drivers/block/rbd.c

   1 /*
   2    rbd.c -- Export ceph rados objects as a Linux block device
   3
   4
   5    based on drivers/block/osdblk.c:
   6
   7    Copyright 2009 Red Hat, Inc.
   8
   9    This program is free software; you can redistribute it and/or modify
  10    it under the terms of the GNU General Public License as published by
  11    the Free Software Foundation.
  12
  13    This program is distributed in the hope that it will be useful,
  14    but WITHOUT ANY WARRANTY; without even the implied warranty of
  15    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16    GNU General Public License for more details.
  17
  18    You should have received a copy of the GNU General Public License
  19    along with this program; see the file COPYING.  If not, write to
  20    the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
  21
  22
  23
  24    For usage instructions, please refer to:
  25
  26                  Documentation/ABI/testing/sysfs-bus-rbd
  27
  28  */
  29
  30 #include <linux/ceph/libceph.h>
  31 #include <linux/ceph/osd_client.h>
  32 #include <linux/ceph/mon_client.h>
  33 #include <linux/ceph/decode.h>
  34 #include <linux/parser.h>
  35
  36 #include <linux/kernel.h>
  37 #include <linux/device.h>
  38 #include <linux/module.h>
  39 #include <linux/fs.h>
  40 #include <linux/blkdev.h>
  41
  42 #include "rbd_types.h"
  43
  44 #define DRV_NAME "rbd"
  45 #define DRV_NAME_LONG "rbd (rados block device)"
  46
  47 #define RBD_MINORS_PER_MAJOR    256             /* max minors per blkdev */
  48
  49 #define RBD_MAX_MD_NAME_LEN     (96 + sizeof(RBD_SUFFIX))
  50 #define RBD_MAX_POOL_NAME_LEN   64
  51 #define RBD_MAX_SNAP_NAME_LEN   32
  52 #define RBD_MAX_OPT_LEN         1024
  53
  54 #define RBD_SNAP_HEAD_NAME      "-"
  55
  56 #define DEV_NAME_LEN            32
  57
  58 #define RBD_NOTIFY_TIMEOUT_DEFAULT 10
  59
  60 /*
  61  * block device image metadata (in-memory version)
  62  */
  63 struct rbd_image_header {
  64         u64 image_size;
  65         char block_name[32];
  66         __u8 obj_order;
  67         __u8 crypt_type;
  68         __u8 comp_type;
  69         struct rw_semaphore snap_rwsem;
  70         struct ceph_snap_context *snapc;
  71         size_t snap_names_len;
  72         u64 snap_seq;
  73         u32 total_snaps;
  74
  75         char *snap_names;
  76         u64 *snap_sizes;
  77
  78         u64 obj_version;
  79 };
  80
  81 struct rbd_options {
  82         int     notify_timeout;
  83 };
  84
  85 /*
  86  * an instance of the client.  multiple devices may share a client.
  87  */
  88 struct rbd_client {
  89         struct ceph_client      *client;
  90         struct rbd_options      *rbd_opts;
  91         struct kref             kref;
  92         struct list_head        node;
  93 };
  94
  95 struct rbd_req_coll;
  96
  97 /*
  98  * a single io request
  99  */
 100 struct rbd_request {
 101         struct request          *rq;            /* blk layer request */
 102         struct bio              *bio;           /* cloned bio */
 103         struct page             **pages;        /* list of used pages */
 104         u64                     len;
 105         int                     coll_index;
 106         struct rbd_req_coll     *coll;
 107 };
 108
 109 struct rbd_req_status {
 110         int done;
 111         int rc;
 112         u64 bytes;
 113 };
 114
 115 /*
 116  * a collection of requests
 117  */
 118 struct rbd_req_coll {
 119         int                     total;
 120         int                     num_done;
 121         struct kref             kref;
 122         struct rbd_req_status   status[0];
 123 };
 124
 125 struct rbd_snap {
 126         struct  device          dev;
 127         const char              *name;
 128         size_t                  size;
 129         struct list_head        node;
 130         u64                     id;
 131 };
 132
 133 /*
 134  * a single device
 135  */
 136 struct rbd_device {
 137         int                     id;             /* blkdev unique id */
 138
 139         int                     major;          /* blkdev assigned major */
 140         struct gendisk          *disk;          /* blkdev's gendisk and rq */
 141         struct request_queue    *q;
 142
 143         struct ceph_client      *client;
 144         struct rbd_client       *rbd_client;
 145
 146         char                    name[DEV_NAME_LEN]; /* blkdev name, e.g. rbd3 */
 147
 148         spinlock_t              lock;           /* queue lock */
 149
 150         struct rbd_image_header header;
 151         char                    obj[RBD_MAX_OBJ_NAME_LEN]; /* rbd image name */
 152         int                     obj_len;
 153         char                    obj_md_name[RBD_MAX_MD_NAME_LEN]; /* hdr nm. */
 154         char                    pool_name[RBD_MAX_POOL_NAME_LEN];
 155         int                     poolid;
 156
 157         struct ceph_osd_event   *watch_event;
 158         struct ceph_osd_request *watch_request;
 159
 160         char                    snap_name[RBD_MAX_SNAP_NAME_LEN];
 161         u32 cur_snap;   /* index+1 of current snapshot within snap context
 162                            0 - for the head */
 163         int read_only;
 164
 165         struct list_head        node;
 166
 167         /* list of snapshots */
 168         struct list_head        snaps;
 169
 170         /* sysfs related */
 171         struct device           dev;
 172 };
 173
 174 static struct bus_type rbd_bus_type = {
 175         .name           = "rbd",
 176 };
 177
 178 static spinlock_t node_lock;      /* protects client get/put */
 179
 180 static DEFINE_MUTEX(ctl_mutex);   /* Serialize open/close/setup/teardown */
 181 static LIST_HEAD(rbd_dev_list);    /* devices */
 182 static LIST_HEAD(rbd_client_list);      /* clients */
 183
 184 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev);
 185 static void rbd_dev_release(struct device *dev);
 186 static ssize_t rbd_snap_rollback(struct device *dev,
 187                                  struct device_attribute *attr,
 188                                  const char *buf,
 189                                  size_t size);
 190 static ssize_t rbd_snap_add(struct device *dev,
 191                             struct device_attribute *attr,
 192                             const char *buf,
 193                             size_t count);
 194 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
 195                                   struct rbd_snap *snap);;
 196
 197
 198 static struct rbd_device *dev_to_rbd(struct device *dev)
 199 {
 200         return container_of(dev, struct rbd_device, dev);
 201 }
 202
 203 static struct device *rbd_get_dev(struct rbd_device *rbd_dev)
 204 {
 205         return get_device(&rbd_dev->dev);
 206 }
 207
 208 static void rbd_put_dev(struct rbd_device *rbd_dev)
 209 {
 210         put_device(&rbd_dev->dev);
 211 }
 212
 213 static int __rbd_update_snaps(struct rbd_device *rbd_dev);
 214
 215 static int rbd_open(struct block_device *bdev, fmode_t mode)
 216 {
 217         struct gendisk *disk = bdev->bd_disk;
 218         struct rbd_device *rbd_dev = disk->private_data;
 219
 220         rbd_get_dev(rbd_dev);
 221
 222         set_device_ro(bdev, rbd_dev->read_only);
 223
 224         if ((mode & FMODE_WRITE) && rbd_dev->read_only)
 225                 return -EROFS;
 226
 227         return 0;
 228 }
 229
 230 static int rbd_release(struct gendisk *disk, fmode_t mode)
 231 {
 232         struct rbd_device *rbd_dev = disk->private_data;
 233
 234         rbd_put_dev(rbd_dev);
 235
 236         return 0;
 237 }
 238
 239 static const struct block_device_operations rbd_bd_ops = {
 240         .owner                  = THIS_MODULE,
 241         .open                   = rbd_open,
 242         .release                = rbd_release,
 243 };
 244
 245 /*
 246  * Initialize an rbd client instance.
 247  * We own *opt.
 248  */
 249 static struct rbd_client *rbd_client_create(struct ceph_options *opt,
 250                                             struct rbd_options *rbd_opts)
 251 {
 252         struct rbd_client *rbdc;
 253         int ret = -ENOMEM;
 254
 255         dout("rbd_client_create\n");
 256         rbdc = kmalloc(sizeof(struct rbd_client), GFP_KERNEL);
 257         if (!rbdc)
 258                 goto out_opt;
 259
 260         kref_init(&rbdc->kref);
 261         INIT_LIST_HEAD(&rbdc->node);
 262
 263         rbdc->client = ceph_create_client(opt, rbdc);
 264         if (IS_ERR(rbdc->client))
 265                 goto out_rbdc;
 266         opt = NULL; /* Now rbdc->client is responsible for opt */
 267
 268         ret = ceph_open_session(rbdc->client);
 269         if (ret < 0)
 270                 goto out_err;
 271
 272         rbdc->rbd_opts = rbd_opts;
 273
 274         spin_lock(&node_lock);
 275         list_add_tail(&rbdc->node, &rbd_client_list);
 276         spin_unlock(&node_lock);
 277
 278         dout("rbd_client_create created %p\n", rbdc);
 279         return rbdc;
 280
 281 out_err:
 282         ceph_destroy_client(rbdc->client);
 283 out_rbdc:
 284         kfree(rbdc);
 285 out_opt:
 286         if (opt)
 287                 ceph_destroy_options(opt);
 288         return ERR_PTR(ret);
 289 }
 290
 291 /*
 292  * Find a ceph client with specific addr and configuration.
 293  */
 294 static struct rbd_client *__rbd_client_find(struct ceph_options *opt)
 295 {
 296         struct rbd_client *client_node;
 297
 298         if (opt->flags & CEPH_OPT_NOSHARE)
 299                 return NULL;
 300
 301         list_for_each_entry(client_node, &rbd_client_list, node)
 302                 if (ceph_compare_options(opt, client_node->client) == 0)
 303                         return client_node;
 304         return NULL;
 305 }
 306
 307 /*
 308  * mount options
 309  */
 310 enum {
 311         Opt_notify_timeout,
 312         Opt_last_int,
 313         /* int args above */
 314         Opt_last_string,
 315         /* string args above */
 316 };
 317
 318 static match_table_t rbdopt_tokens = {
 319         {Opt_notify_timeout, "notify_timeout=%d"},
 320         /* int args above */
 321         /* string args above */
 322         {-1, NULL}
 323 };
 324
 325 static int parse_rbd_opts_token(char *c, void *private)
 326 {
 327         struct rbd_options *rbdopt = private;
 328         substring_t argstr[MAX_OPT_ARGS];
 329         int token, intval, ret;
 330
 331         token = match_token((char *)c, rbdopt_tokens, argstr);
 332         if (token < 0)
 333                 return -EINVAL;
 334
 335         if (token < Opt_last_int) {
 336                 ret = match_int(&argstr[0], &intval);
 337                 if (ret < 0) {
 338                         pr_err("bad mount option arg (not int) "
 339                                "at '%s'\n", c);
 340                         return ret;
 341                 }
 342                 dout("got int token %d val %d\n", token, intval);
 343         } else if (token > Opt_last_int && token < Opt_last_string) {
 344                 dout("got string token %d val %s\n", token,
 345                      argstr[0].from);
 346         } else {
 347                 dout("got token %d\n", token);
 348         }
 349
 350         switch (token) {
 351         case Opt_notify_timeout:
 352                 rbdopt->notify_timeout = intval;
 353                 break;
 354         default:
 355                 BUG_ON(token);
 356         }
 357         return 0;
 358 }
 359
 360 /*
 361  * Get a ceph client with specific addr and configuration, if one does
 362  * not exist create it.
 363  */
 364 static int rbd_get_client(struct rbd_device *rbd_dev, const char *mon_addr,
 365                           char *options)
 366 {
 367         struct rbd_client *rbdc;
 368         struct ceph_options *opt;
 369         int ret;
 370         struct rbd_options *rbd_opts;
 371
 372         rbd_opts = kzalloc(sizeof(*rbd_opts), GFP_KERNEL);
 373         if (!rbd_opts)
 374                 return -ENOMEM;
 375
 376         rbd_opts->notify_timeout = RBD_NOTIFY_TIMEOUT_DEFAULT;
 377
 378         ret = ceph_parse_options(&opt, options, mon_addr,
 379                                  mon_addr + strlen(mon_addr), parse_rbd_opts_token, rbd_opts);
 380         if (ret < 0)
 381                 goto done_err;
 382
 383         spin_lock(&node_lock);
 384         rbdc = __rbd_client_find(opt);
 385         if (rbdc) {
 386                 ceph_destroy_options(opt);
 387
 388                 /* using an existing client */
 389                 kref_get(&rbdc->kref);
 390                 rbd_dev->rbd_client = rbdc;
 391                 rbd_dev->client = rbdc->client;
 392                 spin_unlock(&node_lock);
 393                 return 0;
 394         }
 395         spin_unlock(&node_lock);
 396
 397         rbdc = rbd_client_create(opt, rbd_opts);
 398         if (IS_ERR(rbdc)) {
 399                 ret = PTR_ERR(rbdc);
 400                 goto done_err;
 401         }
 402
 403         rbd_dev->rbd_client = rbdc;
 404         rbd_dev->client = rbdc->client;
 405         return 0;
 406 done_err:
 407         kfree(rbd_opts);
 408         return ret;
 409 }
 410
 411 /*
 412  * Destroy ceph client
 413  */
 414 static void rbd_client_release(struct kref *kref)
 415 {
 416         struct rbd_client *rbdc = container_of(kref, struct rbd_client, kref);
 417
 418         dout("rbd_release_client %p\n", rbdc);
 419         spin_lock(&node_lock);
 420         list_del(&rbdc->node);
 421         spin_unlock(&node_lock);
 422
 423         ceph_destroy_client(rbdc->client);
 424         kfree(rbdc->rbd_opts);
 425         kfree(rbdc);
 426 }
 427
 428 /*
 429  * Drop reference to ceph client node. If it's not referenced anymore, release
 430  * it.
 431  */
 432 static void rbd_put_client(struct rbd_device *rbd_dev)
 433 {
 434         kref_put(&rbd_dev->rbd_client->kref, rbd_client_release);
 435         rbd_dev->rbd_client = NULL;
 436         rbd_dev->client = NULL;
 437 }
 438
 439 /*
 440  * Destroy requests collection
 441  */
 442 static void rbd_coll_release(struct kref *kref)
 443 {
 444         struct rbd_req_coll *coll =
 445                 container_of(kref, struct rbd_req_coll, kref);
 446
 447         dout("rbd_coll_release %p\n", coll);
 448         kfree(coll);
 449 }
 450
 451 /*
 452  * Create a new header structure, translate header format from the on-disk
 453  * header.
 454  */
 455 static int rbd_header_from_disk(struct rbd_image_header *header,
 456                                  struct rbd_image_header_ondisk *ondisk,
 457                                  int allocated_snaps,
 458                                  gfp_t gfp_flags)
 459 {
 460         int i;
 461         u32 snap_count = le32_to_cpu(ondisk->snap_count);
 462         int ret = -ENOMEM;
 463
 464         init_rwsem(&header->snap_rwsem);
 465         header->snap_names_len = le64_to_cpu(ondisk->snap_names_len);
 466         header->snapc = kmalloc(sizeof(struct ceph_snap_context) +
 467                                 snap_count *
 468                                  sizeof(struct rbd_image_snap_ondisk),
 469                                 gfp_flags);
 470         if (!header->snapc)
 471                 return -ENOMEM;
 472         if (snap_count) {
 473                 header->snap_names = kmalloc(header->snap_names_len,
 474                                              GFP_KERNEL);
 475                 if (!header->snap_names)
 476                         goto err_snapc;
 477                 header->snap_sizes = kmalloc(snap_count * sizeof(u64),
 478                                              GFP_KERNEL);
 479                 if (!header->snap_sizes)
 480                         goto err_names;
 481         } else {
 482                 header->snap_names = NULL;
 483                 header->snap_sizes = NULL;
 484         }
 485         memcpy(header->block_name, ondisk->block_name,
 486                sizeof(ondisk->block_name));
 487
 488         header->image_size = le64_to_cpu(ondisk->image_size);
 489         header->obj_order = ondisk->options.order;
 490         header->crypt_type = ondisk->options.crypt_type;
 491         header->comp_type = ondisk->options.comp_type;
 492
 493         atomic_set(&header->snapc->nref, 1);
 494         header->snap_seq = le64_to_cpu(ondisk->snap_seq);
 495         header->snapc->num_snaps = snap_count;
 496         header->total_snaps = snap_count;
 497
 498         if (snap_count &&
 499             allocated_snaps == snap_count) {
 500                 for (i = 0; i < snap_count; i++) {
 501                         header->snapc->snaps[i] =
 502                                 le64_to_cpu(ondisk->snaps[i].id);
 503                         header->snap_sizes[i] =
 504                                 le64_to_cpu(ondisk->snaps[i].image_size);
 505                 }
 506
 507                 /* copy snapshot names */
 508                 memcpy(header->snap_names, &ondisk->snaps[i],
 509                         header->snap_names_len);
 510         }
 511
 512         return 0;
 513
 514 err_names:
 515         kfree(header->snap_names);
 516 err_snapc:
 517         kfree(header->snapc);
 518         return ret;
 519 }
 520
 521 static int snap_index(struct rbd_image_header *header, int snap_num)
 522 {
 523         return header->total_snaps - snap_num;
 524 }
 525
 526 static u64 cur_snap_id(struct rbd_device *rbd_dev)
 527 {
 528         struct rbd_image_header *header = &rbd_dev->header;
 529
 530         if (!rbd_dev->cur_snap)
 531                 return 0;
 532
 533         return header->snapc->snaps[snap_index(header, rbd_dev->cur_snap)];
 534 }
 535
 536 static int snap_by_name(struct rbd_image_header *header, const char *snap_name,
 537                         u64 *seq, u64 *size)
 538 {
 539         int i;
 540         char *p = header->snap_names;
 541
 542         for (i = 0; i < header->total_snaps; i++, p += strlen(p) + 1) {
 543                 if (strcmp(snap_name, p) == 0)
 544                         break;
 545         }
 546         if (i == header->total_snaps)
 547                 return -ENOENT;
 548         if (seq)
 549                 *seq = header->snapc->snaps[i];
 550
 551         if (size)
 552                 *size = header->snap_sizes[i];
 553
 554         return i;
 555 }
 556
 557 static int rbd_header_set_snap(struct rbd_device *dev,
 558                                const char *snap_name,
 559                                u64 *size)
 560 {
 561         struct rbd_image_header *header = &dev->header;
 562         struct ceph_snap_context *snapc = header->snapc;
 563         int ret = -ENOENT;
 564
 565         down_write(&header->snap_rwsem);
 566
 567         if (!snap_name ||
 568             !*snap_name ||
 569             strcmp(snap_name, "-") == 0 ||
 570             strcmp(snap_name, RBD_SNAP_HEAD_NAME) == 0) {
 571                 if (header->total_snaps)
 572                         snapc->seq = header->snap_seq;
 573                 else
 574                         snapc->seq = 0;
 575                 dev->cur_snap = 0;
 576                 dev->read_only = 0;
 577                 if (size)
 578                         *size = header->image_size;
 579         } else {
 580                 ret = snap_by_name(header, snap_name, &snapc->seq, size);
 581                 if (ret < 0)
 582                         goto done;
 583
 584                 dev->cur_snap = header->total_snaps - ret;
 585                 dev->read_only = 1;
 586         }
 587
 588         ret = 0;
 589 done:
 590         up_write(&header->snap_rwsem);
 591         return ret;
 592 }
 593
 594 static void rbd_header_free(struct rbd_image_header *header)
 595 {
 596         kfree(header->snapc);
 597         kfree(header->snap_names);
 598         kfree(header->snap_sizes);
 599 }
 600
 601 /*
 602  * get the actual striped segment name, offset and length
 603  */
 604 static u64 rbd_get_segment(struct rbd_image_header *header,
 605                            const char *block_name,
 606                            u64 ofs, u64 len,
 607                            char *seg_name, u64 *segofs)
 608 {
 609         u64 seg = ofs >> header->obj_order;
 610
 611         if (seg_name)
 612                 snprintf(seg_name, RBD_MAX_SEG_NAME_LEN,
 613                          "%s.%012llx", block_name, seg);
 614
 615         ofs = ofs & ((1 << header->obj_order) - 1);
 616         len = min_t(u64, len, (1 << header->obj_order) - ofs);
 617
 618         if (segofs)
 619                 *segofs = ofs;
 620
 621         return len;
 622 }
 623
 624 static int rbd_get_num_segments(struct rbd_image_header *header,
 625                                 u64 ofs, u64 len)
 626 {
 627         u64 start_seg = ofs >> header->obj_order;
 628         u64 end_seg = (ofs + len - 1) >> header->obj_order;
 629         return end_seg - start_seg + 1;
 630 }
 631
 632 /*
 633  * bio helpers
 634  */
 635
 636 static void bio_chain_put(struct bio *chain)
 637 {
 638         struct bio *tmp;
 639
 640         while (chain) {
 641                 tmp = chain;
 642                 chain = chain->bi_next;
 643                 bio_put(tmp);
 644         }
 645 }
 646
 647 /*
 648  * zeros a bio chain, starting at specific offset
 649  */
 650 static void zero_bio_chain(struct bio *chain, int start_ofs)
 651 {
 652         struct bio_vec *bv;
 653         unsigned long flags;
 654         void *buf;
 655         int i;
 656         int pos = 0;
 657
 658         while (chain) {
 659                 bio_for_each_segment(bv, chain, i) {
 660                         if (pos + bv->bv_len > start_ofs) {
 661                                 int remainder = max(start_ofs - pos, 0);
 662                                 buf = bvec_kmap_irq(bv, &flags);
 663                                 memset(buf + remainder, 0,
 664                                        bv->bv_len - remainder);
 665                                 bvec_kunmap_irq(buf, &flags);
 666                         }
 667                         pos += bv->bv_len;
 668                 }
 669
 670                 chain = chain->bi_next;
 671         }
 672 }
 673
 674 /*
 675  * bio_chain_clone - clone a chain of bios up to a certain length.
 676  * might return a bio_pair that will need to be released.
 677  */
 678 static struct bio *bio_chain_clone(struct bio **old, struct bio **next,
 679                                    struct bio_pair **bp,
 680                                    int len, gfp_t gfpmask)
 681 {
 682         struct bio *tmp, *old_chain = *old, *new_chain = NULL, *tail = NULL;
 683         int total = 0;
 684
 685         if (*bp) {
 686                 bio_pair_release(*bp);
 687                 *bp = NULL;
 688         }
 689
 690         while (old_chain && (total < len)) {
 691                 tmp = bio_kmalloc(gfpmask, old_chain->bi_max_vecs);
 692                 if (!tmp)
 693                         goto err_out;
 694
 695                 if (total + old_chain->bi_size > len) {
 696                         struct bio_pair *bp;
 697
 698                         /*
 699                          * this split can only happen with a single paged bio,
 700                          * split_bio will BUG_ON if this is not the case
 701                          */
 702                         dout("bio_chain_clone split! total=%d remaining=%d"
 703                              "bi_size=%d\n",
 704                              (int)total, (int)len-total,
 705                              (int)old_chain->bi_size);
 706
 707                         /* split the bio. We'll release it either in the next
 708                            call, or it will have to be released outside */
 709                         bp = bio_split(old_chain, (len - total) / 512ULL);
 710                         if (!bp)
 711                                 goto err_out;
 712
 713                         __bio_clone(tmp, &bp->bio1);
 714
 715                         *next = &bp->bio2;
 716                 } else {
 717                         __bio_clone(tmp, old_chain);
 718                         *next = old_chain->bi_next;
 719                 }
 720
 721                 tmp->bi_bdev = NULL;
 722                 gfpmask &= ~__GFP_WAIT;
 723                 tmp->bi_next = NULL;
 724
 725                 if (!new_chain) {
 726                         new_chain = tail = tmp;
 727                 } else {
 728                         tail->bi_next = tmp;
 729                         tail = tmp;
 730                 }
 731                 old_chain = old_chain->bi_next;
 732
 733                 total += tmp->bi_size;
 734         }
 735
 736         BUG_ON(total < len);
 737
 738         if (tail)
 739                 tail->bi_next = NULL;
 740
 741         *old = old_chain;
 742
 743         return new_chain;
 744
 745 err_out:
 746         dout("bio_chain_clone with err\n");
 747         bio_chain_put(new_chain);
 748         return NULL;
 749 }
 750
 751 /*
 752  * helpers for osd request op vectors.
 753  */
 754 static int rbd_create_rw_ops(struct ceph_osd_req_op **ops,
 755                             int num_ops,
 756                             int opcode,
 757                             u32 payload_len)
 758 {
 759         *ops = kzalloc(sizeof(struct ceph_osd_req_op) * (num_ops + 1),
 760                        GFP_NOIO);
 761         if (!*ops)
 762                 return -ENOMEM;
 763         (*ops)[0].op = opcode;
 764         /*
 765          * op extent offset and length will be set later on
 766          * in calc_raw_layout()
 767          */
 768         (*ops)[0].payload_len = payload_len;
 769         return 0;
 770 }
 771
 772 static void rbd_destroy_ops(struct ceph_osd_req_op *ops)
 773 {
 774         kfree(ops);
 775 }
 776
 777 static void rbd_coll_end_req_index(struct request *rq,
 778                                    struct rbd_req_coll *coll,
 779                                    int index,
 780                                    int ret, u64 len)
 781 {
 782         struct request_queue *q;
 783         int min, max, i;
 784
 785         dout("rbd_coll_end_req_index %p index %d ret %d len %lld\n",
 786              coll, index, ret, len);
 787
 788         if (!rq)
 789                 return;
 790
 791         if (!coll) {
 792                 blk_end_request(rq, ret, len);
 793                 return;
 794         }
 795
 796         q = rq->q;
 797
 798         spin_lock_irq(q->queue_lock);
 799         coll->status[index].done = 1;
 800         coll->status[index].rc = ret;
 801         coll->status[index].bytes = len;
 802         max = min = coll->num_done;
 803         while (max < coll->total && coll->status[max].done)
 804                 max++;
 805
 806         for (i = min; i<max; i++) {
 807                 __blk_end_request(rq, coll->status[i].rc,
 808                                   coll->status[i].bytes);
 809                 coll->num_done++;
 810                 kref_put(&coll->kref, rbd_coll_release);
 811         }
 812         spin_unlock_irq(q->queue_lock);
 813 }
 814
 815 static void rbd_coll_end_req(struct rbd_request *req,
 816                              int ret, u64 len)
 817 {
 818         rbd_coll_end_req_index(req->rq, req->coll, req->coll_index, ret, len);
 819 }
 820
 821 /*
 822  * Send ceph osd request
 823  */
 824 static int rbd_do_request(struct request *rq,
 825                           struct rbd_device *dev,
 826                           struct ceph_snap_context *snapc,
 827                           u64 snapid,
 828                           const char *obj, u64 ofs, u64 len,
 829                           struct bio *bio,
 830                           struct page **pages,
 831                           int num_pages,
 832                           int flags,
 833                           struct ceph_osd_req_op *ops,
 834                           int num_reply,
 835                           struct rbd_req_coll *coll,
 836                           int coll_index,
 837                           void (*rbd_cb)(struct ceph_osd_request *req,
 838                                          struct ceph_msg *msg),
 839                           struct ceph_osd_request **linger_req,
 840                           u64 *ver)
 841 {
 842         struct ceph_osd_request *req;
 843         struct ceph_file_layout *layout;
 844         int ret;
 845         u64 bno;
 846         struct timespec mtime = CURRENT_TIME;
 847         struct rbd_request *req_data;
 848         struct ceph_osd_request_head *reqhead;
 849         struct rbd_image_header *header = &dev->header;
 850
 851         req_data = kzalloc(sizeof(*req_data), GFP_NOIO);
 852         if (!req_data) {
 853                 if (coll)
 854                         rbd_coll_end_req_index(rq, coll, coll_index,
 855                                                -ENOMEM, len);
 856                 return -ENOMEM;
 857         }
 858
 859         if (coll) {
 860                 req_data->coll = coll;
 861                 req_data->coll_index = coll_index;
 862         }
 863
 864         dout("rbd_do_request obj=%s ofs=%lld len=%lld\n", obj, len, ofs);
 865
 866         down_read(&header->snap_rwsem);
 867
 868         req = ceph_osdc_alloc_request(&dev->client->osdc, flags,
 869                                       snapc,
 870                                       ops,
 871                                       false,
 872                                       GFP_NOIO, pages, bio);
 873         if (!req) {
 874                 up_read(&header->snap_rwsem);
 875                 ret = -ENOMEM;
 876                 goto done_pages;
 877         }
 878
 879         req->r_callback = rbd_cb;
 880
 881         req_data->rq = rq;
 882         req_data->bio = bio;
 883         req_data->pages = pages;
 884         req_data->len = len;
 885
 886         req->r_priv = req_data;
 887
 888         reqhead = req->r_request->front.iov_base;
 889         reqhead->snapid = cpu_to_le64(CEPH_NOSNAP);
 890
 891         strncpy(req->r_oid, obj, sizeof(req->r_oid));
 892         req->r_oid_len = strlen(req->r_oid);
 893
 894         layout = &req->r_file_layout;
 895         memset(layout, 0, sizeof(*layout));
 896         layout->fl_stripe_unit = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
 897         layout->fl_stripe_count = cpu_to_le32(1);
 898         layout->fl_object_size = cpu_to_le32(1 << RBD_MAX_OBJ_ORDER);
 899         layout->fl_pg_preferred = cpu_to_le32(-1);
 900         layout->fl_pg_pool = cpu_to_le32(dev->poolid);
 901         ceph_calc_raw_layout(&dev->client->osdc, layout, snapid,
 902                              ofs, &len, &bno, req, ops);
 903
 904         ceph_osdc_build_request(req, ofs, &len,
 905                                 ops,
 906                                 snapc,
 907                                 &mtime,
 908                                 req->r_oid, req->r_oid_len);
 909         up_read(&header->snap_rwsem);
 910
 911         if (linger_req) {
 912                 ceph_osdc_set_request_linger(&dev->client->osdc, req);
 913                 *linger_req = req;
 914         }
 915
 916         ret = ceph_osdc_start_request(&dev->client->osdc, req, false);
 917         if (ret < 0)
 918                 goto done_err;
 919
 920         if (!rbd_cb) {
 921                 ret = ceph_osdc_wait_request(&dev->client->osdc, req);
 922                 if (ver)
 923                         *ver = le64_to_cpu(req->r_reassert_version.version);
 924                 dout("reassert_ver=%lld\n",
 925                      le64_to_cpu(req->r_reassert_version.version));
 926                 ceph_osdc_put_request(req);
 927         }
 928         return ret;
 929
 930 done_err:
 931         bio_chain_put(req_data->bio);
 932         ceph_osdc_put_request(req);
 933 done_pages:
 934         rbd_coll_end_req(req_data, ret, len);
 935         kfree(req_data);
 936         return ret;
 937 }
 938
 939 /*
 940  * Ceph osd op callback
 941  */
 942 static void rbd_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
 943 {
 944         struct rbd_request *req_data = req->r_priv;
 945         struct ceph_osd_reply_head *replyhead;
 946         struct ceph_osd_op *op;
 947         __s32 rc;
 948         u64 bytes;
 949         int read_op;
 950
 951         /* parse reply */
 952         replyhead = msg->front.iov_base;
 953         WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
 954         op = (void *)(replyhead + 1);
 955         rc = le32_to_cpu(replyhead->result);
 956         bytes = le64_to_cpu(op->extent.length);
 957         read_op = (le32_to_cpu(op->op) == CEPH_OSD_OP_READ);
 958
 959         dout("rbd_req_cb bytes=%lld readop=%d rc=%d\n", bytes, read_op, rc);
 960
 961         if (rc == -ENOENT && read_op) {
 962                 zero_bio_chain(req_data->bio, 0);
 963                 rc = 0;
 964         } else if (rc == 0 && read_op && bytes < req_data->len) {
 965                 zero_bio_chain(req_data->bio, bytes);
 966                 bytes = req_data->len;
 967         }
 968
 969         rbd_coll_end_req(req_data, rc, bytes);
 970
 971         if (req_data->bio)
 972                 bio_chain_put(req_data->bio);
 973
 974         ceph_osdc_put_request(req);
 975         kfree(req_data);
 976 }
 977
 978 static void rbd_simple_req_cb(struct ceph_osd_request *req, struct ceph_msg *msg)
 979 {
 980         ceph_osdc_put_request(req);
 981 }
 982
 983 /*
 984  * Do a synchronous ceph osd operation
 985  */
 986 static int rbd_req_sync_op(struct rbd_device *dev,
 987                            struct ceph_snap_context *snapc,
 988                            u64 snapid,
 989                            int opcode,
 990                            int flags,
 991                            struct ceph_osd_req_op *orig_ops,
 992                            int num_reply,
 993                            const char *obj,
 994                            u64 ofs, u64 len,
 995                            char *buf,
 996                            struct ceph_osd_request **linger_req,
 997                            u64 *ver)
 998 {
 999         int ret;
1000         struct page **pages;
1001         int num_pages;
1002         struct ceph_osd_req_op *ops = orig_ops;
1003         u32 payload_len;
1004
1005         num_pages = calc_pages_for(ofs , len);
1006         pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1007         if (IS_ERR(pages))
1008                 return PTR_ERR(pages);
1009
1010         if (!orig_ops) {
1011                 payload_len = (flags & CEPH_OSD_FLAG_WRITE ? len : 0);
1012                 ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1013                 if (ret < 0)
1014                         goto done;
1015
1016                 if ((flags & CEPH_OSD_FLAG_WRITE) && buf) {
1017                         ret = ceph_copy_to_page_vector(pages, buf, ofs, len);
1018                         if (ret < 0)
1019                                 goto done_ops;
1020                 }
1021         }
1022
1023         ret = rbd_do_request(NULL, dev, snapc, snapid,
1024                           obj, ofs, len, NULL,
1025                           pages, num_pages,
1026                           flags,
1027                           ops,
1028                           2,
1029                           NULL, 0,
1030                           NULL,
1031                           linger_req, ver);
1032         if (ret < 0)
1033                 goto done_ops;
1034
1035         if ((flags & CEPH_OSD_FLAG_READ) && buf)
1036                 ret = ceph_copy_from_page_vector(pages, buf, ofs, ret);
1037
1038 done_ops:
1039         if (!orig_ops)
1040                 rbd_destroy_ops(ops);
1041 done:
1042         ceph_release_page_vector(pages, num_pages);
1043         return ret;
1044 }
1045
1046 /*
1047  * Do an asynchronous ceph osd operation
1048  */
1049 static int rbd_do_op(struct request *rq,
1050                      struct rbd_device *rbd_dev ,
1051                      struct ceph_snap_context *snapc,
1052                      u64 snapid,
1053                      int opcode, int flags, int num_reply,
1054                      u64 ofs, u64 len,
1055                      struct bio *bio,
1056                      struct rbd_req_coll *coll,
1057                      int coll_index)
1058 {
1059         char *seg_name;
1060         u64 seg_ofs;
1061         u64 seg_len;
1062         int ret;
1063         struct ceph_osd_req_op *ops;
1064         u32 payload_len;
1065
1066         seg_name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO);
1067         if (!seg_name)
1068                 return -ENOMEM;
1069
1070         seg_len = rbd_get_segment(&rbd_dev->header,
1071                                   rbd_dev->header.block_name,
1072                                   ofs, len,
1073                                   seg_name, &seg_ofs);
1074
1075         payload_len = (flags & CEPH_OSD_FLAG_WRITE ? seg_len : 0);
1076
1077         ret = rbd_create_rw_ops(&ops, 1, opcode, payload_len);
1078         if (ret < 0)
1079                 goto done;
1080
1081         /* we've taken care of segment sizes earlier when we
1082            cloned the bios. We should never have a segment
1083            truncated at this point */
1084         BUG_ON(seg_len < len);
1085
1086         ret = rbd_do_request(rq, rbd_dev, snapc, snapid,
1087                              seg_name, seg_ofs, seg_len,
1088                              bio,
1089                              NULL, 0,
1090                              flags,
1091                              ops,
1092                              num_reply,
1093                              coll, coll_index,
1094                              rbd_req_cb, 0, NULL);
1095
1096         rbd_destroy_ops(ops);
1097 done:
1098         kfree(seg_name);
1099         return ret;
1100 }
1101
1102 /*
1103  * Request async osd write
1104  */
1105 static int rbd_req_write(struct request *rq,
1106                          struct rbd_device *rbd_dev,
1107                          struct ceph_snap_context *snapc,
1108                          u64 ofs, u64 len,
1109                          struct bio *bio,
1110                          struct rbd_req_coll *coll,
1111                          int coll_index)
1112 {
1113         return rbd_do_op(rq, rbd_dev, snapc, CEPH_NOSNAP,
1114                          CEPH_OSD_OP_WRITE,
1115                          CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1116                          2,
1117                          ofs, len, bio, coll, coll_index);
1118 }
1119
1120 /*
1121  * Request async osd read
1122  */
1123 static int rbd_req_read(struct request *rq,
1124                          struct rbd_device *rbd_dev,
1125                          u64 snapid,
1126                          u64 ofs, u64 len,
1127                          struct bio *bio,
1128                          struct rbd_req_coll *coll,
1129                          int coll_index)
1130 {
1131         return rbd_do_op(rq, rbd_dev, NULL,
1132                          (snapid ? snapid : CEPH_NOSNAP),
1133                          CEPH_OSD_OP_READ,
1134                          CEPH_OSD_FLAG_READ,
1135                          2,
1136                          ofs, len, bio, coll, coll_index);
1137 }
1138
1139 /*
1140  * Request sync osd read
1141  */
1142 static int rbd_req_sync_read(struct rbd_device *dev,
1143                           struct ceph_snap_context *snapc,
1144                           u64 snapid,
1145                           const char *obj,
1146                           u64 ofs, u64 len,
1147                           char *buf,
1148                           u64 *ver)
1149 {
1150         return rbd_req_sync_op(dev, NULL,
1151                                (snapid ? snapid : CEPH_NOSNAP),
1152                                CEPH_OSD_OP_READ,
1153                                CEPH_OSD_FLAG_READ,
1154                                NULL,
1155                                1, obj, ofs, len, buf, NULL, ver);
1156 }
1157
1158 /*
1159  * Request sync osd watch
1160  */
1161 static int rbd_req_sync_notify_ack(struct rbd_device *dev,
1162                                    u64 ver,
1163                                    u64 notify_id,
1164                                    const char *obj)
1165 {
1166         struct ceph_osd_req_op *ops;
1167         struct page **pages = NULL;
1168         int ret;
1169
1170         ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY_ACK, 0);
1171         if (ret < 0)
1172                 return ret;
1173
1174         ops[0].watch.ver = cpu_to_le64(dev->header.obj_version);
1175         ops[0].watch.cookie = notify_id;
1176         ops[0].watch.flag = 0;
1177
1178         ret = rbd_do_request(NULL, dev, NULL, CEPH_NOSNAP,
1179                           obj, 0, 0, NULL,
1180                           pages, 0,
1181                           CEPH_OSD_FLAG_READ,
1182                           ops,
1183                           1,
1184                           NULL, 0,
1185                           rbd_simple_req_cb, 0, NULL);
1186
1187         rbd_destroy_ops(ops);
1188         return ret;
1189 }
1190
1191 static void rbd_watch_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1192 {
1193         struct rbd_device *dev = (struct rbd_device *)data;
1194         if (!dev)
1195                 return;
1196
1197         dout("rbd_watch_cb %s notify_id=%lld opcode=%d\n", dev->obj_md_name,
1198                 notify_id, (int)opcode);
1199         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1200         __rbd_update_snaps(dev);
1201         mutex_unlock(&ctl_mutex);
1202
1203         rbd_req_sync_notify_ack(dev, ver, notify_id, dev->obj_md_name);
1204 }
1205
1206 /*
1207  * Request sync osd watch
1208  */
1209 static int rbd_req_sync_watch(struct rbd_device *dev,
1210                               const char *obj,
1211                               u64 ver)
1212 {
1213         struct ceph_osd_req_op *ops;
1214         struct ceph_osd_client *osdc = &dev->client->osdc;
1215
1216         int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_WATCH, 0);
1217         if (ret < 0)
1218                 return ret;
1219
1220         ret = ceph_osdc_create_event(osdc, rbd_watch_cb, 0,
1221                                      (void *)dev, &dev->watch_event);
1222         if (ret < 0)
1223                 goto fail;
1224
1225         ops[0].watch.ver = cpu_to_le64(ver);
1226         ops[0].watch.cookie = cpu_to_le64(dev->watch_event->cookie);
1227         ops[0].watch.flag = 1;
1228
1229         ret = rbd_req_sync_op(dev, NULL,
1230                               CEPH_NOSNAP,
1231                               0,
1232                               CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1233                               ops,
1234                               1, obj, 0, 0, NULL,
1235                               &dev->watch_request, NULL);
1236
1237         if (ret < 0)
1238                 goto fail_event;
1239
1240         rbd_destroy_ops(ops);
1241         return 0;
1242
1243 fail_event:
1244         ceph_osdc_cancel_event(dev->watch_event);
1245         dev->watch_event = NULL;
1246 fail:
1247         rbd_destroy_ops(ops);
1248         return ret;
1249 }
1250
1251 struct rbd_notify_info {
1252         struct rbd_device *dev;
1253 };
1254
1255 static void rbd_notify_cb(u64 ver, u64 notify_id, u8 opcode, void *data)
1256 {
1257         struct rbd_device *dev = (struct rbd_device *)data;
1258         if (!dev)
1259                 return;
1260
1261         dout("rbd_notify_cb %s notify_id=%lld opcode=%d\n", dev->obj_md_name,
1262                 notify_id, (int)opcode);
1263 }
1264
1265 /*
1266  * Request sync osd notify
1267  */
1268 static int rbd_req_sync_notify(struct rbd_device *dev,
1269                           const char *obj)
1270 {
1271         struct ceph_osd_req_op *ops;
1272         struct ceph_osd_client *osdc = &dev->client->osdc;
1273         struct ceph_osd_event *event;
1274         struct rbd_notify_info info;
1275         int payload_len = sizeof(u32) + sizeof(u32);
1276         int ret;
1277
1278         ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_NOTIFY, payload_len);
1279         if (ret < 0)
1280                 return ret;
1281
1282         info.dev = dev;
1283
1284         ret = ceph_osdc_create_event(osdc, rbd_notify_cb, 1,
1285                                      (void *)&info, &event);
1286         if (ret < 0)
1287                 goto fail;
1288
1289         ops[0].watch.ver = 1;
1290         ops[0].watch.flag = 1;
1291         ops[0].watch.cookie = event->cookie;
1292         ops[0].watch.prot_ver = RADOS_NOTIFY_VER;
1293         ops[0].watch.timeout = 12;
1294
1295         ret = rbd_req_sync_op(dev, NULL,
1296                                CEPH_NOSNAP,
1297                                0,
1298                                CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1299                                ops,
1300                                1, obj, 0, 0, NULL, NULL, NULL);
1301         if (ret < 0)
1302                 goto fail_event;
1303
1304         ret = ceph_osdc_wait_event(event, CEPH_OSD_TIMEOUT_DEFAULT);
1305         dout("ceph_osdc_wait_event returned %d\n", ret);
1306         rbd_destroy_ops(ops);
1307         return 0;
1308
1309 fail_event:
1310         ceph_osdc_cancel_event(event);
1311 fail:
1312         rbd_destroy_ops(ops);
1313         return ret;
1314 }
1315
1316 /*
1317  * Request sync osd rollback
1318  */
1319 static int rbd_req_sync_rollback_obj(struct rbd_device *dev,
1320                                      u64 snapid,
1321                                      const char *obj)
1322 {
1323         struct ceph_osd_req_op *ops;
1324         int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_ROLLBACK, 0);
1325         if (ret < 0)
1326                 return ret;
1327
1328         ops[0].snap.snapid = snapid;
1329
1330         ret = rbd_req_sync_op(dev, NULL,
1331                                CEPH_NOSNAP,
1332                                0,
1333                                CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1334                                ops,
1335                                1, obj, 0, 0, NULL, NULL, NULL);
1336
1337         rbd_destroy_ops(ops);
1338
1339         return ret;
1340 }
1341
1342 /*
1343  * Request sync osd read
1344  */
1345 static int rbd_req_sync_exec(struct rbd_device *dev,
1346                              const char *obj,
1347                              const char *cls,
1348                              const char *method,
1349                              const char *data,
1350                              int len,
1351                              u64 *ver)
1352 {
1353         struct ceph_osd_req_op *ops;
1354         int cls_len = strlen(cls);
1355         int method_len = strlen(method);
1356         int ret = rbd_create_rw_ops(&ops, 1, CEPH_OSD_OP_CALL,
1357                                     cls_len + method_len + len);
1358         if (ret < 0)
1359                 return ret;
1360
1361         ops[0].cls.class_name = cls;
1362         ops[0].cls.class_len = (__u8)cls_len;
1363         ops[0].cls.method_name = method;
1364         ops[0].cls.method_len = (__u8)method_len;
1365         ops[0].cls.argc = 0;
1366         ops[0].cls.indata = data;
1367         ops[0].cls.indata_len = len;
1368
1369         ret = rbd_req_sync_op(dev, NULL,
1370                                CEPH_NOSNAP,
1371                                0,
1372                                CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ONDISK,
1373                                ops,
1374                                1, obj, 0, 0, NULL, NULL, ver);
1375
1376         rbd_destroy_ops(ops);
1377
1378         dout("cls_exec returned %d\n", ret);
1379         return ret;
1380 }
1381
1382 static struct rbd_req_coll *rbd_alloc_coll(int num_reqs)
1383 {
1384         struct rbd_req_coll *coll =
1385                         kzalloc(sizeof(struct rbd_req_coll) +
1386                                 sizeof(struct rbd_req_status) * num_reqs,
1387                                 GFP_ATOMIC);
1388
1389         if (!coll)
1390                 return NULL;
1391         coll->total = num_reqs;
1392         kref_init(&coll->kref);
1393         return coll;
1394 }
1395
1396 /*
1397  * block device queue callback
1398  */
1399 static void rbd_rq_fn(struct request_queue *q)
1400 {
1401         struct rbd_device *rbd_dev = q->queuedata;
1402         struct request *rq;
1403         struct bio_pair *bp = NULL;
1404
1405         rq = blk_fetch_request(q);
1406
1407         while (1) {
1408                 struct bio *bio;
1409                 struct bio *rq_bio, *next_bio = NULL;
1410                 bool do_write;
1411                 int size, op_size = 0;
1412                 u64 ofs;
1413                 int num_segs, cur_seg = 0;
1414                 struct rbd_req_coll *coll;
1415
1416                 /* peek at request from block layer */
1417                 if (!rq)
1418                         break;
1419
1420                 dout("fetched request\n");
1421
1422                 /* filter out block requests we don't understand */
1423                 if ((rq->cmd_type != REQ_TYPE_FS)) {
1424                         __blk_end_request_all(rq, 0);
1425                         goto next;
1426                 }
1427
1428                 /* deduce our operation (read, write) */
1429                 do_write = (rq_data_dir(rq) == WRITE);
1430
1431                 size = blk_rq_bytes(rq);
1432                 ofs = blk_rq_pos(rq) * 512ULL;
1433                 rq_bio = rq->bio;
1434                 if (do_write && rbd_dev->read_only) {
1435                         __blk_end_request_all(rq, -EROFS);
1436                         goto next;
1437                 }
1438
1439                 spin_unlock_irq(q->queue_lock);
1440
1441                 dout("%s 0x%x bytes at 0x%llx\n",
1442                      do_write ? "write" : "read",
1443                      size, blk_rq_pos(rq) * 512ULL);
1444
1445                 num_segs = rbd_get_num_segments(&rbd_dev->header, ofs, size);
1446                 coll = rbd_alloc_coll(num_segs);
1447                 if (!coll) {
1448                         spin_lock_irq(q->queue_lock);
1449                         __blk_end_request_all(rq, -ENOMEM);
1450                         goto next;
1451                 }
1452
1453                 do {
1454                         /* a bio clone to be passed down to OSD req */
1455                         dout("rq->bio->bi_vcnt=%d\n", rq->bio->bi_vcnt);
1456                         op_size = rbd_get_segment(&rbd_dev->header,
1457                                                   rbd_dev->header.block_name,
1458                                                   ofs, size,
1459                                                   NULL, NULL);
1460                         kref_get(&coll->kref);
1461                         bio = bio_chain_clone(&rq_bio, &next_bio, &bp,
1462                                               op_size, GFP_ATOMIC);
1463                         if (!bio) {
1464                                 rbd_coll_end_req_index(rq, coll, cur_seg,
1465                                                        -ENOMEM, op_size);
1466                                 goto next_seg;
1467                         }
1468
1469
1470                         /* init OSD command: write or read */
1471                         if (do_write)
1472                                 rbd_req_write(rq, rbd_dev,
1473                                               rbd_dev->header.snapc,
1474                                               ofs,
1475                                               op_size, bio,
1476                                               coll, cur_seg);
1477                         else
1478                                 rbd_req_read(rq, rbd_dev,
1479                                              cur_snap_id(rbd_dev),
1480                                              ofs,
1481                                              op_size, bio,
1482                                              coll, cur_seg);
1483
1484 next_seg:
1485                         size -= op_size;
1486                         ofs += op_size;
1487
1488                         cur_seg++;
1489                         rq_bio = next_bio;
1490                 } while (size > 0);
1491                 kref_put(&coll->kref, rbd_coll_release);
1492
1493                 if (bp)
1494                         bio_pair_release(bp);
1495                 spin_lock_irq(q->queue_lock);
1496 next:
1497                 rq = blk_fetch_request(q);
1498         }
1499 }
1500
1501 /*
1502  * a queue callback. Makes sure that we don't create a bio that spans across
1503  * multiple osd objects. One exception would be with a single page bios,
1504  * which we handle later at bio_chain_clone
1505  */
1506 static int rbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
1507                           struct bio_vec *bvec)
1508 {
1509         struct rbd_device *rbd_dev = q->queuedata;
1510         unsigned int chunk_sectors = 1 << (rbd_dev->header.obj_order - 9);
1511         sector_t sector = bmd->bi_sector + get_start_sect(bmd->bi_bdev);
1512         unsigned int bio_sectors = bmd->bi_size >> 9;
1513         int max;
1514
1515         max =  (chunk_sectors - ((sector & (chunk_sectors - 1))
1516                                  + bio_sectors)) << 9;
1517         if (max < 0)
1518                 max = 0; /* bio_add cannot handle a negative return */
1519         if (max <= bvec->bv_len && bio_sectors == 0)
1520                 return bvec->bv_len;
1521         return max;
1522 }
1523
1524 static void rbd_free_disk(struct rbd_device *rbd_dev)
1525 {
1526         struct gendisk *disk = rbd_dev->disk;
1527
1528         if (!disk)
1529                 return;
1530
1531         rbd_header_free(&rbd_dev->header);
1532
1533         if (disk->flags & GENHD_FL_UP)
1534                 del_gendisk(disk);
1535         if (disk->queue)
1536                 blk_cleanup_queue(disk->queue);
1537         put_disk(disk);
1538 }
1539
1540 /*
1541  * reload the ondisk the header
1542  */
1543 static int rbd_read_header(struct rbd_device *rbd_dev,
1544                            struct rbd_image_header *header)
1545 {
1546         ssize_t rc;
1547         struct rbd_image_header_ondisk *dh;
1548         int snap_count = 0;
1549         u64 snap_names_len = 0;
1550         u64 ver;
1551
1552         while (1) {
1553                 int len = sizeof(*dh) +
1554                           snap_count * sizeof(struct rbd_image_snap_ondisk) +
1555                           snap_names_len;
1556
1557                 rc = -ENOMEM;
1558                 dh = kmalloc(len, GFP_KERNEL);
1559                 if (!dh)
1560                         return -ENOMEM;
1561
1562                 rc = rbd_req_sync_read(rbd_dev,
1563                                        NULL, CEPH_NOSNAP,
1564                                        rbd_dev->obj_md_name,
1565                                        0, len,
1566                                        (char *)dh, &ver);
1567                 if (rc < 0)
1568                         goto out_dh;
1569
1570                 rc = rbd_header_from_disk(header, dh, snap_count, GFP_KERNEL);
1571                 if (rc < 0)
1572                         goto out_dh;
1573
1574                 if (snap_count != header->total_snaps) {
1575                         snap_count = header->total_snaps;
1576                         snap_names_len = header->snap_names_len;
1577                         rbd_header_free(header);
1578                         kfree(dh);
1579                         continue;
1580                 }
1581                 break;
1582         }
1583         header->obj_version = ver;
1584
1585 out_dh:
1586         kfree(dh);
1587         return rc;
1588 }
1589
1590 /*
1591  * create a snapshot
1592  */
1593 static int rbd_header_add_snap(struct rbd_device *dev,
1594                                const char *snap_name,
1595                                gfp_t gfp_flags)
1596 {
1597         int name_len = strlen(snap_name);
1598         u64 new_snapid;
1599         int ret;
1600         void *data, *data_start, *data_end;
1601         u64 ver;
1602
1603         /* we should create a snapshot only if we're pointing at the head */
1604         if (dev->cur_snap)
1605                 return -EINVAL;
1606
1607         ret = ceph_monc_create_snapid(&dev->client->monc, dev->poolid,
1608                                       &new_snapid);
1609         dout("created snapid=%lld\n", new_snapid);
1610         if (ret < 0)
1611                 return ret;
1612
1613         data = kmalloc(name_len + 16, gfp_flags);
1614         if (!data)
1615                 return -ENOMEM;
1616
1617         data_start = data;
1618         data_end = data + name_len + 16;
1619
1620         ceph_encode_string_safe(&data, data_end, snap_name, name_len, bad);
1621         ceph_encode_64_safe(&data, data_end, new_snapid, bad);
1622
1623         ret = rbd_req_sync_exec(dev, dev->obj_md_name, "rbd", "snap_add",
1624                                 data_start, data - data_start, &ver);
1625
1626         kfree(data_start);
1627
1628         if (ret < 0)
1629                 return ret;
1630
1631         dev->header.snapc->seq =  new_snapid;
1632
1633         return 0;
1634 bad:
1635         return -ERANGE;
1636 }
1637
1638 static void __rbd_remove_all_snaps(struct rbd_device *rbd_dev)
1639 {
1640         struct rbd_snap *snap;
1641
1642         while (!list_empty(&rbd_dev->snaps)) {
1643                 snap = list_first_entry(&rbd_dev->snaps, struct rbd_snap, node);
1644                 __rbd_remove_snap_dev(rbd_dev, snap);
1645         }
1646 }
1647
1648 /*
1649  * only read the first part of the ondisk header, without the snaps info
1650  */
1651 static int __rbd_update_snaps(struct rbd_device *rbd_dev)
1652 {
1653         int ret;
1654         struct rbd_image_header h;
1655         u64 snap_seq;
1656         int follow_seq = 0;
1657
1658         ret = rbd_read_header(rbd_dev, &h);
1659         if (ret < 0)
1660                 return ret;
1661
1662         down_write(&rbd_dev->header.snap_rwsem);
1663
1664         snap_seq = rbd_dev->header.snapc->seq;
1665         if (rbd_dev->header.total_snaps &&
1666             rbd_dev->header.snapc->snaps[0] == snap_seq)
1667                 /* pointing at the head, will need to follow that
1668                    if head moves */
1669                 follow_seq = 1;
1670
1671         kfree(rbd_dev->header.snapc);
1672         kfree(rbd_dev->header.snap_names);
1673         kfree(rbd_dev->header.snap_sizes);
1674
1675         rbd_dev->header.total_snaps = h.total_snaps;
1676         rbd_dev->header.snapc = h.snapc;
1677         rbd_dev->header.snap_names = h.snap_names;
1678         rbd_dev->header.snap_names_len = h.snap_names_len;
1679         rbd_dev->header.snap_sizes = h.snap_sizes;
1680         if (follow_seq)
1681                 rbd_dev->header.snapc->seq = rbd_dev->header.snapc->snaps[0];
1682         else
1683                 rbd_dev->header.snapc->seq = snap_seq;
1684
1685         ret = __rbd_init_snaps_header(rbd_dev);
1686
1687         up_write(&rbd_dev->header.snap_rwsem);
1688
1689         return ret;
1690 }
1691
1692 static int rbd_init_disk(struct rbd_device *rbd_dev)
1693 {
1694         struct gendisk *disk;
1695         struct request_queue *q;
1696         int rc;
1697         u64 total_size = 0;
1698
1699         /* contact OSD, request size info about the object being mapped */
1700         rc = rbd_read_header(rbd_dev, &rbd_dev->header);
1701         if (rc)
1702                 return rc;
1703
1704         /* no need to lock here, as rbd_dev is not registered yet */
1705         rc = __rbd_init_snaps_header(rbd_dev);
1706         if (rc)
1707                 return rc;
1708
1709         rc = rbd_header_set_snap(rbd_dev, rbd_dev->snap_name, &total_size);
1710         if (rc)
1711                 return rc;
1712
1713         /* create gendisk info */
1714         rc = -ENOMEM;
1715         disk = alloc_disk(RBD_MINORS_PER_MAJOR);
1716         if (!disk)
1717                 goto out;
1718
1719         sprintf(disk->disk_name, DRV_NAME "%d", rbd_dev->id);
1720         disk->major = rbd_dev->major;
1721         disk->first_minor = 0;
1722         disk->fops = &rbd_bd_ops;
1723         disk->private_data = rbd_dev;
1724
1725         /* init rq */
1726         rc = -ENOMEM;
1727         q = blk_init_queue(rbd_rq_fn, &rbd_dev->lock);
1728         if (!q)
1729                 goto out_disk;
1730         blk_queue_merge_bvec(q, rbd_merge_bvec);
1731         disk->queue = q;
1732
1733         q->queuedata = rbd_dev;
1734
1735         rbd_dev->disk = disk;
1736         rbd_dev->q = q;
1737
1738         /* finally, announce the disk to the world */
1739         set_capacity(disk, total_size / 512ULL);
1740         add_disk(disk);
1741
1742         pr_info("%s: added with size 0x%llx\n",
1743                 disk->disk_name, (unsigned long long)total_size);
1744         return 0;
1745
1746 out_disk:
1747         put_disk(disk);
1748 out:
1749         return rc;
1750 }
1751
1752 /*
1753   sysfs
1754 */
1755
1756 static ssize_t rbd_size_show(struct device *dev,
1757                              struct device_attribute *attr, char *buf)
1758 {
1759         struct rbd_device *rbd_dev = dev_to_rbd(dev);
1760
1761         return sprintf(buf, "%llu\n", (unsigned long long)rbd_dev->header.image_size);
1762 }
1763
1764 static ssize_t rbd_major_show(struct device *dev,
1765                               struct device_attribute *attr, char *buf)
1766 {
1767         struct rbd_device *rbd_dev = dev_to_rbd(dev);
1768
1769         return sprintf(buf, "%d\n", rbd_dev->major);
1770 }
1771
1772 static ssize_t rbd_client_id_show(struct device *dev,
1773                                   struct device_attribute *attr, char *buf)
1774 {
1775         struct rbd_device *rbd_dev = dev_to_rbd(dev);
1776
1777         return sprintf(buf, "client%lld\n", ceph_client_id(rbd_dev->client));
1778 }
1779
1780 static ssize_t rbd_pool_show(struct device *dev,
1781                              struct device_attribute *attr, char *buf)
1782 {
1783         struct rbd_device *rbd_dev = dev_to_rbd(dev);
1784
1785         return sprintf(buf, "%s\n", rbd_dev->pool_name);
1786 }
1787
1788 static ssize_t rbd_name_show(struct device *dev,
1789                              struct device_attribute *attr, char *buf)
1790 {
1791         struct rbd_device *rbd_dev = dev_to_rbd(dev);
1792
1793         return sprintf(buf, "%s\n", rbd_dev->obj);
1794 }
1795
1796 static ssize_t rbd_snap_show(struct device *dev,
1797                              struct device_attribute *attr,
1798                              char *buf)
1799 {
1800         struct rbd_device *rbd_dev = dev_to_rbd(dev);
1801
1802         return sprintf(buf, "%s\n", rbd_dev->snap_name);
1803 }
1804
1805 static ssize_t rbd_image_refresh(struct device *dev,
1806                                  struct device_attribute *attr,
1807                                  const char *buf,
1808                                  size_t size)
1809 {
1810         struct rbd_device *rbd_dev = dev_to_rbd(dev);
1811         int rc;
1812         int ret = size;
1813
1814         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
1815
1816         rc = __rbd_update_snaps(rbd_dev);
1817         if (rc < 0)
1818                 ret = rc;
1819
1820         mutex_unlock(&ctl_mutex);
1821         return ret;
1822 }
1823
1824 static DEVICE_ATTR(size, S_IRUGO, rbd_size_show, NULL);
1825 static DEVICE_ATTR(major, S_IRUGO, rbd_major_show, NULL);
1826 static DEVICE_ATTR(client_id, S_IRUGO, rbd_client_id_show, NULL);
1827 static DEVICE_ATTR(pool, S_IRUGO, rbd_pool_show, NULL);
1828 static DEVICE_ATTR(name, S_IRUGO, rbd_name_show, NULL);
1829 static DEVICE_ATTR(refresh, S_IWUSR, NULL, rbd_image_refresh);
1830 static DEVICE_ATTR(current_snap, S_IRUGO, rbd_snap_show, NULL);
1831 static DEVICE_ATTR(create_snap, S_IWUSR, NULL, rbd_snap_add);
1832 static DEVICE_ATTR(rollback_snap, S_IWUSR, NULL, rbd_snap_rollback);
1833
1834 static struct attribute *rbd_attrs[] = {
1835         &dev_attr_size.attr,
1836         &dev_attr_major.attr,
1837         &dev_attr_client_id.attr,
1838         &dev_attr_pool.attr,
1839         &dev_attr_name.attr,
1840         &dev_attr_current_snap.attr,
1841         &dev_attr_refresh.attr,
1842         &dev_attr_create_snap.attr,
1843         &dev_attr_rollback_snap.attr,
1844         NULL
1845 };
1846
1847 static struct attribute_group rbd_attr_group = {
1848         .attrs = rbd_attrs,
1849 };
1850
1851 static const struct attribute_group *rbd_attr_groups[] = {
1852         &rbd_attr_group,
1853         NULL
1854 };
1855
1856 static void rbd_sysfs_dev_release(struct device *dev)
1857 {
1858 }
1859
1860 static struct device_type rbd_device_type = {
1861         .name           = "rbd",
1862         .groups         = rbd_attr_groups,
1863         .release        = rbd_sysfs_dev_release,
1864 };
1865
1866
1867 /*
1868   sysfs - snapshots
1869 */
1870
1871 static ssize_t rbd_snap_size_show(struct device *dev,
1872                                   struct device_attribute *attr,
1873                                   char *buf)
1874 {
1875         struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1876
1877         return sprintf(buf, "%lld\n", (long long)snap->size);
1878 }
1879
1880 static ssize_t rbd_snap_id_show(struct device *dev,
1881                                 struct device_attribute *attr,
1882                                 char *buf)
1883 {
1884         struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1885
1886         return sprintf(buf, "%lld\n", (long long)snap->id);
1887 }
1888
1889 static DEVICE_ATTR(snap_size, S_IRUGO, rbd_snap_size_show, NULL);
1890 static DEVICE_ATTR(snap_id, S_IRUGO, rbd_snap_id_show, NULL);
1891
1892 static struct attribute *rbd_snap_attrs[] = {
1893         &dev_attr_snap_size.attr,
1894         &dev_attr_snap_id.attr,
1895         NULL,
1896 };
1897
1898 static struct attribute_group rbd_snap_attr_group = {
1899         .attrs = rbd_snap_attrs,
1900 };
1901
1902 static void rbd_snap_dev_release(struct device *dev)
1903 {
1904         struct rbd_snap *snap = container_of(dev, struct rbd_snap, dev);
1905         kfree(snap->name);
1906         kfree(snap);
1907 }
1908
1909 static const struct attribute_group *rbd_snap_attr_groups[] = {
1910         &rbd_snap_attr_group,
1911         NULL
1912 };
1913
1914 static struct device_type rbd_snap_device_type = {
1915         .groups         = rbd_snap_attr_groups,
1916         .release        = rbd_snap_dev_release,
1917 };
1918
1919 static void __rbd_remove_snap_dev(struct rbd_device *rbd_dev,
1920                                   struct rbd_snap *snap)
1921 {
1922         list_del(&snap->node);
1923         device_unregister(&snap->dev);
1924 }
1925
1926 static int rbd_register_snap_dev(struct rbd_device *rbd_dev,
1927                                   struct rbd_snap *snap,
1928                                   struct device *parent)
1929 {
1930         struct device *dev = &snap->dev;
1931         int ret;
1932
1933         dev->type = &rbd_snap_device_type;
1934         dev->parent = parent;
1935         dev->release = rbd_snap_dev_release;
1936         dev_set_name(dev, "snap_%s", snap->name);
1937         ret = device_register(dev);
1938
1939         return ret;
1940 }
1941
1942 static int __rbd_add_snap_dev(struct rbd_device *rbd_dev,
1943                               int i, const char *name,
1944                               struct rbd_snap **snapp)
1945 {
1946         int ret;
1947         struct rbd_snap *snap = kzalloc(sizeof(*snap), GFP_KERNEL);
1948         if (!snap)
1949                 return -ENOMEM;
1950         snap->name = kstrdup(name, GFP_KERNEL);
1951         snap->size = rbd_dev->header.snap_sizes[i];
1952         snap->id = rbd_dev->header.snapc->snaps[i];
1953         if (device_is_registered(&rbd_dev->dev)) {
1954                 ret = rbd_register_snap_dev(rbd_dev, snap,
1955                                              &rbd_dev->dev);
1956                 if (ret < 0)
1957                         goto err;
1958         }
1959         *snapp = snap;
1960         return 0;
1961 err:
1962         kfree(snap->name);
1963         kfree(snap);
1964         return ret;
1965 }
1966
1967 /*
1968  * search for the previous snap in a null delimited string list
1969  */
1970 const char *rbd_prev_snap_name(const char *name, const char *start)
1971 {
1972         if (name < start + 2)
1973                 return NULL;
1974
1975         name -= 2;
1976         while (*name) {
1977                 if (name == start)
1978                         return start;
1979                 name--;
1980         }
1981         return name + 1;
1982 }
1983
1984 /*
1985  * compare the old list of snapshots that we have to what's in the header
1986  * and update it accordingly. Note that the header holds the snapshots
1987  * in a reverse order (from newest to oldest) and we need to go from
1988  * older to new so that we don't get a duplicate snap name when
1989  * doing the process (e.g., removed snapshot and recreated a new
1990  * one with the same name.
1991  */
1992 static int __rbd_init_snaps_header(struct rbd_device *rbd_dev)
1993 {
1994         const char *name, *first_name;
1995         int i = rbd_dev->header.total_snaps;
1996         struct rbd_snap *snap, *old_snap = NULL;
1997         int ret;
1998         struct list_head *p, *n;
1999
2000         first_name = rbd_dev->header.snap_names;
2001         name = first_name + rbd_dev->header.snap_names_len;
2002
2003         list_for_each_prev_safe(p, n, &rbd_dev->snaps) {
2004                 u64 cur_id;
2005
2006                 old_snap = list_entry(p, struct rbd_snap, node);
2007
2008                 if (i)
2009                         cur_id = rbd_dev->header.snapc->snaps[i - 1];
2010
2011                 if (!i || old_snap->id < cur_id) {
2012                         /* old_snap->id was skipped, thus was removed */
2013                         __rbd_remove_snap_dev(rbd_dev, old_snap);
2014                         continue;
2015                 }
2016                 if (old_snap->id == cur_id) {
2017                         /* we have this snapshot already */
2018                         i--;
2019                         name = rbd_prev_snap_name(name, first_name);
2020                         continue;
2021                 }
2022                 for (; i > 0;
2023                      i--, name = rbd_prev_snap_name(name, first_name)) {
2024                         if (!name) {
2025                                 WARN_ON(1);
2026                                 return -EINVAL;
2027                         }
2028                         cur_id = rbd_dev->header.snapc->snaps[i];
2029                         /* snapshot removal? handle it above */
2030                         if (cur_id >= old_snap->id)
2031                                 break;
2032                         /* a new snapshot */
2033                         ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2034                         if (ret < 0)
2035                                 return ret;
2036
2037                         /* note that we add it backward so using n and not p */
2038                         list_add(&snap->node, n);
2039                         p = &snap->node;
2040                 }
2041         }
2042         /* we're done going over the old snap list, just add what's left */
2043         for (; i > 0; i--) {
2044                 name = rbd_prev_snap_name(name, first_name);
2045                 if (!name) {
2046                         WARN_ON(1);
2047                         return -EINVAL;
2048                 }
2049                 ret = __rbd_add_snap_dev(rbd_dev, i - 1, name, &snap);
2050                 if (ret < 0)
2051                         return ret;
2052                 list_add(&snap->node, &rbd_dev->snaps);
2053         }
2054
2055         return 0;
2056 }
2057
2058
2059 static void rbd_root_dev_release(struct device *dev)
2060 {
2061 }
2062
2063 static struct device rbd_root_dev = {
2064         .init_name =    "rbd",
2065         .release =      rbd_root_dev_release,
2066 };
2067
2068 static int rbd_bus_add_dev(struct rbd_device *rbd_dev)
2069 {
2070         int ret = -ENOMEM;
2071         struct device *dev;
2072         struct rbd_snap *snap;
2073
2074         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2075         dev = &rbd_dev->dev;
2076
2077         dev->bus = &rbd_bus_type;
2078         dev->type = &rbd_device_type;
2079         dev->parent = &rbd_root_dev;
2080         dev->release = rbd_dev_release;
2081         dev_set_name(dev, "%d", rbd_dev->id);
2082         ret = device_register(dev);
2083         if (ret < 0)
2084                 goto done_free;
2085
2086         list_for_each_entry(snap, &rbd_dev->snaps, node) {
2087                 ret = rbd_register_snap_dev(rbd_dev, snap,
2088                                              &rbd_dev->dev);
2089                 if (ret < 0)
2090                         break;
2091         }
2092
2093         mutex_unlock(&ctl_mutex);
2094         return 0;
2095 done_free:
2096         mutex_unlock(&ctl_mutex);
2097         return ret;
2098 }
2099
2100 static void rbd_bus_del_dev(struct rbd_device *rbd_dev)
2101 {
2102         device_unregister(&rbd_dev->dev);
2103 }
2104
2105 static int rbd_init_watch_dev(struct rbd_device *rbd_dev)
2106 {
2107         int ret, rc;
2108
2109         do {
2110                 ret = rbd_req_sync_watch(rbd_dev, rbd_dev->obj_md_name,
2111                                          rbd_dev->header.obj_version);
2112                 if (ret == -ERANGE) {
2113                         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2114                         rc = __rbd_update_snaps(rbd_dev);
2115                         mutex_unlock(&ctl_mutex);
2116                         if (rc < 0)
2117                                 return rc;
2118                 }
2119         } while (ret == -ERANGE);
2120
2121         return ret;
2122 }
2123
2124 static ssize_t rbd_add(struct bus_type *bus,
2125                        const char *buf,
2126                        size_t count)
2127 {
2128         struct ceph_osd_client *osdc;
2129         struct rbd_device *rbd_dev;
2130         ssize_t rc = -ENOMEM;
2131         int irc, new_id = 0;
2132         struct list_head *tmp;
2133         char *mon_dev_name;
2134         char *options;
2135
2136         if (!try_module_get(THIS_MODULE))
2137                 return -ENODEV;
2138
2139         mon_dev_name = kmalloc(RBD_MAX_OPT_LEN, GFP_KERNEL);
2140         if (!mon_dev_name)
2141                 goto err_out_mod;
2142
2143         options = kmalloc(RBD_MAX_OPT_LEN, GFP_KERNEL);
2144         if (!options)
2145                 goto err_mon_dev;
2146
2147         /* new rbd_device object */
2148         rbd_dev = kzalloc(sizeof(*rbd_dev), GFP_KERNEL);
2149         if (!rbd_dev)
2150                 goto err_out_opt;
2151
2152         /* static rbd_device initialization */
2153         spin_lock_init(&rbd_dev->lock);
2154         INIT_LIST_HEAD(&rbd_dev->node);
2155         INIT_LIST_HEAD(&rbd_dev->snaps);
2156
2157         /* generate unique id: find highest unique id, add one */
2158         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2159
2160         list_for_each(tmp, &rbd_dev_list) {
2161                 struct rbd_device *rbd_dev;
2162
2163                 rbd_dev = list_entry(tmp, struct rbd_device, node);
2164                 if (rbd_dev->id >= new_id)
2165                         new_id = rbd_dev->id + 1;
2166         }
2167
2168         rbd_dev->id = new_id;
2169
2170         /* add to global list */
2171         list_add_tail(&rbd_dev->node, &rbd_dev_list);
2172
2173         /* parse add command */
2174         if (sscanf(buf, "%" __stringify(RBD_MAX_OPT_LEN) "s "
2175                    "%" __stringify(RBD_MAX_OPT_LEN) "s "
2176                    "%" __stringify(RBD_MAX_POOL_NAME_LEN) "s "
2177                    "%" __stringify(RBD_MAX_OBJ_NAME_LEN) "s"
2178                    "%" __stringify(RBD_MAX_SNAP_NAME_LEN) "s",
2179                    mon_dev_name, options, rbd_dev->pool_name,
2180                    rbd_dev->obj, rbd_dev->snap_name) < 4) {
2181                 rc = -EINVAL;
2182                 goto err_out_slot;
2183         }
2184
2185         if (rbd_dev->snap_name[0] == 0)
2186                 rbd_dev->snap_name[0] = '-';
2187
2188         rbd_dev->obj_len = strlen(rbd_dev->obj);
2189         snprintf(rbd_dev->obj_md_name, sizeof(rbd_dev->obj_md_name), "%s%s",
2190                  rbd_dev->obj, RBD_SUFFIX);
2191
2192         /* initialize rest of new object */
2193         snprintf(rbd_dev->name, DEV_NAME_LEN, DRV_NAME "%d", rbd_dev->id);
2194         rc = rbd_get_client(rbd_dev, mon_dev_name, options);
2195         if (rc < 0)
2196                 goto err_out_slot;
2197
2198         mutex_unlock(&ctl_mutex);
2199
2200         /* pick the pool */
2201         osdc = &rbd_dev->client->osdc;
2202         rc = ceph_pg_poolid_by_name(osdc->osdmap, rbd_dev->pool_name);
2203         if (rc < 0)
2204                 goto err_out_client;
2205         rbd_dev->poolid = rc;
2206
2207         /* register our block device */
2208         irc = register_blkdev(0, rbd_dev->name);
2209         if (irc < 0) {
2210                 rc = irc;
2211                 goto err_out_client;
2212         }
2213         rbd_dev->major = irc;
2214
2215         rc = rbd_bus_add_dev(rbd_dev);
2216         if (rc)
2217                 goto err_out_blkdev;
2218
2219         /* set up and announce blkdev mapping */
2220         rc = rbd_init_disk(rbd_dev);
2221         if (rc)
2222                 goto err_out_bus;
2223
2224         rc = rbd_init_watch_dev(rbd_dev);
2225         if (rc)
2226                 goto err_out_bus;
2227
2228         return count;
2229
2230 err_out_bus:
2231         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2232         list_del_init(&rbd_dev->node);
2233         mutex_unlock(&ctl_mutex);
2234
2235         /* this will also clean up rest of rbd_dev stuff */
2236
2237         rbd_bus_del_dev(rbd_dev);
2238         kfree(options);
2239         kfree(mon_dev_name);
2240         return rc;
2241
2242 err_out_blkdev:
2243         unregister_blkdev(rbd_dev->major, rbd_dev->name);
2244 err_out_client:
2245         rbd_put_client(rbd_dev);
2246         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2247 err_out_slot:
2248         list_del_init(&rbd_dev->node);
2249         mutex_unlock(&ctl_mutex);
2250
2251         kfree(rbd_dev);
2252 err_out_opt:
2253         kfree(options);
2254 err_mon_dev:
2255         kfree(mon_dev_name);
2256 err_out_mod:
2257         dout("Error adding device %s\n", buf);
2258         module_put(THIS_MODULE);
2259         return rc;
2260 }
2261
2262 static struct rbd_device *__rbd_get_dev(unsigned long id)
2263 {
2264         struct list_head *tmp;
2265         struct rbd_device *rbd_dev;
2266
2267         list_for_each(tmp, &rbd_dev_list) {
2268                 rbd_dev = list_entry(tmp, struct rbd_device, node);
2269                 if (rbd_dev->id == id)
2270                         return rbd_dev;
2271         }
2272         return NULL;
2273 }
2274
2275 static void rbd_dev_release(struct device *dev)
2276 {
2277         struct rbd_device *rbd_dev =
2278                         container_of(dev, struct rbd_device, dev);
2279
2280         if (rbd_dev->watch_request)
2281                 ceph_osdc_unregister_linger_request(&rbd_dev->client->osdc,
2282                                                     rbd_dev->watch_request);
2283         if (rbd_dev->watch_event)
2284                 ceph_osdc_cancel_event(rbd_dev->watch_event);
2285
2286         rbd_put_client(rbd_dev);
2287
2288         /* clean up and free blkdev */
2289         rbd_free_disk(rbd_dev);
2290         unregister_blkdev(rbd_dev->major, rbd_dev->name);
2291         kfree(rbd_dev);
2292
2293         /* release module ref */
2294         module_put(THIS_MODULE);
2295 }
2296
2297 static ssize_t rbd_remove(struct bus_type *bus,
2298                           const char *buf,
2299                           size_t count)
2300 {
2301         struct rbd_device *rbd_dev = NULL;
2302         int target_id, rc;
2303         unsigned long ul;
2304         int ret = count;
2305
2306         rc = strict_strtoul(buf, 10, &ul);
2307         if (rc)
2308                 return rc;
2309
2310         /* convert to int; abort if we lost anything in the conversion */
2311         target_id = (int) ul;
2312         if (target_id != ul)
2313                 return -EINVAL;
2314
2315         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2316
2317         rbd_dev = __rbd_get_dev(target_id);
2318         if (!rbd_dev) {
2319                 ret = -ENOENT;
2320                 goto done;
2321         }
2322
2323         list_del_init(&rbd_dev->node);
2324
2325         __rbd_remove_all_snaps(rbd_dev);
2326         rbd_bus_del_dev(rbd_dev);
2327
2328 done:
2329         mutex_unlock(&ctl_mutex);
2330         return ret;
2331 }
2332
2333 static ssize_t rbd_snap_add(struct device *dev,
2334                             struct device_attribute *attr,
2335                             const char *buf,
2336                             size_t count)
2337 {
2338         struct rbd_device *rbd_dev = dev_to_rbd(dev);
2339         int ret;
2340         char *name = kmalloc(count + 1, GFP_KERNEL);
2341         if (!name)
2342                 return -ENOMEM;
2343
2344         snprintf(name, count, "%s", buf);
2345
2346         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2347
2348         ret = rbd_header_add_snap(rbd_dev,
2349                                   name, GFP_KERNEL);
2350         if (ret < 0)
2351                 goto err_unlock;
2352
2353         ret = __rbd_update_snaps(rbd_dev);
2354         if (ret < 0)
2355                 goto err_unlock;
2356
2357         /* shouldn't hold ctl_mutex when notifying.. notify might
2358            trigger a watch callback that would need to get that mutex */
2359         mutex_unlock(&ctl_mutex);
2360
2361         /* make a best effort, don't error if failed */
2362         rbd_req_sync_notify(rbd_dev, rbd_dev->obj_md_name);
2363
2364         ret = count;
2365         kfree(name);
2366         return ret;
2367
2368 err_unlock:
2369         mutex_unlock(&ctl_mutex);
2370         kfree(name);
2371         return ret;
2372 }
2373
2374 static ssize_t rbd_snap_rollback(struct device *dev,
2375                                  struct device_attribute *attr,
2376                                  const char *buf,
2377                                  size_t count)
2378 {
2379         struct rbd_device *rbd_dev = dev_to_rbd(dev);
2380         int ret;
2381         u64 snapid;
2382         u64 cur_ofs;
2383         char *seg_name = NULL;
2384         char *snap_name = kmalloc(count + 1, GFP_KERNEL);
2385         ret = -ENOMEM;
2386         if (!snap_name)
2387                 return ret;
2388
2389         /* parse snaps add command */
2390         snprintf(snap_name, count, "%s", buf);
2391         seg_name = kmalloc(RBD_MAX_SEG_NAME_LEN + 1, GFP_NOIO);
2392         if (!seg_name)
2393                 goto done;
2394
2395         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2396
2397         ret = snap_by_name(&rbd_dev->header, snap_name, &snapid, NULL);
2398         if (ret < 0)
2399                 goto done_unlock;
2400
2401         dout("snapid=%lld\n", snapid);
2402
2403         cur_ofs = 0;
2404         while (cur_ofs < rbd_dev->header.image_size) {
2405                 cur_ofs += rbd_get_segment(&rbd_dev->header,
2406                                            rbd_dev->obj,
2407                                            cur_ofs, (u64)-1,
2408                                            seg_name, NULL);
2409                 dout("seg_name=%s\n", seg_name);
2410
2411                 ret = rbd_req_sync_rollback_obj(rbd_dev, snapid, seg_name);
2412                 if (ret < 0)
2413                         pr_warning("could not roll back obj %s err=%d\n",
2414                                    seg_name, ret);
2415         }
2416
2417         ret = __rbd_update_snaps(rbd_dev);
2418         if (ret < 0)
2419                 goto done_unlock;
2420
2421         ret = count;
2422
2423 done_unlock:
2424         mutex_unlock(&ctl_mutex);
2425 done:
2426         kfree(seg_name);
2427         kfree(snap_name);
2428
2429         return ret;
2430 }
2431
2432 static struct bus_attribute rbd_bus_attrs[] = {
2433         __ATTR(add, S_IWUSR, NULL, rbd_add),
2434         __ATTR(remove, S_IWUSR, NULL, rbd_remove),
2435         __ATTR_NULL
2436 };
2437
2438 /*
2439  * create control files in sysfs
2440  * /sys/bus/rbd/...
2441  */
2442 static int rbd_sysfs_init(void)
2443 {
2444         int ret;
2445
2446         rbd_bus_type.bus_attrs = rbd_bus_attrs;
2447
2448         ret = bus_register(&rbd_bus_type);
2449          if (ret < 0)
2450                 return ret;
2451
2452         ret = device_register(&rbd_root_dev);
2453
2454         return ret;
2455 }
2456
2457 static void rbd_sysfs_cleanup(void)
2458 {
2459         device_unregister(&rbd_root_dev);
2460         bus_unregister(&rbd_bus_type);
2461 }
2462
2463 int __init rbd_init(void)
2464 {
2465         int rc;
2466
2467         rc = rbd_sysfs_init();
2468         if (rc)
2469                 return rc;
2470         spin_lock_init(&node_lock);
2471         pr_info("loaded " DRV_NAME_LONG "\n");
2472         return 0;
2473 }
2474
2475 void __exit rbd_exit(void)
2476 {
2477         rbd_sysfs_cleanup();
2478 }
2479
2480 module_init(rbd_init);
2481 module_exit(rbd_exit);
2482
2483 MODULE_AUTHOR("Sage Weil <sage@newdream.net>");
2484 MODULE_AUTHOR("Yehuda Sadeh <yehuda@hq.newdream.net>");
2485 MODULE_DESCRIPTION("rados block device");
2486
2487 /* following authorship retained from original osdblk.c */
2488 MODULE_AUTHOR("Jeff Garzik <jeff@garzik.org>");
2489
2490 MODULE_LICENSE("GPL");