fs/ceph/locks.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/file.h>
   4 #include <linux/namei.h>
   5 #include <linux/random.h>
   6
   7 #include "super.h"
   8 #include "mds_client.h"
   9 #include <linux/ceph/pagelist.h>
  10
  11 static u64 lock_secret;
  12 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
  13                                          struct ceph_mds_request *req);
  14
  15 static inline u64 secure_addr(void *addr)
  16 {
  17         u64 v = lock_secret ^ (u64)(unsigned long)addr;
  18         /*
  19          * Set the most significant bit, so that MDS knows the 'owner'
  20          * is sufficient to identify the owner of lock. (old code uses
  21          * both 'owner' and 'pid')
  22          */
  23         v |= (1ULL << 63);
  24         return v;
  25 }
  26
  27 void __init ceph_flock_init(void)
  28 {
  29         get_random_bytes(&lock_secret, sizeof(lock_secret));
  30 }
  31
  32 /**
  33  * Implement fcntl and flock locking functions.
  34  */
  35 static int ceph_lock_message(u8 lock_type, u16 operation, struct file *file,
  36                              int cmd, u8 wait, struct file_lock *fl)
  37 {
  38         struct inode *inode = file_inode(file);
  39         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
  40         struct ceph_mds_request *req;
  41         int err;
  42         u64 length = 0;
  43         u64 owner;
  44
  45         if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK)
  46                 wait = 0;
  47
  48         req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
  49         if (IS_ERR(req))
  50                 return PTR_ERR(req);
  51         req->r_inode = inode;
  52         ihold(inode);
  53         req->r_num_caps = 1;
  54
  55         /* mds requires start and length rather than start and end */
  56         if (LLONG_MAX == fl->fl_end)
  57                 length = 0;
  58         else
  59                 length = fl->fl_end - fl->fl_start + 1;
  60
  61         owner = secure_addr(fl->fl_owner);
  62
  63         dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
  64              "start: %llu, length: %llu, wait: %d, type: %d", (int)lock_type,
  65              (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
  66              wait, fl->fl_type);
  67
  68         req->r_args.filelock_change.rule = lock_type;
  69         req->r_args.filelock_change.type = cmd;
  70         req->r_args.filelock_change.owner = cpu_to_le64(owner);
  71         req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
  72         req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
  73         req->r_args.filelock_change.length = cpu_to_le64(length);
  74         req->r_args.filelock_change.wait = wait;
  75
  76         if (wait)
  77                 req->r_wait_for_completion = ceph_lock_wait_for_completion;
  78
  79         err = ceph_mdsc_do_request(mdsc, inode, req);
  80
  81         if (operation == CEPH_MDS_OP_GETFILELOCK) {
  82                 fl->fl_pid = -le64_to_cpu(req->r_reply_info.filelock_reply->pid);
  83                 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
  84                         fl->fl_type = F_RDLCK;
  85                 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
  86                         fl->fl_type = F_WRLCK;
  87                 else
  88                         fl->fl_type = F_UNLCK;
  89
  90                 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
  91                 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
  92                                                  le64_to_cpu(req->r_reply_info.filelock_reply->length);
  93                 if (length >= 1)
  94                         fl->fl_end = length -1;
  95                 else
  96                         fl->fl_end = 0;
  97
  98         }
  99         ceph_mdsc_put_request(req);
 100         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
 101              "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
 102              (int)operation, (u64)fl->fl_pid, fl->fl_start,
 103              length, wait, fl->fl_type, err);
 104         return err;
 105 }
 106
 107 static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
 108                                          struct ceph_mds_request *req)
 109 {
 110         struct ceph_mds_request *intr_req;
 111         struct inode *inode = req->r_inode;
 112         int err, lock_type;
 113
 114         BUG_ON(req->r_op != CEPH_MDS_OP_SETFILELOCK);
 115         if (req->r_args.filelock_change.rule == CEPH_LOCK_FCNTL)
 116                 lock_type = CEPH_LOCK_FCNTL_INTR;
 117         else if (req->r_args.filelock_change.rule == CEPH_LOCK_FLOCK)
 118                 lock_type = CEPH_LOCK_FLOCK_INTR;
 119         else
 120                 BUG_ON(1);
 121         BUG_ON(req->r_args.filelock_change.type == CEPH_LOCK_UNLOCK);
 122
 123         err = wait_for_completion_interruptible(&req->r_completion);
 124         if (!err)
 125                 return 0;
 126
 127         dout("ceph_lock_wait_for_completion: request %llu was interrupted\n",
 128              req->r_tid);
 129
 130         mutex_lock(&mdsc->mutex);
 131         if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
 132                 err = 0;
 133         } else {
 134                 /*
 135                  * ensure we aren't running concurrently with
 136                  * ceph_fill_trace or ceph_readdir_prepopulate, which
 137                  * rely on locks (dir mutex) held by our caller.
 138                  */
 139                 mutex_lock(&req->r_fill_mutex);
 140                 req->r_err = err;
 141                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
 142                 mutex_unlock(&req->r_fill_mutex);
 143
 144                 if (!req->r_session) {
 145                         // haven't sent the request
 146                         err = 0;
 147                 }
 148         }
 149         mutex_unlock(&mdsc->mutex);
 150         if (!err)
 151                 return 0;
 152
 153         intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK,
 154                                             USE_AUTH_MDS);
 155         if (IS_ERR(intr_req))
 156                 return PTR_ERR(intr_req);
 157
 158         intr_req->r_inode = inode;
 159         ihold(inode);
 160         intr_req->r_num_caps = 1;
 161
 162         intr_req->r_args.filelock_change = req->r_args.filelock_change;
 163         intr_req->r_args.filelock_change.rule = lock_type;
 164         intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK;
 165
 166         err = ceph_mdsc_do_request(mdsc, inode, intr_req);
 167         ceph_mdsc_put_request(intr_req);
 168
 169         if (err && err != -ERESTARTSYS)
 170                 return err;
 171
 172         wait_for_completion_killable(&req->r_safe_completion);
 173         return 0;
 174 }
 175
 176 /**
 177  * Attempt to set an fcntl lock.
 178  * For now, this just goes away to the server. Later it may be more awesome.
 179  */
 180 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 181 {
 182         u8 lock_cmd;
 183         int err;
 184         u8 wait = 0;
 185         u16 op = CEPH_MDS_OP_SETFILELOCK;
 186
 187         if (!(fl->fl_flags & FL_POSIX))
 188                 return -ENOLCK;
 189         /* No mandatory locks */
 190         if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
 191                 return -ENOLCK;
 192
 193         dout("ceph_lock, fl_owner: %p", fl->fl_owner);
 194
 195         /* set wait bit as appropriate, then make command as Ceph expects it*/
 196         if (IS_GETLK(cmd))
 197                 op = CEPH_MDS_OP_GETFILELOCK;
 198         else if (IS_SETLKW(cmd))
 199                 wait = 1;
 200
 201         if (F_RDLCK == fl->fl_type)
 202                 lock_cmd = CEPH_LOCK_SHARED;
 203         else if (F_WRLCK == fl->fl_type)
 204                 lock_cmd = CEPH_LOCK_EXCL;
 205         else
 206                 lock_cmd = CEPH_LOCK_UNLOCK;
 207
 208         err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
 209         if (!err) {
 210                 if (op != CEPH_MDS_OP_GETFILELOCK) {
 211                         dout("mds locked, locking locally");
 212                         err = posix_lock_file(file, fl, NULL);
 213                         if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
 214                                 /* undo! This should only happen if
 215                                  * the kernel detects local
 216                                  * deadlock. */
 217                                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 218                                                   CEPH_LOCK_UNLOCK, 0, fl);
 219                                 dout("got %d on posix_lock_file, undid lock",
 220                                      err);
 221                         }
 222                 }
 223         }
 224         return err;
 225 }
 226
 227 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 228 {
 229         u8 lock_cmd;
 230         int err;
 231         u8 wait = 0;
 232
 233         if (!(fl->fl_flags & FL_FLOCK))
 234                 return -ENOLCK;
 235         /* No mandatory locks */
 236         if (fl->fl_type & LOCK_MAND)
 237                 return -EOPNOTSUPP;
 238
 239         dout("ceph_flock, fl_file: %p", fl->fl_file);
 240
 241         if (IS_SETLKW(cmd))
 242                 wait = 1;
 243
 244         if (F_RDLCK == fl->fl_type)
 245                 lock_cmd = CEPH_LOCK_SHARED;
 246         else if (F_WRLCK == fl->fl_type)
 247                 lock_cmd = CEPH_LOCK_EXCL;
 248         else
 249                 lock_cmd = CEPH_LOCK_UNLOCK;
 250
 251         err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
 252                                 file, lock_cmd, wait, fl);
 253         if (!err) {
 254                 err = locks_lock_file_wait(file, fl);
 255                 if (err) {
 256                         ceph_lock_message(CEPH_LOCK_FLOCK,
 257                                           CEPH_MDS_OP_SETFILELOCK,
 258                                           file, CEPH_LOCK_UNLOCK, 0, fl);
 259                         dout("got %d on locks_lock_file_wait, undid lock", err);
 260                 }
 261         }
 262         return err;
 263 }
 264
 265 /*
 266  * Fills in the passed counter variables, so you can prepare pagelist metadata
 267  * before calling ceph_encode_locks.
 268  */
 269 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
 270 {
 271         struct file_lock *lock;
 272         struct file_lock_context *ctx;
 273
 274         *fcntl_count = 0;
 275         *flock_count = 0;
 276
 277         ctx = inode->i_flctx;
 278         if (ctx) {
 279                 spin_lock(&ctx->flc_lock);
 280                 list_for_each_entry(lock, &ctx->flc_posix, fl_list)
 281                         ++(*fcntl_count);
 282                 list_for_each_entry(lock, &ctx->flc_flock, fl_list)
 283                         ++(*flock_count);
 284                 spin_unlock(&ctx->flc_lock);
 285         }
 286         dout("counted %d flock locks and %d fcntl locks",
 287              *flock_count, *fcntl_count);
 288 }
 289
 290 /**
 291  * Encode the flock and fcntl locks for the given inode into the ceph_filelock
 292  * array. Must be called with inode->i_lock already held.
 293  * If we encounter more of a specific lock type than expected, return -ENOSPC.
 294  */
 295 int ceph_encode_locks_to_buffer(struct inode *inode,
 296                                 struct ceph_filelock *flocks,
 297                                 int num_fcntl_locks, int num_flock_locks)
 298 {
 299         struct file_lock *lock;
 300         struct file_lock_context *ctx = inode->i_flctx;
 301         int err = 0;
 302         int seen_fcntl = 0;
 303         int seen_flock = 0;
 304         int l = 0;
 305
 306         dout("encoding %d flock and %d fcntl locks", num_flock_locks,
 307              num_fcntl_locks);
 308
 309         if (!ctx)
 310                 return 0;
 311
 312         spin_lock(&ctx->flc_lock);
 313         list_for_each_entry(lock, &ctx->flc_posix, fl_list) {
 314                 ++seen_fcntl;
 315                 if (seen_fcntl > num_fcntl_locks) {
 316                         err = -ENOSPC;
 317                         goto fail;
 318                 }
 319                 err = lock_to_ceph_filelock(lock, &flocks[l]);
 320                 if (err)
 321                         goto fail;
 322                 ++l;
 323         }
 324         list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
 325                 ++seen_flock;
 326                 if (seen_flock > num_flock_locks) {
 327                         err = -ENOSPC;
 328                         goto fail;
 329                 }
 330                 err = lock_to_ceph_filelock(lock, &flocks[l]);
 331                 if (err)
 332                         goto fail;
 333                 ++l;
 334         }
 335 fail:
 336         spin_unlock(&ctx->flc_lock);
 337         return err;
 338 }
 339
 340 /**
 341  * Copy the encoded flock and fcntl locks into the pagelist.
 342  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
 343  * sequential flock locks.
 344  * Returns zero on success.
 345  */
 346 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
 347                            struct ceph_pagelist *pagelist,
 348                            int num_fcntl_locks, int num_flock_locks)
 349 {
 350         int err = 0;
 351         __le32 nlocks;
 352
 353         nlocks = cpu_to_le32(num_fcntl_locks);
 354         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 355         if (err)
 356                 goto out_fail;
 357
 358         err = ceph_pagelist_append(pagelist, flocks,
 359                                    num_fcntl_locks * sizeof(*flocks));
 360         if (err)
 361                 goto out_fail;
 362
 363         nlocks = cpu_to_le32(num_flock_locks);
 364         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 365         if (err)
 366                 goto out_fail;
 367
 368         err = ceph_pagelist_append(pagelist,
 369                                    &flocks[num_fcntl_locks],
 370                                    num_flock_locks * sizeof(*flocks));
 371 out_fail:
 372         return err;
 373 }
 374
 375 /*
 376  * Given a pointer to a lock, convert it to a ceph filelock
 377  */
 378 int lock_to_ceph_filelock(struct file_lock *lock,
 379                           struct ceph_filelock *cephlock)
 380 {
 381         int err = 0;
 382         cephlock->start = cpu_to_le64(lock->fl_start);
 383         cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
 384         cephlock->client = cpu_to_le64(0);
 385         cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
 386         cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
 387
 388         switch (lock->fl_type) {
 389         case F_RDLCK:
 390                 cephlock->type = CEPH_LOCK_SHARED;
 391                 break;
 392         case F_WRLCK:
 393                 cephlock->type = CEPH_LOCK_EXCL;
 394                 break;
 395         case F_UNLCK:
 396                 cephlock->type = CEPH_LOCK_UNLOCK;
 397                 break;
 398         default:
 399                 dout("Have unknown lock type %d", lock->fl_type);
 400                 err = -EINVAL;
 401         }
 402
 403         return err;
 404 }