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         intr_req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETFILELOCK,
 131                                             USE_AUTH_MDS);
 132         if (IS_ERR(intr_req))
 133                 return PTR_ERR(intr_req);
 134
 135         intr_req->r_inode = inode;
 136         ihold(inode);
 137         intr_req->r_num_caps = 1;
 138
 139         intr_req->r_args.filelock_change = req->r_args.filelock_change;
 140         intr_req->r_args.filelock_change.rule = lock_type;
 141         intr_req->r_args.filelock_change.type = CEPH_LOCK_UNLOCK;
 142
 143         err = ceph_mdsc_do_request(mdsc, inode, intr_req);
 144         ceph_mdsc_put_request(intr_req);
 145
 146         if (err && err != -ERESTARTSYS)
 147                 return err;
 148
 149         wait_for_completion(&req->r_completion);
 150         return 0;
 151 }
 152
 153 /**
 154  * Attempt to set an fcntl lock.
 155  * For now, this just goes away to the server. Later it may be more awesome.
 156  */
 157 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 158 {
 159         u8 lock_cmd;
 160         int err;
 161         u8 wait = 0;
 162         u16 op = CEPH_MDS_OP_SETFILELOCK;
 163
 164         if (!(fl->fl_flags & FL_POSIX))
 165                 return -ENOLCK;
 166         /* No mandatory locks */
 167         if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
 168                 return -ENOLCK;
 169
 170         dout("ceph_lock, fl_owner: %p", fl->fl_owner);
 171
 172         /* set wait bit as appropriate, then make command as Ceph expects it*/
 173         if (IS_GETLK(cmd))
 174                 op = CEPH_MDS_OP_GETFILELOCK;
 175         else if (IS_SETLKW(cmd))
 176                 wait = 1;
 177
 178         if (F_RDLCK == fl->fl_type)
 179                 lock_cmd = CEPH_LOCK_SHARED;
 180         else if (F_WRLCK == fl->fl_type)
 181                 lock_cmd = CEPH_LOCK_EXCL;
 182         else
 183                 lock_cmd = CEPH_LOCK_UNLOCK;
 184
 185         err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
 186         if (!err) {
 187                 if (op != CEPH_MDS_OP_GETFILELOCK) {
 188                         dout("mds locked, locking locally");
 189                         err = posix_lock_file(file, fl, NULL);
 190                         if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
 191                                 /* undo! This should only happen if
 192                                  * the kernel detects local
 193                                  * deadlock. */
 194                                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 195                                                   CEPH_LOCK_UNLOCK, 0, fl);
 196                                 dout("got %d on posix_lock_file, undid lock",
 197                                      err);
 198                         }
 199                 }
 200         }
 201         return err;
 202 }
 203
 204 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 205 {
 206         u8 lock_cmd;
 207         int err;
 208         u8 wait = 0;
 209
 210         if (!(fl->fl_flags & FL_FLOCK))
 211                 return -ENOLCK;
 212         /* No mandatory locks */
 213         if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
 214                 return -ENOLCK;
 215
 216         dout("ceph_flock, fl_file: %p", fl->fl_file);
 217
 218         if (IS_SETLKW(cmd))
 219                 wait = 1;
 220
 221         if (F_RDLCK == fl->fl_type)
 222                 lock_cmd = CEPH_LOCK_SHARED;
 223         else if (F_WRLCK == fl->fl_type)
 224                 lock_cmd = CEPH_LOCK_EXCL;
 225         else
 226                 lock_cmd = CEPH_LOCK_UNLOCK;
 227
 228         err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
 229                                 file, lock_cmd, wait, fl);
 230         if (!err) {
 231                 err = flock_lock_file_wait(file, fl);
 232                 if (err) {
 233                         ceph_lock_message(CEPH_LOCK_FLOCK,
 234                                           CEPH_MDS_OP_SETFILELOCK,
 235                                           file, CEPH_LOCK_UNLOCK, 0, fl);
 236                         dout("got %d on flock_lock_file_wait, undid lock", err);
 237                 }
 238         }
 239         return err;
 240 }
 241
 242 /*
 243  * Fills in the passed counter variables, so you can prepare pagelist metadata
 244  * before calling ceph_encode_locks.
 245  *
 246  * FIXME: add counters to struct file_lock_context so we don't need to do this?
 247  */
 248 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
 249 {
 250         struct file_lock *lock;
 251         struct file_lock_context *ctx;
 252
 253         *fcntl_count = 0;
 254         *flock_count = 0;
 255
 256         spin_lock(&inode->i_lock);
 257         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 258                 if (lock->fl_flags & FL_POSIX)
 259                         ++(*fcntl_count);
 260         }
 261
 262         ctx = inode->i_flctx;
 263         if (ctx) {
 264                 list_for_each_entry(lock, &ctx->flc_flock, fl_list)
 265                         ++(*flock_count);
 266         }
 267         spin_unlock(&inode->i_lock);
 268         dout("counted %d flock locks and %d fcntl locks",
 269              *flock_count, *fcntl_count);
 270 }
 271
 272 /**
 273  * Encode the flock and fcntl locks for the given inode into the ceph_filelock
 274  * array. Must be called with inode->i_lock already held.
 275  * If we encounter more of a specific lock type than expected, return -ENOSPC.
 276  */
 277 int ceph_encode_locks_to_buffer(struct inode *inode,
 278                                 struct ceph_filelock *flocks,
 279                                 int num_fcntl_locks, int num_flock_locks)
 280 {
 281         struct file_lock *lock;
 282         struct file_lock_context *ctx;
 283         int err = 0;
 284         int seen_fcntl = 0;
 285         int seen_flock = 0;
 286         int l = 0;
 287
 288         dout("encoding %d flock and %d fcntl locks", num_flock_locks,
 289              num_fcntl_locks);
 290
 291         spin_lock(&inode->i_lock);
 292         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 293                 if (lock->fl_flags & FL_POSIX) {
 294                         ++seen_fcntl;
 295                         if (seen_fcntl > num_fcntl_locks) {
 296                                 err = -ENOSPC;
 297                                 goto fail;
 298                         }
 299                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 300                         if (err)
 301                                 goto fail;
 302                         ++l;
 303                 }
 304         }
 305
 306         ctx = inode->i_flctx;
 307         if (ctx) {
 308                 list_for_each_entry(lock, &ctx->flc_flock, fl_list) {
 309                         ++seen_flock;
 310                         if (seen_flock > num_flock_locks) {
 311                                 err = -ENOSPC;
 312                                 goto fail;
 313                         }
 314                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 315                         if (err)
 316                                 goto fail;
 317                         ++l;
 318                 }
 319         }
 320 fail:
 321         spin_unlock(&inode->i_lock);
 322         return err;
 323 }
 324
 325 /**
 326  * Copy the encoded flock and fcntl locks into the pagelist.
 327  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
 328  * sequential flock locks.
 329  * Returns zero on success.
 330  */
 331 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
 332                            struct ceph_pagelist *pagelist,
 333                            int num_fcntl_locks, int num_flock_locks)
 334 {
 335         int err = 0;
 336         __le32 nlocks;
 337
 338         nlocks = cpu_to_le32(num_fcntl_locks);
 339         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 340         if (err)
 341                 goto out_fail;
 342
 343         err = ceph_pagelist_append(pagelist, flocks,
 344                                    num_fcntl_locks * sizeof(*flocks));
 345         if (err)
 346                 goto out_fail;
 347
 348         nlocks = cpu_to_le32(num_flock_locks);
 349         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 350         if (err)
 351                 goto out_fail;
 352
 353         err = ceph_pagelist_append(pagelist,
 354                                    &flocks[num_fcntl_locks],
 355                                    num_flock_locks * sizeof(*flocks));
 356 out_fail:
 357         return err;
 358 }
 359
 360 /*
 361  * Given a pointer to a lock, convert it to a ceph filelock
 362  */
 363 int lock_to_ceph_filelock(struct file_lock *lock,
 364                           struct ceph_filelock *cephlock)
 365 {
 366         int err = 0;
 367         cephlock->start = cpu_to_le64(lock->fl_start);
 368         cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
 369         cephlock->client = cpu_to_le64(0);
 370         cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
 371         cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
 372
 373         switch (lock->fl_type) {
 374         case F_RDLCK:
 375                 cephlock->type = CEPH_LOCK_SHARED;
 376                 break;
 377         case F_WRLCK:
 378                 cephlock->type = CEPH_LOCK_EXCL;
 379                 break;
 380         case F_UNLCK:
 381                 cephlock->type = CEPH_LOCK_UNLOCK;
 382                 break;
 383         default:
 384                 dout("Have unknown lock type %d", lock->fl_type);
 385                 err = -EINVAL;
 386         }
 387
 388         return err;
 389 }