static unsigned long *minors; 238 static DEFINE_SPINLOCK(minor_lock); 239 240 #define PARTS_PER_DISK 16 241 #define PARTS_PER_EXT_DISK 256 242 243 #define BLKIF_MAJOR(dev) ((dev)>>8) 244 #define BLKIF_MINOR(dev) ((dev) & 0xff) 245 246 #define EXT_SHIFT 28 247 #define EXTENDED (1<<EXT_SHIFT) 248 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED)) 249 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED)) 250 #define EMULATED_HD_DISK_MINOR_OFFSET (0) 251 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256) 252 #define EMULATED_SD_DISK_MINOR_OFFSET (0) 253 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256) 254 255 #define DEV_NAME "xvd" /* name in /dev */ 256 257 /* 258 * Grants are always the same size as a Xen page (i.e 4KB). 259 * A physical segment is always the same size as a Linux page. 260 * Number of grants per physical segment 261 */ 262 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE) 263 264 #define GRANTS_PER_INDIRECT_FRAME \ 265 (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment)) 266 267 #define INDIRECT_GREFS(_grants) \ 268 DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME) 269 270 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo); 271 static void blkfront_gather_backend_features(struct blkfront_info *info); 272 static int negotiate_mq(struct blkfront_info *info); 273 274 #define for_each_rinfo(info, ptr, idx) \ 275 for ((ptr) = (info)->rinfo, (idx) = 0; \ 276 (idx) < (info)->nr_rings; \ 277 (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size) 278 279 static inline struct blkfront_ring_info * 280 get_rinfo(const struct blkfront_info *info, unsigned int i) 281 { 282 BUG_ON(i >= info->nr_rings); 283 return (void *)info->rinfo + i * info->rinfo_size; 284 } 285 286 static int get_id_from_freelist(struct blkfront_ring_info *rinfo) 287 { 288 unsigned long free = rinfo->shadow_free; 289 290 BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info)); 291 rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id; 292 rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */ 293 return free; 294 } 295 296 static int add_id_to_freelist(struct blkfront_ring_info *rinfo, 297 unsigned long id) 298 { 299 if (rinfo->shadow[id].req.u.rw.id != id) 300 return -EINVAL; 301 if (rinfo->shadow[id].request == NULL) 302 return -EINVAL; 303 rinfo->shadow[id].req.u.rw.id = rinfo->shadow_free; 304 rinfo->shadow[id].request = NULL; 305 rinfo->shadow_free = id; 306 return 0; 307 } 308 309 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num) 310 { 311 struct blkfront_info *info = rinfo->dev_info; 312 struct page *granted_page; 313 struct grant *gnt_list_entry, *n; 314 int i = 0; 315 316 while (i < num) { 317 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO); 318 if (!gnt_list_entry) 319 goto out_of_memory; 320 321 if (info->bounce) { 322 granted_page = alloc_page(GFP_NOIO | __GFP_ZERO); 323 if (!granted_page) { 324 kfree(gnt_list_entry); 325 goto out_of_memory; 326 } 327 gnt_list_entry->page = granted_page; 328 } 329 330 gnt_list_entry->gref = INVALID_GRANT_REF; 331 list_add(&gnt_list_entry->node, &rinfo->grants); 332 i++; 333 } 334 335 return 0; 336 337 out_of_memory: 338 list_for_each_entry_safe(gnt_list_entry, n, 339 &rinfo->grants, node) { 340 list_del(&gnt_list_entry->node); 341 if (info->bounce) 342 __free_page(gnt_list_entry->page); 343 kfree(gnt_list_entry); 344 i--; 345 } 346 BUG_ON(i != 0); 347 return -ENOMEM; 348 } 349 350 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo) 351 { 352 struct grant *gnt_list_entry; 353 354 BUG_ON(list_empty(&rinfo->grants)); 355 gnt_list_entry = list_first_entry(&rinfo->grants, struct grant, 356 node); 357 list_del(&gnt_list_entry->node); 358 359 if (gnt_list_entry->gref != INVALID_GRANT_REF) 360 rinfo->persistent_gnts_c--; 361 362 return gnt_list_entry; 363 } 364 365 static inline void grant_foreign_access(const struct grant *gnt_list_entry, 366 const struct blkfront_info *info) 367 { 368 gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref, 369 info->xbdev->otherend_id, 370 gnt_list_entry->page, 371 0); 372 } 373 374 static struct grant *get_grant(grant_ref_t *gref_head, 375 unsigned long gfn, 376 struct blkfront_ring_info *rinfo) 377 { 378 struct grant *gnt_list_entry = get_free_grant(rinfo); 379 struct blkfront_info *info = rinfo->dev_info; 380 381 if (gnt_list_entry->gref != INVALID_GRANT_REF) 382 return gnt_list_entry; 383 384 /* Assign a gref to this page */ 385 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head); 386 BUG_ON(gnt_list_entry->gref == -ENOSPC); 387 if (info->bounce) 388 grant_foreign_access(gnt_list_entry, info); 389 else { 390 /* Grant access to the GFN passed by the caller */ 391 gnttab_grant_foreign_access_ref(gnt_list_entry->gref, 392 info->xbdev->otherend_id, 393 gfn, 0); 394 } 395 396 return gnt_list_entry; 397 } 398 399 static struct grant *get_indirect_grant(grant_ref_t *gref_head, 400 struct blkfront_ring_info *rinfo) 401 { 402 struct grant *gnt_list_entry = get_free_grant(rinfo); 403 struct blkfront_info *info = rinfo->dev_info; 404 405 if (gnt_list_entry->gref != INVALID_GRANT_REF) 406 return gnt_list_entry; 407 408 /* Assign a gref to this page */ 409 gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head); 410 BUG_ON(gnt_list_entry->gref == -ENOSPC); 411 if (!info->bounce) { 412 struct page *indirect_page; 413 414 /* Fetch a pre-allocated page to use for indirect grefs */ 415 BUG_ON(list_empty(&rinfo->indirect_pages)); 416 indirect_page = list_first_entry(&rinfo->indirect_pages, 417 struct page, lru); 418 list_del(&indirect_page->lru); 419 gnt_list_entry->page = indirect_page; 420 } 421 grant_foreign_access(gnt_list_entry, info); 422 423 return gnt_list_entry; 424 } 425 426 static const char *op_name(int op) 427 { 428 static const char *const names[] = { 429 [BLKIF_OP_READ] = "read", 430 [BLKIF_OP_WRITE] = "write", 431 [BLKIF_OP_WRITE_BARRIER] = "barrier", 432 [BLKIF_OP_FLUSH_DISKCACHE] = "flush", 433 [BLKIF_OP_DISCARD] = "discard" }; 434 435 if (op < 0 || op >= ARRAY_SIZE(names)) 436 return "unknown"; 437 438 if (!names[op]) 439 return "reserved"; 440 441 return names[op]; 442 } 443 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr) 444 { 445 unsigned int end = minor + nr; 446 int rc; 447 448 if (end > nr_minors) { 449 unsigned long *bitmap, *old; 450 451 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap), 452 GFP_KERNEL); 453 if (bitmap == NULL) 454 return -ENOMEM; 455 456 spin_lock(&minor_lock); 457 if (end > nr_minors) { 458 old = minors; 459 memcpy(bitmap, minors, 460 BITS_TO_LONGS(nr_minors) * sizeof(*bitmap)); 461 minors = bitmap; 462 nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG; 463 } else 464 old = bitmap; 465 spin_unlock(&minor_lock); 466 kfree(old); 467 } 468 469 spin_lock(&minor_lock); 470 if (find_next_bit(minors, end, minor) >= end) { 471 bitmap_set(minors, minor, nr); 472 rc = 0; 473 } else 474 rc = -EBUSY; 475 spin_unlock(&minor_lock); 476 477 return rc; 478 } 479 480 static void xlbd_release_minors(unsigned int minor, unsigned int nr) 481 { 482 unsigned int end = minor + nr; 483 484 BUG_ON(end > nr_minors); 485 spin_lock(&minor_lock); 486 bitmap_clear(minors, minor, nr); 487 spin_unlock(&minor_lock); 488 } 489 490 static void blkif_restart_queue_callback(void *arg) 491 { 492 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg; 493 schedule_work(&rinfo->work); 494 } 495 496 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg) 497 { 498 /* We don't have real geometry info, but let's at least return 499 values consistent with the size of the device */ 500 sector_t nsect = get_capacity(bd->bd_disk); 501 sector_t cylinders = nsect; 502 503 hg->heads = 0xff; 504 hg->sectors = 0x3f; 505 sector_div(cylinders, hg->heads * hg->sectors); 506 hg->cylinders = cylinders; 507 if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect) 508 hg->cylinders = 0xffff; 509 return 0; 510 } 511 512 static int blkif_ioctl(struct block_device *bdev, blk_mode_t mode, 513 unsigned command, unsigned long argument) 514 { 515 struct blkfront_info *info = bdev->bd_disk->private_data; 516 int i; 517 518 switch (command) { 519 case CDROMMULTISESSION: 520 for (i = 0; i < sizeof(struct cdrom_multisession); i++) 521 if (put_user(0, (char __user *)(argument + i))) 522 return -EFAULT; 523 return 0; 524 case CDROM_GET_CAPABILITY: 525 if (!(info->vdisk_info & VDISK_CDROM)) 526 return -EINVAL; 527 return 0; 528 default: 529 return -EINVAL; 530 } 531 } 532 533 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo, 534 struct request *req, 535 struct blkif_request **ring_req) 536 { 537 unsigned long id; 538 539 *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt); 540 rinfo->ring.req_prod_pvt++; 541 542 id = get_id_from_freelist(rinfo); 543 rinfo->shadow[id].request = req; 544 rinfo->shadow[id].status = REQ_PROCESSING; 545 rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID; 546 547 rinfo->shadow[id].req.u.rw.id = id; 548 549 return id; 550 } 551 552 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo) 553 { 554 struct blkfront_info *info = rinfo->dev_info; 555 struct blkif_request *ring_req, *final_ring_req; 556 unsigned long id; 557 558 /* Fill out a communications ring structure. */ 559 id = blkif_ring_get_request(rinfo, req, &final_ring_req); 560 ring_req = &rinfo->shadow[id].req; 561 562 ring_req->operation = BLKIF_OP_DISCARD; 563 ring_req->u.discard.nr_sectors = blk_rq_sectors(req); 564 ring_req->u.discard.id = id; 565 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req); 566 if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard) 567 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE; 568 else 569 ring_req->u.discard.flag = 0; 570 571 /* Copy the request to the ring page. */ 572 *final_ring_req = *ring_req; 573 rinfo->shadow[id].status = REQ_WAITING; 574 575 return 0; 576 } 577 578 struct setup_rw_req { 579 unsigned int grant_idx; 580 struct blkif_request_segment *segments; 581 struct blkfront_ring_info *rinfo; 582 struct blkif_request *ring_req; 583 grant_ref_t gref_head; 584 unsigned int id; 585 /* Only used when persistent grant is used and it's a write request */ 586 bool need_copy; 587 unsigned int bvec_off; 588 char *bvec_data; 589 590 bool require_extra_req; 591 struct blkif_request *extra_ring_req; 592 }; 593 594 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset, 595 unsigned int len, void *data) 596 { 597 struct setup_rw_req *setup = data; 598 int n, ref; 599 struct grant *gnt_list_entry; 600 unsigned int fsect, lsect; 601 /* Convenient aliases */ 602 unsigned int grant_idx = setup->grant_idx; 603 struct blkif_request *ring_req = setup->ring_req; 604 struct blkfront_ring_info *rinfo = setup->rinfo; 605 /* 606 * We always use the shadow of the first request to store the list 607 * of grant associated to the block I/O request. This made the 608 * completion more easy to handle even if the block I/O request is 609 * split. 610 */ 611 struct blk_shadow *shadow = &rinfo->shadow[setup->id]; 612 613 if (unlikely(setup->require_extra_req && 614 grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) { 615 /* 616 * We are using the second request, setup grant_idx 617 * to be the index of the segment array. 618 */ 619 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST; 620 ring_req = setup->extra_ring_req; 621 } 622 623 if ((ring_req->operation == BLKIF_OP_INDIRECT) && 624 (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) { 625 if (setup->segments) 626 kunmap_atomic(setup->segments); 627 628 n = grant_idx / GRANTS_PER_INDIRECT_FRAME; 629 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo); 630 shadow->indirect_grants[n] = gnt_list_entry; 631 setup->segments = kmap_atomic(gnt_list_entry->page); 632 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref; 633 } 634 635 gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo); 636 ref = gnt_list_entry->gref; 637 /* 638 * All the grants are stored in the shadow of the first 639 * request. Therefore we have to use the global index. 640 */ 641 shadow->grants_used[setup->grant_idx] = gnt_list_entry; 642 643 if (setup->need_copy) { 644 void *shared_data; 645 646 shared_data = kmap_atomic(gnt_list_entry->page); 647 /* 648 * this does not wipe data stored outside the 649 * range sg->offset..sg->offset+sg->length. 650 * Therefore, blkback *could* see data from 651 * previous requests. This is OK as long as 652 * persistent grants are shared with just one 653 * domain. It may need refactoring if this 654 * changes 655 */ 656 memcpy(shared_data + offset, 657 setup->bvec_data + setup->bvec_off, 658 len); 659 660 kunmap_atomic(shared_data); 661 setup->bvec_off += len; 662 } 663 664 fsect = offset >> 9; 665 lsect = fsect + (len >> 9) - 1; 666 if (ring_req->operation != BLKIF_OP_INDIRECT) { 667 ring_req->u.rw.seg[grant_idx] = 668 (struct blkif_request_segment) { 669 .gref = ref, 670 .first_sect = fsect, 671 .last_sect = lsect }; 672 } else { 673 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] = 674 (struct blkif_request_segment) { 675 .gref = ref, 676 .first_sect = fsect, 677 .last_sect = lsect }; 678 } 679 680 (setup->grant_idx)++; 681 } 682 683 static void blkif_setup_extra_req(struct blkif_request *first, 684 struct blkif_request *second) 685 { 686 uint16_t nr_segments = first->u.rw.nr_segments; 687 688 /* 689 * The second request is only present when the first request uses 690 * all its segments. It's always the continuity of the first one. 691 */ 692 first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST; 693 694 second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST; 695 second->u.rw.sector_number = first->u.rw.sector_number + 696 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512; 697 698 second->u.rw.handle = first->u.rw.handle; 699 second->operation = first->operation; 700 } 701 702 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo) 703 { 704 struct blkfront_info *info = rinfo->dev_info; 705 struct blkif_request *ring_req, *extra_ring_req = NULL; 706 struct blkif_request *final_ring_req, *final_extra_ring_req = NULL; 707 unsigned long id, extra_id = NO_ASSOCIATED_ID; 708 bool require_extra_req = false; 709 int i; 710 struct setup_rw_req setup = { 711 .grant_idx = 0, 712 .segments = NULL, 713 .rinfo = rinfo, 714 .need_copy = rq_data_dir(req) && info->bounce, 715 }; 716 717 /* 718 * Used to store if we are able to queue the request by just using 719 * existing persistent grants, or if we have to get new grants, 720 * as there are not sufficiently many free. 721 */ 722 bool new_persistent_gnts = false; 723 struct scatterlist *sg; 724 int num_sg, max_grefs, num_grant; 725 726 max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG; 727 if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST) 728 /* 729 * If we are using indirect segments we need to account 730 * for the indirect grefs used in the request. 731 */ 732 max_grefs += INDIRECT_GREFS(max_grefs); 733 734 /* Check if we have enough persistent grants to allocate a requests */ 735 if (rinfo->persistent_gnts_c < max_grefs) { 736 new_persistent_gnts = true; 737 738 if (gnttab_alloc_grant_references( 739 max_grefs - rinfo->persistent_gnts_c, 740 &setup.gref_head) < 0) { 741 gnttab_request_free_callback( 742 &rinfo->callback, 743 blkif_restart_queue_callback, 744 rinfo, 745 max_grefs - rinfo->persistent_gnts_c); 746 return 1; 747 } 748 } 749 750 /* Fill out a communications ring structure. */ 751 id = blkif_ring_get_request(rinfo, req, &final_ring_req); 752 ring_req = &rinfo->shadow[id].req; 753 754 num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg); 755 num_grant = 0; 756 /* Calculate the number of grant used */ 757 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) 758 num_grant += gnttab_count_grant(sg->offset, sg->length); 759 760 require_extra_req = info->max_indirect_segments == 0 && 761 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST; 762 BUG_ON(!HAS_EXTRA_REQ && require_extra_req); 763 764 rinfo->shadow[id].num_sg = num_sg; 765 if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST && 766 likely(!require_extra_req)) { 767 /* 768 * The indirect operation can only be a BLKIF_OP_READ or 769 * BLKIF_OP_WRITE 770 */ 771 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA); 772 ring_req->operation = BLKIF_OP_INDIRECT; 773 ring_req->u.indirect.indirect_op = rq_data_dir(req) ? 774 BLKIF_OP_WRITE : BLKIF_OP_READ; 775 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req); 776 ring_req->u.indirect.handle = info->handle; 777 ring_req->u.indirect.nr_segments = num_grant; 778 } else { 779 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req); 780 ring_req->u.rw.handle = info->handle; 781 ring_req->operation = rq_data_dir(req) ? 782 BLKIF_OP_WRITE : BLKIF_OP_READ; 783 if (req_op(req) == REQ_OP_FLUSH || 784 (req_op(req) == REQ_OP_WRITE && (req->cmd_flags & REQ_FUA))) { 785 /* 786 * Ideally we can do an unordered flush-to-disk. 787 * In case the backend onlysupports barriers, use that. 788 * A barrier request a superset of FUA, so we can 789 * implement it the same way. (It's also a FLUSH+FUA, 790 * since it is guaranteed ordered WRT previous writes.) 791 * 792 * Note that can end up here with a FUA write and the 793 * flags cleared. This happens when the flag was 794 * run-time disabled after a failing I/O, and we'll 795 * simplify submit it as a normal write. 796 */ 797 if (info->feature_flush && info->feature_fua) 798 ring_req->operation = 799 BLKIF_OP_WRITE_BARRIER; 800 else if (info->feature_flush) 801 ring_req->operation = 802 BLKIF_OP_FLUSH_DISKCACHE; 803 } 804 ring_req->u.rw.nr_segments = num_grant; 805 if (unlikely(require_extra_req)) { 806 extra_id = blkif_ring_get_request(rinfo, req, 807 &final_extra_ring_req); 808 extra_ring_req = &rinfo->shadow[extra_id].req; 809 810 /* 811 * Only the first request contains the scatter-gather 812 * list. 813 */ 814 rinfo->shadow[extra_id].num_sg = 0; 815 816 blkif_setup_extra_req(ring_req, extra_ring_req); 817 818 /* Link the 2 requests together */ 819 rinfo->shadow[extra_id].associated_id = id; 820 rinfo->shadow[id].associated_id = extra_id; 821 } 822 } 823 824 setup.ring_req = ring_req; 825 setup.id = id; 826 827 setup.require_extra_req = require_extra_req; 828 if (unlikely(require_extra_req)) 829 setup.extra_ring_req = extra_ring_req; 830 831 for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) { 832 BUG_ON(sg->offset + sg->length > PAGE_SIZE); 833 834 if (setup.need_copy) { 835 setup.bvec_off = sg->offset; 836 setup.bvec_data = kmap_atomic(sg_page(sg)); 837 } 838 839 gnttab_foreach_grant_in_range(sg_page(sg), 840 sg->offset, 841 sg->length, 842 blkif_setup_rw_req_grant, 843 &setup); 844 845 if (setup.need_copy) 846 kunmap_atomic(setup.bvec_data); 847 } 848 if (setup.segments) 849 kunmap_atomic(setup.segments); 850 851 /* Copy request(s) to the ring page. */ 852 *final_ring_req = *ring_req; 853 rinfo->shadow[id].status = REQ_WAITING; 854 if (unlikely(require_extra_req)) { 855 *final_extra_ring_req = *extra_ring_req; 856 rinfo->shadow[extra_id].status = REQ_WAITING; 857 } 858 859 if (new_persistent_gnts) 860 gnttab_free_grant_references(setup.gref_head); 861 862 return 0; 863 } 864 865 /* 866 * Generate a Xen blkfront IO request from a blk layer request. Reads 867 * and writes are handled as expected. 868 * 869 * @req: a request struct 870 */ 871 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo) 872 { 873 if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED)) 874 return 1; 875 876 if (unlikely(req_op(req) == REQ_OP_DISCARD || 877 req_op(req) == REQ_OP_SECURE_ERASE)) 878 return blkif_queue_discard_req(req, rinfo); 879 else 880 return blkif_queue_rw_req(req, rinfo); 881 } 882 883 static inline void flush_requests(struct blkfront_ring_info *rinfo) 884 { 885 int notify; 886 887 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify); 888 889 if (notify) 890 notify_remote_via_irq(rinfo->irq); 891 } 892 893 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx, 894 const struct blk_mq_queue_data *qd) 895 { 896 unsigned long flags; 897 int qid = hctx->queue_num; 898 struct blkfront_info *info = hctx->queue->queuedata; 899 struct blkfront_ring_info *rinfo = NULL; 900 901 rinfo = get_rinfo(info, qid); 902 blk_mq_start_request(qd->rq); 903 spin_lock_irqsave(&rinfo->ring_lock, flags); 904 905 /* 906 * Check if the backend actually supports flushes. 907 * 908 * While the block layer won't send us flushes if we don't claim to 909 * support them, the Xen protocol allows the backend to revoke support 910 * at any time. That is of course a really bad idea and dangerous, but 911 * has been allowed for 10+ years. In that case we simply clear the 912 * flags, and directly return here for an empty flush and ignore the 913 * FUA flag later on. 914 */ 915 if (unlikely(req_op(qd->rq) == REQ_OP_FLUSH && !info->feature_flush)) 916 goto complete; 917 918 if (RING_FULL(&rinfo->ring)) 919 goto out_busy; 920 if (blkif_queue_request(qd->rq, rinfo)) 921 goto out_busy; 922 923 flush_requests(rinfo); 924 spin_unlock_irqrestore(&rinfo->ring_lock, flags); 925 return BLK_STS_OK; 926 927 out_busy: 928 blk_mq_stop_hw_queue(hctx); 929 spin_unlock_irqrestore(&rinfo->ring_lock, flags); 930 return BLK_STS_DEV_RESOURCE; 931 complete: 932 spin_unlock_irqrestore(&rinfo->ring_lock, flags); 933 blk_mq_end_request(qd->rq, BLK_STS_OK); 934 return BLK_STS_OK; 935 } 936 937 static void blkif_complete_rq(struct request *rq) 938 { 939 blk_mq_end_request(rq, blkif_req(rq)->error); 940 } 941 942 static const struct blk_mq_ops blkfront_mq_ops = { 943 .queue_rq = blkif_queue_rq, 944 .complete = blkif_complete_rq, 945 }; 946 947 static void blkif_set_queue_limits(const struct blkfront_info *info, 948 struct queue_limits *lim) 949 { 950 unsigned int segments = info->max_indirect_segments ? : 951 BLKIF_MAX_SEGMENTS_PER_REQUEST; 952 953 if (info->feature_discard) { 954 lim->max_hw_discard_sectors = UINT_MAX; 955 if (info->discard_granularity) 956 lim->discard_granularity = info->discard_granularity; 957 lim->discard_alignment = info->discard_alignment; 958 if (info->feature_secdiscard) 959 lim->max_secure_erase_sectors = UINT_MAX; 960 } 961 962 if (info->feature_flush) { 963 lim->features |= BLK_FEAT_WRITE_CACHE; 964 if (info->feature_fua) 965 lim->features |= BLK_FEAT_FUA; 966 } 967 968 /* Hard sector size and max sectors impersonate the equiv. hardware. */ 969 lim->logical_block_size = info->sector_size; 970 lim->physical_block_size = info->physical_sector_size; 971 lim->max_hw_sectors = (segments * XEN_PAGE_SIZE) / 512; 972 973 /* Each segment in a request is up to an aligned page in size. */ 974 lim->seg_boundary_mask = PAGE_SIZE - 1; 975 lim->max_segment_size = PAGE_SIZE; 976 977 /* Ensure a merged request will fit in a single I/O ring slot. */ 978 lim->max_segments = segments / GRANTS_PER_PSEG; 979 980 /* Make sure buffer addresses are sector-aligned. */ 981 lim->dma_alignment = 511; 982 } 983 984 static const char *flush_info(struct blkfront_info *info) 985 { 986 if (info->feature_flush && info->feature_fua) 987 return "barrier: enabled;"; 988 else if (info->feature_flush) 989 return "flush diskcache: enabled;"; 990 else 991 return "barrier or flush: disabled;"; 992 } 993 994 static void xlvbd_flush(struct blkfront_info *info) 995 { 996 pr_info("blkfront: %s: %s %s %s %s %s %s %s\n", 997 info->gd->disk_name, flush_info(info), 998 "persistent grants:", info->feature_persistent ? 999 "enabled;" : "disabled;", "indirect descriptors:", 1000 info->max_indirect_segments ? "enabled;" : "disabled;", 1001 "bounce buffer:", info->bounce ? "enabled" : "disabled;"); 1002 } 1003 1004 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset) 1005 { 1006 int major; 1007 major = BLKIF_MAJOR(vdevice); 1008 *minor = BLKIF_MINOR(vdevice); 1009 switch (major) { 1010 case XEN_IDE0_MAJOR: 1011 *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET; 1012 *minor = ((*minor / 64) * PARTS_PER_DISK) + 1013 EMULATED_HD_DISK_MINOR_OFFSET; 1014 break; 1015 case XEN_IDE1_MAJOR: 1016 *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET; 1017 *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) + 1018 EMULATED_HD_DISK_MINOR_OFFSET; 1019 break; 1020 case XEN_SCSI_DISK0_MAJOR: 1021 *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET; 1022 *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET; 1023 break; 1024 case XEN_SCSI_DISK1_MAJOR: 1025 case XEN_SCSI_DISK2_MAJOR: 1026 case XEN_SCSI_DISK3_MAJOR: 1027 case XEN_SCSI_DISK4_MAJOR: 1028 case XEN_SCSI_DISK5_MAJOR: 1029 case XEN_SCSI_DISK6_MAJOR: 1030 case XEN_SCSI_DISK7_MAJOR: 1031 *offset = (*minor / PARTS_PER_DISK) + 1032 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) + 1033 EMULATED_SD_DISK_NAME_OFFSET; 1034 *minor = *minor + 1035 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) + 1036 EMULATED_SD_DISK_MINOR_OFFSET; 1037 break; 1038 case XEN_SCSI_DISK8_MAJOR: 1039 case XEN_SCSI_DISK9_MAJOR: 1040 case XEN_SCSI_DISK10_MAJOR: 1041 case XEN_SCSI_DISK11_MAJOR: 1042 case XEN_SCSI_DISK12_MAJOR: 1043 case XEN_SCSI_DISK13_MAJOR: 1044 case XEN_SCSI_DISK14_MAJOR: 1045 case XEN_SCSI_DISK15_MAJOR: 1046 *offset = (*minor / PARTS_PER_DISK) + 1047 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) + 1048 EMULATED_SD_DISK_NAME_OFFSET; 1049 *minor = *minor + 1050 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) + 1051 EMULATED_SD_DISK_MINOR_OFFSET; 1052 break; 1053 case XENVBD_MAJOR: 1054 *offset = *minor / PARTS_PER_DISK; 1055 break; 1056 default: 1057 printk(KERN_WARNING "blkfront: your disk configuration is " 1058 "incorrect, please use an xvd device instead\n"); 1059 return -ENODEV; 1060 } 1061 return 0; 1062 } 1063 1064 static char *encode_disk_name(char *ptr, unsigned int n) 1065 { 1066 if (n >= 26) 1067 ptr = encode_disk_name(ptr, n / 26 - 1); 1068 *ptr = 'a' + n % 26; 1069 return ptr + 1; 1070 } 1071 1072 static int xlvbd_alloc_gendisk(blkif_sector_t capacity, 1073 struct blkfront_info *info, u16 sector_size, 1074 unsigned int physical_sector_size) 1075 { 1076 struct queue_limits lim = {}; 1077 struct gendisk *gd; 1078 int nr_minors = 1; 1079 int err; 1080 unsigned int offset; 1081 int minor; 1082 int nr_parts; 1083 char *ptr; 1084 1085 BUG_ON(info->gd != NULL); 1086 BUG_ON(info->rq != NULL); 1087 1088 if ((info->vdevice>>EXT_SHIFT) > 1) { 1089 /* this is above the extended range; something is wrong */ 1090 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice); 1091 return -ENODEV; 1092 } 1093 1094 if (!VDEV_IS_EXTENDED(info->vdevice)) { 1095 err = xen_translate_vdev(info->vdevice, &minor, &offset); 1096 if (err) 1097 return err; 1098 nr_parts = PARTS_PER_DISK; 1099 } else { 1100 minor = BLKIF_MINOR_EXT(info->vdevice); 1101 nr_parts = PARTS_PER_EXT_DISK; 1102 offset = minor / nr_parts; 1103 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4) 1104 printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with " 1105 "emulated IDE disks,\n\t choose an xvd device name" 1106 "from xvde on\n", info->vdevice); 1107 } 1108 if (minor >> MINORBITS) { 1109 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n", 1110 info->vdevice, minor); 1111 return -ENODEV; 1112 } 1113 1114 if ((minor % nr_parts) == 0) 1115 nr_minors = nr_parts; 1116 1117 err = xlbd_reserve_minors(minor, nr_minors); 1118 if (err) 1119 return err; 1120 1121 memset(&info->tag_set, 0, sizeof(info->tag_set)); 1122 info->tag_set.ops = &blkfront_mq_ops; 1123 info->tag_set.nr_hw_queues = info->nr_rings; 1124 if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) { 1125 /* 1126 * When indirect descriptior is not supported, the I/O request 1127 * will be split between multiple request in the ring. 1128 * To avoid problems when sending the request, divide by 1129 * 2 the depth of the queue. 1130 */ 1131 info->tag_set.queue_depth = BLK_RING_SIZE(info) / 2; 1132 } else 1133 info->tag_set.queue_depth = BLK_RING_SIZE(info); 1134 info->tag_set.numa_node = NUMA_NO_NODE; 1135 info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; 1136 info->tag_set.cmd_size = sizeof(struct blkif_req); 1137 info->tag_set.driver_data = info; 1138 1139 err = blk_mq_alloc_tag_set(&info->tag_set); 1140 if (err) 1141 goto out_release_minors; 1142 1143 blkif_set_queue_limits(info, &lim); 1144 gd = blk_mq_alloc_disk(&info->tag_set, &lim, info); 1145 if (IS_ERR(gd)) { 1146 err = PTR_ERR(gd); 1147 goto out_free_tag_set; 1148 } 1149 1150 strcpy(gd->disk_name, DEV_NAME); 1151 ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset); 1152 BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN); 1153 if (nr_minors > 1) 1154 *ptr = 0; 1155 else 1156 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr, 1157 "%d", minor & (nr_parts - 1)); 1158 1159 gd->major = XENVBD_MAJOR; 1160 gd->first_minor = minor; 1161 gd->minors = nr_minors; 1162 gd->fops = &xlvbd_block_fops; 1163 gd->private_data = info; 1164 set_capacity(gd, capacity); 1165 1166 info->rq = gd->queue; 1167 info->gd = gd; 1168 info->sector_size = sector_size; 1169 info->physical_sector_size = physical_sector_size; 1170 1171 xlvbd_flush(info); 1172 1173 if (info->vdisk_info & VDISK_READONLY) 1174 set_disk_ro(gd, 1); 1175 if (info->vdisk_info & VDISK_REMOVABLE) 1176 gd->flags |= GENHD_FL_REMOVABLE; 1177 1178 return 0; 1179 1180 out_free_tag_set: 1181 blk_mq_free_tag_set(&info->tag_set); 1182 out_release_minors: 1183 xlbd_release_minors(minor, nr_minors); 1184 return err; 1185 } 1186 1187 /* Already hold rinfo->ring_lock. */ 1188 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo) 1189 { 1190 if (!RING_FULL(&rinfo->ring)) 1191 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true); 1192 } 1193 1194 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo) 1195 { 1196 unsigned long flags; 1197 1198 spin_lock_irqsave(&rinfo->ring_lock, flags); 1199 kick_pending_request_queues_locked(rinfo); 1200 spin_unlock_irqrestore(&rinfo->ring_lock, flags); 1201 } 1202 1203 static void blkif_restart_queue(struct work_struct *work) 1204 { 1205 struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work); 1206 1207 if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED) 1208 kick_pending_request_queues(rinfo); 1209 } 1210 1211 static void blkif_free_ring(struct blkfront_ring_info *rinfo) 1212 { 1213 struct grant *persistent_gnt, *n; 1214 struct blkfront_info *info = rinfo->dev_info; 1215 int i, j, segs; 1216 1217 /* 1218 * Remove indirect pages, this only happens when using indirect 1219 * descriptors but not persistent grants 1220 */ 1221 if (!list_empty(&rinfo->indirect_pages)) { 1222 struct page *indirect_page, *n; 1223 1224 BUG_ON(info->bounce); 1225 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) { 1226 list_del(&indirect_page->lru); 1227 __free_page(indirect_page); 1228 } 1229 } 1230 1231 /* Remove all persistent grants. */ 1232 if (!list_empty(&rinfo->grants)) { 1233 list_for_each_entry_safe(persistent_gnt, n, 1234 &rinfo->grants, node) { 1235 list_del(&persistent_gnt->node); 1236 if (persistent_gnt->gref != INVALID_GRANT_REF) { 1237 gnttab_end_foreign_access(persistent_gnt->gref, 1238 NULL); 1239 rinfo->persistent_gnts_c--; 1240 } 1241 if (info->bounce) 1242 __free_page(persistent_gnt->page); 1243 kfree(persistent_gnt); 1244 } 1245 } 1246 BUG_ON(rinfo->persistent_gnts_c != 0); 1247 1248 for (i = 0; i < BLK_RING_SIZE(info); i++) { 1249 /* 1250 * Clear persistent grants present in requests already 1251 * on the shared ring 1252 */ 1253 if (!rinfo->shadow[i].request) 1254 goto free_shadow; 1255 1256 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ? 1257 rinfo->shadow[i].req.u.indirect.nr_segments : 1258 rinfo->shadow[i].req.u.rw.nr_segments; 1259 for (j = 0; j < segs; j++) { 1260 persistent_gnt = rinfo->shadow[i].grants_used[j]; 1261 gnttab_end_foreign_access(persistent_gnt->gref, NULL); 1262 if (info->bounce) 1263 __free_page(persistent_gnt->page); 1264 kfree(persistent_gnt); 1265 } 1266 1267 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT) 1268 /* 1269 * If this is not an indirect operation don't try to 1270 * free indirect segments 1271 */ 1272 goto free_shadow; 1273 1274 for (j = 0; j < INDIRECT_GREFS(segs); j++) { 1275 persistent_gnt = rinfo->shadow[i].indirect_grants[j]; 1276 gnttab_end_foreign_access(persistent_gnt->gref, NULL); 1277 __free_page(persistent_gnt->page); 1278 kfree(persistent_gnt); 1279 } 1280 1281 free_shadow: 1282 kvfree(rinfo->shadow[i].grants_used); 1283 rinfo->shadow[i].grants_used = NULL; 1284 kvfree(rinfo->shadow[i].indirect_grants); 1285 rinfo->shadow[i].indirect_grants = NULL; 1286 kvfree(rinfo->shadow[i].sg); 1287 rinfo->shadow[i].sg = NULL; 1288 } 1289 1290 /* No more gnttab callback work. */ 1291 gnttab_cancel_free_callback(&rinfo->callback); 1292 1293 /* Flush gnttab callback work. Must be done with no locks held. */ 1294 flush_work(&rinfo->work); 1295 1296 /* Free resources associated with old device channel. */ 1297 xenbus_teardown_ring((void **)&rinfo->ring.sring, info->nr_ring_pages, 1298 rinfo->ring_ref); 1299 1300 if (rinfo->irq) 1301 unbind_from_irqhandler(rinfo->irq, rinfo); 1302 rinfo->evtchn = rinfo->irq = 0; 1303 } 1304 1305 static void blkif_free(struct blkfront_info *info, int suspend) 1306 { 1307 unsigned int i; 1308 struct blkfront_ring_info *rinfo; 1309 1310 /* Prevent new requests being issued until we fix things up. */ 1311 info->connected = suspend ? 1312 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED; 1313 /* No more blkif_request(). */ 1314 if (info->rq) 1315 blk_mq_stop_hw_queues(info->rq); 1316 1317 for_each_rinfo(info, rinfo, i) 1318 blkif_free_ring(rinfo); 1319 1320 kvfree(info->rinfo); 1321 info->rinfo = NULL; 1322 info->nr_rings = 0; 1323 } 1324 1325 struct copy_from_grant { 1326 const struct blk_shadow *s; 1327 unsigned int grant_idx; 1328 unsigned int bvec_offset; 1329 char *bvec_data; 1330 }; 1331 1332 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset, 1333 unsigned int len, void *data) 1334 { 1335 struct copy_from_grant *info = data; 1336 char *shared_data; 1337 /* Convenient aliases */ 1338 const struct blk_shadow *s = info->s; 1339 1340 shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page); 1341 1342 memcpy(info->bvec_data + info->bvec_offset, 1343 shared_data + offset, len); 1344 1345 info->bvec_offset += len; 1346 info->grant_idx++; 1347 1348 kunmap_atomic(shared_data); 1349 } 1350 1351 static enum blk_req_status blkif_rsp_to_req_status(int rsp) 1352 { 1353 switch (rsp) 1354 { 1355 case BLKIF_RSP_OKAY: 1356 return REQ_DONE; 1357 case BLKIF_RSP_EOPNOTSUPP: 1358 return REQ_EOPNOTSUPP; 1359 case BLKIF_RSP_ERROR: 1360 default: 1361 return REQ_ERROR; 1362 } 1363 } 1364 1365 /* 1366 * Get the final status of the block request based on two ring response 1367 */ 1368 static int blkif_get_final_status(enum blk_req_status s1, 1369 enum blk_req_status s2) 1370 { 1371 BUG_ON(s1 < REQ_DONE); 1372 BUG_ON(s2 < REQ_DONE); 1373 1374 if (s1 == REQ_ERROR || s2 == REQ_ERROR) 1375 return BLKIF_RSP_ERROR; 1376 else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP) 1377 return BLKIF_RSP_EOPNOTSUPP; 1378 return BLKIF_RSP_OKAY; 1379 } 1380 1381 /* 1382 * Return values: 1383 * 1 response processed. 1384 * 0 missing further responses. 1385 * -1 error while processing. 1386 */ 1387 static int blkif_completion(unsigned long *id, 1388 struct blkfront_ring_info *rinfo, 1389 struct blkif_response *bret) 1390 { 1391 int i = 0; 1392 struct scatterlist *sg; 1393 int num_sg, num_grant; 1394 struct blkfront_info *info = rinfo->dev_info; 1395 struct blk_shadow *s = &rinfo->shadow[*id]; 1396 struct copy_from_grant data = { 1397 .grant_idx = 0, 1398 }; 1399 1400 num_grant = s->req.operation == BLKIF_OP_INDIRECT ? 1401 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments; 1402 1403 /* The I/O request may be split in two. */ 1404 if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) { 1405 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id]; 1406 1407 /* Keep the status of the current response in shadow. */ 1408 s->status = blkif_rsp_to_req_status(bret->status); 1409 1410 /* Wait the second response if not yet here. */ 1411 if (s2->status < REQ_DONE) 1412 return 0; 1413 1414 bret->status = blkif_get_final_status(s->status, 1415 s2->status); 1416 1417 /* 1418 * All the grants is stored in the first shadow in order 1419 * to make the completion code simpler. 1420 */ 1421 num_grant += s2->req.u.rw.nr_segments; 1422 1423 /* 1424 * The two responses may not come in order. Only the 1425 * first request will store the scatter-gather list. 1426 */ 1427 if (s2->num_sg != 0) { 1428 /* Update "id" with the ID of the first response. */ 1429 *id = s->associated_id; 1430 s = s2; 1431 } 1432 1433 /* 1434 * We don't need anymore the second request, so recycling 1435 * it now. 1436 */ 1437 if (add_id_to_freelist(rinfo, s->associated_id)) 1438 WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n", 1439 info->gd->disk_name, s->associated_id); 1440 } 1441 1442 data.s = s; 1443 num_sg = s->num_sg; 1444 1445 if (bret->operation == BLKIF_OP_READ && info->bounce) { 1446 for_each_sg(s->sg, sg, num_sg, i) { 1447 BUG_ON(sg->offset + sg->length > PAGE_SIZE); 1448 1449 data.bvec_offset = sg->offset; 1450 data.bvec_data = kmap_atomic(sg_page(sg)); 1451 1452 gnttab_foreach_grant_in_range(sg_page(sg), 1453 sg->offset, 1454 sg->length, 1455 blkif_copy_from_grant, 1456 &data); 1457 1458 kunmap_atomic(data.bvec_data); 1459 } 1460 } 1461 /* Add the persistent grant into the list of free grants */ 1462 for (i = 0; i < num_grant; i++) { 1463 if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) { 1464 /* 1465 * If the grant is still mapped by the backend (the 1466 * backend has chosen to make this grant persistent) 1467 * we add it at the head of the list, so it will be 1468 * reused first. 1469 */ 1470 if (!info->feature_persistent) { 1471 pr_alert("backed has not unmapped grant: %u\n", 1472 s->grants_used[i]->gref); 1473 return -1; 1474 } 1475 list_add(&s->grants_used[i]->node, &rinfo->grants); 1476 rinfo->persistent_gnts_c++; 1477 } else { 1478 /* 1479 * If the grant is not mapped by the backend we add it 1480 * to the tail of the list, so it will not be picked 1481 * again unless we run out of persistent grants. 1482 */ 1483 s->grants_used[i]->gref = INVALID_GRANT_REF; 1484 list_add_tail(&s->grants_used[i]->node, &rinfo->grants); 1485 } 1486 } 1487 if (s->req.operation == BLKIF_OP_INDIRECT) { 1488 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) { 1489 if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) { 1490 if (!info->feature_persistent) { 1491 pr_alert("backed has not unmapped grant: %u\n", 1492 s->indirect_grants[i]->gref); 1493 return -1; 1494 } 1495 list_add(&s->indirect_grants[i]->node, &rinfo->grants); 1496 rinfo->persistent_gnts_c++; 1497 } else { 1498 struct page *indirect_page; 1499 1500 /* 1501 * Add the used indirect page back to the list of 1502 * available pages for indirect grefs. 1503 */ 1504 if (!info->bounce) { 1505 indirect_page = s->indirect_grants[i]->page; 1506 list_add(&indirect_page->lru, &rinfo->indirect_pages); 1507 } 1508 s->indirect_grants[i]->gref = INVALID_GRANT_REF; 1509 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants); 1510 } 1511 } 1512 } 1513 1514 return 1; 1515 } 1516 1517 static irqreturn_t blkif_interrupt(int irq, void *dev_id) 1518 { 1519 struct request *req; 1520 struct blkif_response bret; 1521 RING_IDX i, rp; 1522 unsigned long flags; 1523 struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id; 1524 struct blkfront_info *info = rinfo->dev_info; 1525 unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS; 1526 1527 if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) { 1528 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS); 1529 return IRQ_HANDLED; 1530 } 1531 1532 spin_lock_irqsave(&rinfo->ring_lock, flags); 1533 again: 1534 rp = READ_ONCE(rinfo->ring.sring->rsp_prod); 1535 virt_rmb(); /* Ensure we see queued responses up to 'rp'. */ 1536 if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) { 1537 pr_alert("%s: illegal number of responses %u\n", 1538 info->gd->disk_name, rp - rinfo->ring.rsp_cons); 1539 goto err; 1540 } 1541 1542 for (i = rinfo->ring.rsp_cons; i != rp; i++) { 1543 unsigned long id; 1544 unsigned int op; 1545 1546 eoiflag = 0; 1547 1548 RING_COPY_RESPONSE(&rinfo->ring, i, &bret); 1549 id = bret.id; 1550 1551 /* 1552 * The backend has messed up and given us an id that we would 1553 * never have given to it (we stamp it up to BLK_RING_SIZE - 1554 * look in get_id_from_freelist. 1555 */ 1556 if (id >= BLK_RING_SIZE(info)) { 1557 pr_alert("%s: response has incorrect id (%ld)\n", 1558 info->gd->disk_name, id); 1559 goto err; 1560 } 1561 if (rinfo->shadow[id].status != REQ_WAITING) { 1562 pr_alert("%s: response references no pending request\n", 1563 info->gd->disk_name); 1564 goto err; 1565 } 1566 1567 rinfo->shadow[id].status = REQ_PROCESSING; 1568 req = rinfo->shadow[id].request; 1569 1570 op = rinfo->shadow[id].req.operation; 1571 if (op == BLKIF_OP_INDIRECT) 1572 op = rinfo->shadow[id].req.u.indirect.indirect_op; 1573 if (bret.operation != op) { 1574 pr_alert("%s: response has wrong operation (%u instead of %u)\n", 1575 info->gd->disk_name, bret.operation, op); 1576 goto err; 1577 } 1578 1579 if (bret.operation != BLKIF_OP_DISCARD) { 1580 int ret; 1581 1582 /* 1583 * We may need to wait for an extra response if the 1584 * I/O request is split in 2 1585 */ 1586 ret = blkif_completion(&id, rinfo, &bret); 1587 if (!ret) 1588 continue; 1589 if (unlikely(ret < 0)) 1590 goto err; 1591 } 1592 1593 if (add_id_to_freelist(rinfo, id)) { 1594 WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n", 1595 info->gd->disk_name, op_name(bret.operation), id); 1596 continue; 1597 } 1598 1599 if (bret.status == BLKIF_RSP_OKAY) 1600 blkif_req(req)->error = BLK_STS_OK; 1601 else 1602 blkif_req(req)->error = BLK_STS_IOERR; 1603 1604 switch (bret.operation) { 1605 case BLKIF_OP_DISCARD: 1606 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) { 1607 struct request_queue *rq = info->rq; 1608 1609 pr_warn_ratelimited("blkfront: %s: %s op failed\n", 1610 info->gd->disk_name, op_name(bret.operation)); 1611 blkif_req(req)->error = BLK_STS_NOTSUPP; 1612 info->feature_discard = 0; 1613 info->feature_secdiscard = 0; 1614 blk_queue_disable_discard(rq); 1615 blk_queue_disable_secure_erase(rq); 1616 } 1617 break; 1618 case BLKIF_OP_FLUSH_DISKCACHE: 1619 case BLKIF_OP_WRITE_BARRIER: 1620 if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) { 1621 pr_warn_ratelimited("blkfront: %s: %s op failed\n", 1622 info->gd->disk_name, op_name(bret.operation)); 1623 blkif_req(req)->error = BLK_STS_NOTSUPP; 1624 } 1625 if (unlikely(bret.status == BLKIF_RSP_ERROR && 1626 rinfo->shadow[id].req.u.rw.nr_segments == 0)) { 1627 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n", 1628 info->gd->disk_name, op_name(bret.operation)); 1629 blkif_req(req)->error = BLK_STS_NOTSUPP; 1630 } 1631 if (unlikely(blkif_req(req)->error)) { 1632 if (blkif_req(req)->error == BLK_STS_NOTSUPP) 1633 blkif_req(req)->error = BLK_STS_OK; 1634 info->feature_fua = 0; 1635 info->feature_flush = 0; 1636 } 1637 fallthrough; 1638 case BLKIF_OP_READ: 1639 case BLKIF_OP_WRITE: 1640 if (unlikely(bret.status != BLKIF_RSP_OKAY)) 1641 dev_dbg_ratelimited(&info->xbdev->dev, 1642 "Bad return from blkdev data request: %#x\n", 1643 bret.status); 1644 1645 break; 1646 default: 1647 BUG(); 1648 } 1649 1650 if (likely(!blk_should_fake_timeout(req->q))) 1651 blk_mq_complete_request(req); 1652 } 1653 1654 rinfo->ring.rsp_cons = i; 1655 1656 if (i != rinfo->ring.req_prod_pvt) { 1657 int more_to_do; 1658 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do); 1659 if (more_to_do) 1660 goto again; 1661 } else 1662 rinfo->ring.sring->rsp_event = i + 1; 1663 1664 kick_pending_request_queues_locked(rinfo); 1665 1666 spin_unlock_irqrestore(&rinfo->ring_lock, flags); 1667 1668 xen_irq_lateeoi(irq, eoiflag); 1669 1670 return IRQ_HANDLED; 1671 1672 err: 1673 info->connected = BLKIF_STATE_ERROR; 1674 1675 spin_unlock_irqrestore(&rinfo->ring_lock, flags); 1676 1677 /* No EOI in order to avoid further interrupts. */ 1678 1679 pr_alert("%s disabled for further use\n", info->gd->disk_name); 1680 return IRQ_HANDLED; 1681 } 1682 1683 1684 static int setup_blkring(struct xenbus_device *dev, 1685 struct blkfront_ring_info *rinfo) 1686 { 1687 struct blkif_sring *sring; 1688 int err; 1689 struct blkfront_info *info = rinfo->dev_info; 1690 unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE; 1691 1692 err = xenbus_setup_ring(dev, GFP_NOIO, (void **)&sring, 1693 info->nr_ring_pages, rinfo->ring_ref); 1694 if (err) 1695 goto fail; 1696 1697 XEN_FRONT_RING_INIT(&rinfo->ring, sring, ring_size); 1698 1699 err = xenbus_alloc_evtchn(dev, &rinfo->evtchn); 1700 if (err) 1701 goto fail; 1702 1703 err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt, 1704 0, "blkif", rinfo); 1705 if (err <= 0) { 1706 xenbus_dev_fatal(dev, err, 1707 "bind_evtchn_to_irqhandler failed"); 1708 goto fail; 1709 } 1710 rinfo->irq = err; 1711 1712 return 0; 1713 fail: 1714 blkif_free(info, 0); 1715 return err; 1716 } 1717 1718 /* 1719 * Write out per-ring/queue nodes including ring-ref and event-channel, and each 1720 * ring buffer may have multi pages depending on ->nr_ring_pages. 1721 */ 1722 static int write_per_ring_nodes(struct xenbus_transaction xbt, 1723 struct blkfront_ring_info *rinfo, const char *dir) 1724 { 1725 int err; 1726 unsigned int i; 1727 const char *message = NULL; 1728 struct blkfront_info *info = rinfo->dev_info; 1729 1730 if (info->nr_ring_pages == 1) { 1731 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]); 1732 if (err) { 1733 message = "writing ring-ref"; 1734 goto abort_transaction; 1735 } 1736 } else { 1737 for (i = 0; i < info->nr_ring_pages; i++) { 1738 char ring_ref_name[RINGREF_NAME_LEN]; 1739 1740 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i); 1741 err = xenbus_printf(xbt, dir, ring_ref_name, 1742 "%u", rinfo->ring_ref[i]); 1743 if (err) { 1744 message = "writing ring-ref"; 1745 goto abort_transaction; 1746 } 1747 } 1748 } 1749 1750 err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn); 1751 if (err) { 1752 message = "writing event-channel"; 1753 goto abort_transaction; 1754 } 1755 1756 return 0; 1757 1758 abort_transaction: 1759 xenbus_transaction_end(xbt, 1); 1760 if (message) 1761 xenbus_dev_fatal(info->xbdev, err, "%s", message); 1762 1763 return err; 1764 } 1765 1766 /* Enable the persistent grants feature. */ 1767 static bool feature_persistent = true; 1768 module_param(feature_persistent, bool, 0644); 1769 MODULE_PARM_DESC(feature_persistent, 1770 "Enables the persistent grants feature"); 1771 1772 /* Common code used when first setting up, and when resuming. */ 1773 static int talk_to_blkback(struct xenbus_device *dev, 1774 struct blkfront_info *info) 1775 { 1776 const char *message = NULL; 1777 struct xenbus_transaction xbt; 1778 int err; 1779 unsigned int i, max_page_order; 1780 unsigned int ring_page_order; 1781 struct blkfront_ring_info *rinfo; 1782 1783 if (!info) 1784 return -ENODEV; 1785 1786 /* Check if backend is trusted. */ 1787 info->bounce = !xen_blkif_trusted || 1788 !xenbus_read_unsigned(dev->nodename, "trusted", 1); 1789 1790 max_page_order = xenbus_read_unsigned(info->xbdev->otherend, 1791 "max-ring-page-order", 0); 1792 ring_page_order = min(xen_blkif_max_ring_order, max_page_order); 1793 info->nr_ring_pages = 1 << ring_page_order; 1794 1795 err = negotiate_mq(info); 1796 if (err) 1797 goto destroy_blkring; 1798 1799 for_each_rinfo(info, rinfo, i) { 1800 /* Create shared ring, alloc event channel. */ 1801 err = setup_blkring(dev, rinfo); 1802 if (err) 1803 goto destroy_blkring; 1804 } 1805 1806 again: 1807 err = xenbus_transaction_start(&xbt); 1808 if (err) { 1809 xenbus_dev_fatal(dev, err, "starting transaction"); 1810 goto destroy_blkring; 1811 } 1812 1813 if (info->nr_ring_pages > 1) { 1814 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u", 1815 ring_page_order); 1816 if (err) { 1817 message = "writing ring-page-order"; 1818 goto abort_transaction; 1819 } 1820 } 1821 1822 /* We already got the number of queues/rings in _probe */ 1823 if (info->nr_rings == 1) { 1824 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename); 1825 if (err) 1826 goto destroy_blkring; 1827 } else { 1828 char *path; 1829 size_t pathsize; 1830 1831 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u", 1832 info->nr_rings); 1833 if (err) { 1834 message = "writing multi-queue-num-queues"; 1835 goto abort_transaction; 1836 } 1837 1838 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN; 1839 path = kmalloc(pathsize, GFP_KERNEL); 1840 if (!path) { 1841 err = -ENOMEM; 1842 message = "ENOMEM while writing ring references"; 1843 goto abort_transaction; 1844 } 1845 1846 for_each_rinfo(info, rinfo, i) { 1847 memset(path, 0, pathsize); 1848 snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i); 1849 err = write_per_ring_nodes(xbt, rinfo, path); 1850 if (err) { 1851 kfree(path); 1852 goto destroy_blkring; 1853 } 1854 } 1855 kfree(path); 1856 } 1857 err = xenbus_printf(xbt, dev->nodename, "protocol", "%s", 1858 XEN_IO_PROTO_ABI_NATIVE); 1859 if (err) { 1860 message = "writing protocol"; 1861 goto abort_transaction; 1862 } 1863 info->feature_persistent_parm = feature_persistent; 1864 err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1865 info->feature_persistent_parm); 1866 if (err) 1867 dev_warn(&dev->dev, 1868 "writing persistent grants feature to xenbus"); 1869 1870 err = xenbus_transaction_end(xbt, 0); 1871 if (err) { 1872 if (err == -EAGAIN) 1873 goto again; 1874 xenbus_dev_fatal(dev, err, "completing transaction"); 1875 goto destroy_blkring; 1876 } 1877 1878 for_each_rinfo(info, rinfo, i) { 1879 unsigned int j; 1880 1881 for (j = 0; j < BLK_RING_SIZE(info); j++) 1882 rinfo->shadow[j].req.u.rw.id = j + 1; 1883 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff; 1884 } 1885 xenbus_switch_state(dev, XenbusStateInitialised); 1886 1887 return 0; 1888 1889 abort_transaction: 1890 xenbus_transaction_end(xbt, 1); 1891 if (message) 1892 xenbus_dev_fatal(dev, err, "%s", message); 1893 destroy_blkring: 1894 blkif_free(info, 0); 1895 return err; 1896 } 1897 1898 static int negotiate_mq(struct blkfront_info *info) 1899 { 1900 unsigned int backend_max_queues; 1901 unsigned int i; 1902 struct blkfront_ring_info *rinfo; 1903 1904 BUG_ON(info->nr_rings); 1905 1906 /* Check if backend supports multiple queues. */ 1907 backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend, 1908 "multi-queue-max-queues", 1); 1909 info->nr_rings = min(backend_max_queues, xen_blkif_max_queues); 1910 /* We need at least one ring. */ 1911 if (!info->nr_rings) 1912 info->nr_rings = 1; 1913 1914 info->rinfo_size = struct_size(info->rinfo, shadow, 1915 BLK_RING_SIZE(info)); 1916 info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL); 1917 if (!info->rinfo) { 1918 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure"); 1919 info->nr_rings = 0; 1920 return -ENOMEM; 1921 } 1922 1923 for_each_rinfo(info, rinfo, i) { 1924 INIT_LIST_HEAD(&rinfo->indirect_pages); 1925 INIT_LIST_HEAD(&rinfo->grants); 1926 rinfo->dev_info = info; 1927 INIT_WORK(&rinfo->work, blkif_restart_queue); 1928 spin_lock_init(&rinfo->ring_lock); 1929 } 1930 return 0; 1931 } 1932 1933 /* 1934 * Entry point to this code when a new device is created. Allocate the basic 1935 * structures and the ring buffer for communication with the backend, and 1936 * inform the backend of the appropriate details for those. Switch to 1937 * Initialised state. 1938 */ 1939 static int blkfront_probe(struct xenbus_device *dev, 1940 const struct xenbus_device_id *id) 1941 { 1942 int err, vdevice; 1943 struct blkfront_info *info; 1944 1945 /* FIXME: Use dynamic device id if this is not set. */ 1946 err = xenbus_scanf(XBT_NIL, dev->nodename, 1947 "virtual-device", "%i", &vdevice); 1948 if (err != 1) { 1949 /* go looking in the extended area instead */ 1950 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext", 1951 "%i", &vdevice); 1952 if (err != 1) { 1953 xenbus_dev_fatal(dev, err, "reading virtual-device"); 1954 return err; 1955 } 1956 } 1957 1958 if (xen_hvm_domain()) { 1959 char *type; 1960 int len; 1961 /* no unplug has been done: do not hook devices != xen vbds */ 1962 if (xen_has_pv_and_legacy_disk_devices()) { 1963 int major; 1964 1965 if (!VDEV_IS_EXTENDED(vdevice)) 1966 major = BLKIF_MAJOR(vdevice); 1967 else 1968 major = XENVBD_MAJOR; 1969 1970 if (major != XENVBD_MAJOR) { 1971 printk(KERN_INFO 1972 "%s: HVM does not support vbd %d as xen block device\n", 1973 __func__, vdevice); 1974 return -ENODEV; 1975 } 1976 } 1977 /* do not create a PV cdrom device if we are an HVM guest */ 1978 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len); 1979 if (IS_ERR(type)) 1980 return -ENODEV; 1981 if (strncmp(type, "cdrom", 5) == 0) { 1982 kfree(type); 1983 return -ENODEV; 1984 } 1985 kfree(type); 1986 } 1987 info = kzalloc(sizeof(*info), GFP_KERNEL); 1988 if (!info) { 1989 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure"); 1990 return -ENOMEM; 1991 } 1992 1993 info->xbdev = dev; 1994 1995 mutex_init(&info->mutex); 1996 info->vdevice = vdevice; 1997 info->connected = BLKIF_STATE_DISCONNECTED; 1998 1999 /* Front end dir is a number, which is used as the id. */ 2000 info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0); 2001 dev_set_drvdata(&dev->dev, info); 2002 2003 mutex_lock(&blkfront_mutex); 2004 list_add(&info->info_list, &info_list); 2005 mutex_unlock(&blkfront_mutex); 2006 2007 return 0; 2008 } 2009 2010 static int blkif_recover(struct blkfront_info *info) 2011 { 2012 struct queue_limits lim; 2013 unsigned int r_index; 2014 struct request *req, *n; 2015 int rc; 2016 struct bio *bio; 2017 struct blkfront_ring_info *rinfo; 2018 2019 lim = queue_limits_start_update(info->rq); 2020 blkfront_gather_backend_features(info); 2021 blkif_set_queue_limits(info, &lim); 2022 rc = queue_limits_commit_update(info->rq, &lim); 2023 if (rc) 2024 return rc; 2025 2026 for_each_rinfo(info, rinfo, r_index) { 2027 rc = blkfront_setup_indirect(rinfo); 2028 if (rc) 2029 return rc; 2030 } 2031 xenbus_switch_state(info->xbdev, XenbusStateConnected); 2032 2033 /* Now safe for us to use the shared ring */ 2034 info->connected = BLKIF_STATE_CONNECTED; 2035 2036 for_each_rinfo(info, rinfo, r_index) { 2037 /* Kick any other new requests queued since we resumed */ 2038 kick_pending_request_queues(rinfo); 2039 } 2040 2041 list_for_each_entry_safe(req, n, &info->requests, queuelist) { 2042 /* Requeue pending requests (flush or discard) */ 2043 list_del_init(&req->queuelist); 2044 BUG_ON(req->nr_phys_segments > 2045 (info->max_indirect_segments ? : 2046 BLKIF_MAX_SEGMENTS_PER_REQUEST)); 2047 blk_mq_requeue_request(req, false); 2048 } 2049 blk_mq_start_stopped_hw_queues(info->rq, true); 2050 blk_mq_kick_requeue_list(info->rq); 2051 2052 while ((bio = bio_list_pop(&info->bio_list)) != NULL) { 2053 /* Traverse the list of pending bios and re-queue them */ 2054 submit_bio(bio); 2055 } 2056 2057 return 0; 2058 } 2059 2060 /* 2061 * We are reconnecting to the backend, due to a suspend/resume, or a backend 2062 * driver restart. We tear down our blkif structure and recreate it, but 2063 * leave the device-layer structures intact so that this is transparent to the 2064 * rest of the kernel. 2065 */ 2066 static int blkfront_resume(struct xenbus_device *dev) 2067 { 2068 struct blkfront_info *info = dev_get_drvdata(&dev->dev); 2069 int err = 0; 2070 unsigned int i, j; 2071 struct blkfront_ring_info *rinfo; 2072 2073 dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename); 2074 2075 bio_list_init(&info->bio_list); 2076 INIT_LIST_HEAD(&info->requests); 2077 for_each_rinfo(info, rinfo, i) { 2078 struct bio_list merge_bio; 2079 struct blk_shadow *shadow = rinfo->shadow; 2080 2081 for (j = 0; j < BLK_RING_SIZE(info); j++) { 2082 /* Not in use? */ 2083 if (!shadow[j].request) 2084 continue; 2085 2086 /* 2087 * Get the bios in the request so we can re-queue them. 2088 */ 2089 if (req_op(shadow[j].request) == REQ_OP_FLUSH || 2090 req_op(shadow[j].request) == REQ_OP_DISCARD || 2091 req_op(shadow[j].request) == REQ_OP_SECURE_ERASE || 2092 shadow[j].request->cmd_flags & REQ_FUA) { 2093 /* 2094 * Flush operations don't contain bios, so 2095 * we need to requeue the whole request 2096 * 2097 * XXX: but this doesn't make any sense for a 2098 * write with the FUA flag set.. 2099 */ 2100 list_add(&shadow[j].request->queuelist, &info->requests); 2101 continue; 2102 } 2103 merge_bio.head = shadow[j].request->bio; 2104 merge_bio.tail = shadow[j].request->biotail; 2105 bio_list_merge(&info->bio_list, &merge_bio); 2106 shadow[j].request->bio = NULL; 2107 blk_mq_end_request(shadow[j].request, BLK_STS_OK); 2108 } 2109 } 2110 2111 blkif_free(info, info->connected == BLKIF_STATE_CONNECTED); 2112 2113 err = talk_to_blkback(dev, info); 2114 if (!err) 2115 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings); 2116 2117 /* 2118 * We have to wait for the backend to switch to 2119 * connected state, since we want to read which 2120 * features it supports. 2121 */ 2122 2123 return err; 2124 } 2125 2126 static void blkfront_closing(struct blkfront_info *info) 2127 { 2128 struct xenbus_device *xbdev = info->xbdev; 2129 struct blkfront_ring_info *rinfo; 2130 unsigned int i; 2131 2132 if (xbdev->state == XenbusStateClosing) 2133 return; 2134 2135 /* No more blkif_request(). */ 2136 if (info->rq && info->gd) { 2137 blk_mq_stop_hw_queues(info->rq); 2138 blk_mark_disk_dead(info->gd); 2139 } 2140 2141 for_each_rinfo(info, rinfo, i) { 2142 /* No more gnttab callback work. */ 2143 gnttab_cancel_free_callback(&rinfo->callback); 2144 2145 /* Flush gnttab callback work. Must be done with no locks held. */ 2146 flush_work(&rinfo->work); 2147 } 2148 2149 xenbus_frontend_closed(xbdev); 2150 } 2151 2152 static void blkfront_setup_discard(struct blkfront_info *info) 2153 { 2154 info->feature_discard = 1; 2155 info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend, 2156 "discard-granularity", 2157 0); 2158 info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend, 2159 "discard-alignment", 0); 2160 info->feature_secdiscard = 2161 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure", 2162 0); 2163 } 2164 2165 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo) 2166 { 2167 unsigned int psegs, grants, memflags; 2168 int err, i; 2169 struct blkfront_info *info = rinfo->dev_info; 2170 2171 memflags = memalloc_noio_save(); 2172 2173 if (info->max_indirect_segments == 0) { 2174 if (!HAS_EXTRA_REQ) 2175 grants = BLKIF_MAX_SEGMENTS_PER_REQUEST; 2176 else { 2177 /* 2178 * When an extra req is required, the maximum 2179 * grants supported is related to the size of the 2180 * Linux block segment. 2181 */ 2182 grants = GRANTS_PER_PSEG; 2183 } 2184 } 2185 else 2186 grants = info->max_indirect_segments; 2187 psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG); 2188 2189 err = fill_grant_buffer(rinfo, 2190 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info)); 2191 if (err) 2192 goto out_of_memory; 2193 2194 if (!info->bounce && info->max_indirect_segments) { 2195 /* 2196 * We are using indirect descriptors but don't have a bounce 2197 * buffer, we need to allocate a set of pages that can be 2198 * used for mapping indirect grefs 2199 */ 2200 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info); 2201 2202 BUG_ON(!list_empty(&rinfo->indirect_pages)); 2203 for (i = 0; i < num; i++) { 2204 struct page *indirect_page = alloc_page(GFP_KERNEL | 2205 __GFP_ZERO); 2206 if (!indirect_page) 2207 goto out_of_memory; 2208 list_add(&indirect_page->lru, &rinfo->indirect_pages); 2209 } 2210 } 2211 2212 for (i = 0; i < BLK_RING_SIZE(info); i++) { 2213 rinfo->shadow[i].grants_used = 2214 kvcalloc(grants, 2215 sizeof(rinfo->shadow[i].grants_used[0]), 2216 GFP_KERNEL); 2217 rinfo->shadow[i].sg = kvcalloc(psegs, 2218 sizeof(rinfo->shadow[i].sg[0]), 2219 GFP_KERNEL); 2220 if (info->max_indirect_segments) 2221 rinfo->shadow[i].indirect_grants = 2222 kvcalloc(INDIRECT_GREFS(grants), 2223 sizeof(rinfo->shadow[i].indirect_grants[0]), 2224 GFP_KERNEL); 2225 if ((rinfo->shadow[i].grants_used == NULL) || 2226 (rinfo->shadow[i].sg == NULL) || 2227 (info->max_indirect_segments && 2228 (rinfo->shadow[i].indirect_grants == NULL))) 2229 goto out_of_memory; 2230 sg_init_table(rinfo->shadow[i].sg, psegs); 2231 } 2232 2233 memalloc_noio_restore(memflags); 2234 2235 return 0; 2236 2237 out_of_memory: 2238 for (i = 0; i < BLK_RING_SIZE(info); i++) { 2239 kvfree(rinfo->shadow[i].grants_used); 2240 rinfo->shadow[i].grants_used = NULL; 2241 kvfree(rinfo->shadow[i].sg); 2242 rinfo->shadow[i].sg = NULL; 2243 kvfree(rinfo->shadow[i].indirect_grants); 2244 rinfo->shadow[i].indirect_grants = NULL; 2245 } 2246 if (!list_empty(&rinfo->indirect_pages)) { 2247 struct page *indirect_page, *n; 2248 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) { 2249 list_del(&indirect_page->lru); 2250 __free_page(indirect_page); 2251 } 2252 } 2253 2254 memalloc_noio_restore(memflags); 2255 2256 return -ENOMEM; 2257 } 2258 2259 /* 2260 * Gather all backend feature-* 2261 */ 2262 static void blkfront_gather_backend_features(struct blkfront_info *info) 2263 { 2264 unsigned int indirect_segments; 2265 2266 info->feature_flush = 0; 2267 info->feature_fua = 0; 2268 2269 /* 2270 * If there's no "feature-barrier" defined, then it means 2271 * we're dealing with a very old backend which writes 2272 * synchronously; nothing to do. 2273 * 2274 * If there are barriers, then we use flush. 2275 */ 2276 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) { 2277 info->feature_flush = 1; 2278 info->feature_fua = 1; 2279 } 2280 2281 /* 2282 * And if there is "feature-flush-cache" use that above 2283 * barriers. 2284 */ 2285 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache", 2286 0)) { 2287 info->feature_flush = 1; 2288 info->feature_fua = 0; 2289 } 2290 2291 if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0)) 2292 blkfront_setup_discard(info); 2293 2294 if (info->feature_persistent_parm) 2295 info->feature_persistent = 2296 !!xenbus_read_unsigned(info->xbdev->otherend, 2297 "feature-persistent", 0); 2298 if (info->feature_persistent) 2299 info->bounce = true; 2300 2301 indirect_segments = xenbus_read_unsigned(info->xbdev->otherend, 2302 "feature-max-indirect-segments", 0); 2303 if (indirect_segments > xen_blkif_max_segments) 2304 indirect_segments = xen_blkif_max_segments; 2305 if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST) 2306 indirect_segments = 0; 2307 info->max_indirect_segments = indirect_segments; 2308 2309 if (info->feature_persistent) { 2310 mutex_lock(&blkfront_mutex); 2311 schedule_delayed_work(&blkfront_work, HZ * 10); 2312 mutex_unlock(&blkfront_mutex); 2313 } 2314 } 2315 2316 /* 2317 * Invoked when the backend is finally 'ready' (and has told produced 2318 * the details about the physical device - #sectors, size, etc). 2319 */ 2320 static void blkfront_connect(struct blkfront_info *info) 2321 { 2322 unsigned long long sectors; 2323 unsigned long sector_size; 2324 unsigned int physical_sector_size; 2325 int err, i; 2326 struct blkfront_ring_info *rinfo; 2327 2328 switch (info->connected) { 2329 case BLKIF_STATE_CONNECTED: 2330 /* 2331 * Potentially, the back-end may be signalling 2332 * a capacity change; update the capacity. 2333 */ 2334 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, 2335 "sectors", "%Lu", §ors); 2336 if (XENBUS_EXIST_ERR(err)) 2337 return; 2338 printk(KERN_INFO "Setting capacity to %Lu\n", 2339 sectors); 2340 set_capacity_and_notify(info->gd, sectors); 2341 2342 return; 2343 case BLKIF_STATE_SUSPENDED: 2344 /* 2345 * If we are recovering from suspension, we need to wait 2346 * for the backend to announce it's features before 2347 * reconnecting, at least we need to know if the backend 2348 * supports indirect descriptors, and how many. 2349 */ 2350 blkif_recover(info); 2351 return; 2352 2353 default: 2354 break; 2355 } 2356 2357 dev_dbg(&info->xbdev->dev, "%s:%s.\n", 2358 __func__, info->xbdev->otherend); 2359 2360 err = xenbus_gather(XBT_NIL, info->xbdev->otherend, 2361 "sectors", "%llu", §ors, 2362 "info", "%u", &info->vdisk_info, 2363 "sector-size", "%lu", §or_size, 2364 NULL); 2365 if (err) { 2366 xenbus_dev_fatal(info->xbdev, err, 2367 "reading backend fields at %s", 2368 info->xbdev->otherend); 2369 return; 2370 } 2371 2372 /* 2373 * physical-sector-size is a newer field, so old backends may not 2374 * provide this. Assume physical sector size to be the same as 2375 * sector_size in that case. 2376 */ 2377 physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend, 2378 "physical-sector-size", 2379 sector_size); 2380 blkfront_gather_backend_features(info); 2381 for_each_rinfo(info, rinfo, i) { 2382 err = blkfront_setup_indirect(rinfo); 2383 if (err) { 2384 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s", 2385 info->xbdev->otherend); 2386 blkif_free(info, 0); 2387 break; 2388 } 2389 } 2390 2391 err = xlvbd_alloc_gendisk(sectors, info, sector_size, 2392 physical_sector_size); 2393 if (err) { 2394 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s", 2395 info->xbdev->otherend); 2396 goto fail; 2397 } 2398 2399 xenbus_switch_state(info->xbdev, XenbusStateConnected); 2400 2401 /* Kick pending requests. */ 2402 info->connected = BLKIF_STATE_CONNECTED; 2403 for_each_rinfo(info, rinfo, i) 2404 kick_pending_request_queues(rinfo); 2405 2406 err = device_add_disk(&info->xbdev->dev, info->gd, NULL); 2407 if (err) { 2408 put_disk(info->gd); 2409 blk_mq_free_tag_set(&info->tag_set); 2410 info->rq = NULL; 2411 goto fail; 2412 } 2413 2414 info->is_ready = 1; 2415 return; 2416 2417 fail: 2418 blkif_free(info, 0); 2419 return; 2420 } 2421 2422 /* 2423 * Callback received when the backend's state changes. 2424 */ 2425 static void blkback_changed(struct xenbus_device *dev, 2426 enum xenbus_state backend_state) 2427 { 2428 struct blkfront_info *info = dev_get_drvdata(&dev->dev); 2429 2430 dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state); 2431 2432 switch (backend_state) { 2433 case XenbusStateInitWait: 2434 if (dev->state != XenbusStateInitialising) 2435 break; 2436 if (talk_to_blkback(dev, info)) 2437 break; 2438 break; 2439 case XenbusStateInitialising: 2440 case XenbusStateInitialised: 2441 case XenbusStateReconfiguring: 2442 case XenbusStateReconfigured: 2443 case XenbusStateUnknown: 2444 break; 2445 2446 case XenbusStateConnected: 2447 /* 2448 * talk_to_blkback sets state to XenbusStateInitialised 2449 * and blkfront_connect sets it to XenbusStateConnected 2450 * (if connection went OK). 2451 * 2452 * If the backend (or toolstack) decides to poke at backend 2453 * state (and re-trigger the watch by setting the state repeatedly 2454 * to XenbusStateConnected (4)) we need to deal with this. 2455 * This is allowed as this is used to communicate to the guest 2456 * that the size of disk has changed! 2457 */ 2458 if ((dev->state != XenbusStateInitialised) && 2459 (dev->state != XenbusStateConnected)) { 2460 if (talk_to_blkback(dev, info)) 2461 break; 2462 } 2463 2464 blkfront_connect(info); 2465 break; 2466 2467 case XenbusStateClosed: 2468 if (dev->state == XenbusStateClosed) 2469 break; 2470 fallthrough; 2471 case XenbusStateClosing: 2472 blkfront_closing(info); 2473 break; 2474 } 2475 } 2476 2477 static void blkfront_remove(struct xenbus_device *xbdev) 2478 { 2479 struct blkfront_info *info = dev_get_drvdata(&xbdev->dev); 2480 2481 dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename); 2482 2483 if (info->gd) 2484 del_gendisk(info->gd); 2485 2486 mutex_lock(&blkfront_mutex); 2487 list_del(&info->info_list); 2488 mutex_unlock(&blkfront_mutex); 2489 2490 blkif_free(info, 0); 2491 if (info->gd) { 2492 xlbd_release_minors(info->gd->first_minor, info->gd->minors); 2493 put_disk(info->gd); 2494 blk_mq_free_tag_set(&info->tag_set); 2495 } 2496 2497 kfree(info); 2498 } 2499 2500 static int blkfront_is_ready(struct xenbus_device *dev) 2501 { 2502 struct blkfront_info *info = dev_get_drvdata(&dev->dev); 2503 2504 return info->is_ready && info->xbdev; 2505 } 2506 2507 static const struct block_device_operations xlvbd_block_fops = 2508 { 2509 .owner = THIS_MODULE, 2510 .getgeo = blkif_getgeo, 2511 .ioctl = blkif_ioctl, 2512 .compat_ioctl = blkdev_compat_ptr_ioctl, 2513 }; 2514 2515 2516 static const struct xenbus_device_id blkfront_ids[] = { 2517 { "vbd" }, 2518 { "" } 2519 }; 2520 2521 static struct xenbus_driver blkfront_driver = { 2522 .ids = blkfront_ids, 2523 .probe = blkfront_probe, 2524 .remove = blkfront_remove, 2525 .resume = blkfront_resume, 2526 .otherend_changed = blkback_changed, 2527 .is_ready = blkfront_is_ready, 2528 }; 2529 2530 static void purge_persistent_grants(struct blkfront_info *info) 2531 { 2532 unsigned int i; 2533 unsigned long flags; 2534 struct blkfront_ring_info *rinfo; 2535 2536 for_each_rinfo(info, rinfo, i) { 2537 struct grant *gnt_list_entry, *tmp; 2538 LIST_HEAD(grants); 2539 2540 spin_lock_irqsave(&rinfo->ring_lock, flags); 2541 2542 if (rinfo->persistent_gnts_c == 0) { 2543 spin_unlock_irqrestore(&rinfo->ring_lock, flags); 2544 continue; 2545 } 2546 2547 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants, 2548 node) { 2549 if (gnt_list_entry->gref == INVALID_GRANT_REF || 2550 !gnttab_try_end_foreign_access(gnt_list_entry->gref)) 2551 continue; 2552 2553 list_del(&gnt_list_entry->node); 2554 rinfo->persistent_gnts_c--; 2555 gnt_list_entry->gref = INVALID_GRANT_REF; 2556 list_add_tail(&gnt_list_entry->node, &grants); 2557 } 2558 2559 list_splice_tail(&grants, &rinfo->grants); 2560 2561 spin_unlock_irqrestore(&rinfo->ring_lock, flags); 2562 } 2563 } 2564 2565 static void blkfront_delay_work(struct work_struct *work) 2566 { 2567 struct blkfront_info *info; 2568 bool need_schedule_work = false; 2569 2570 /* 2571 * Note that when using bounce buffers but not persistent grants 2572 * there's no need to run blkfront_delay_work because grants are 2573 * revoked in blkif_completion or else an error is reported and the 2574 * connection is closed. 2575 */ 2576 2577 mutex_lock(&blkfront_mutex); 2578 2579 list_for_each_entry(info, &info_list, info_list) { 2580 if (info->feature_persistent) { 2581 need_schedule_work = true; 2582 mutex_lock(&info->mutex); 2583 purge_persistent_grants(info); 2584 mutex_unlock(&info->mutex); 2585 } 2586 } 2587 2588 if (need_schedule_work) 2589 schedule_delayed_work(&blkfront_work, HZ * 10); 2590 2591 mutex_unlock(&blkfront_mutex); 2592 } 2593 2594 static int __init xlblk_init(void) 2595 { 2596 int ret; 2597 int nr_cpus = num_online_cpus(); 2598 2599 if (!xen_domain()) 2600 return -ENODEV; 2601 2602 if (!xen_has_pv_disk_devices()) 2603 return -ENODEV; 2604 2605 if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) { 2606 pr_warn("xen_blk: can't get major %d with name %s\n", 2607 XENVBD_MAJOR, DEV_NAME); 2608 return -ENODEV; 2609 } 2610 2611 if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST) 2612 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST; 2613 2614 if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) { 2615 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n", 2616 xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER); 2617 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER; 2618 } 2619 2620 if (xen_blkif_max_queues > nr_cpus) { 2621 pr_info("Invalid max_queues (%d), will use default max: %d.\n", 2622 xen_blkif_max_queues, nr_cpus); 2623 xen_blkif_max_queues = nr_cpus; 2624 } 2625 2626 INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work); 2627 2628 ret = xenbus_register_frontend(&blkfront_driver); 2629 if (ret) { 2630 unregister_blkdev(XENVBD_MAJOR, DEV_NAME); 2631 return ret; 2632 } 2633 2634 return 0; 2635 } 2636 module_init(xlblk_init); 2637 2638 2639 static void __exit xlblk_exit(void) 2640 { 2641 cancel_delayed_work_sync(&blkfront_work); 2642 2643 xenbus_unregister_driver(&blkfront_driver); 2644 unregister_blkdev(XENVBD_MAJOR, DEV_NAME); 2645 kfree(minors); 2646 } 2647 module_exit(xlblk_exit); 2648 2649 MODULE_DESCRIPTION("Xen virtual block device frontend"); 2650 MODULE_LICENSE("GPL"); 2651 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR); 2652 MODULE_ALIAS("xen:vbd"); 2653 MODULE_ALIAS("xenblk"); 2654