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
  13 static inline u64 secure_addr(void *addr)
  14 {
  15         u64 v = lock_secret ^ (u64)(unsigned long)addr;
  16         /*
  17          * Set the most significant bit, so that MDS knows the 'owner'
  18          * is sufficient to identify the owner of lock. (old code uses
  19          * both 'owner' and 'pid')
  20          */
  21         v |= (1ULL << 63);
  22         return v;
  23 }
  24
  25 void __init ceph_flock_init(void)
  26 {
  27         get_random_bytes(&lock_secret, sizeof(lock_secret));
  28 }
  29
  30 /**
  31  * Implement fcntl and flock locking functions.
  32  */
  33 static int ceph_lock_message(u8 lock_type, u16 operation, struct file *file,
  34                              int cmd, u8 wait, struct file_lock *fl)
  35 {
  36         struct inode *inode = file_inode(file);
  37         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
  38         struct ceph_mds_request *req;
  39         int err;
  40         u64 length = 0;
  41         u64 owner;
  42
  43         req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
  44         if (IS_ERR(req))
  45                 return PTR_ERR(req);
  46         req->r_inode = inode;
  47         ihold(inode);
  48         req->r_num_caps = 1;
  49
  50         /* mds requires start and length rather than start and end */
  51         if (LLONG_MAX == fl->fl_end)
  52                 length = 0;
  53         else
  54                 length = fl->fl_end - fl->fl_start + 1;
  55
  56         owner = secure_addr(fl->fl_owner);
  57
  58         dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
  59              "start: %llu, length: %llu, wait: %d, type: %d", (int)lock_type,
  60              (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
  61              wait, fl->fl_type);
  62
  63         req->r_args.filelock_change.rule = lock_type;
  64         req->r_args.filelock_change.type = cmd;
  65         req->r_args.filelock_change.owner = cpu_to_le64(owner);
  66         req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
  67         req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
  68         req->r_args.filelock_change.length = cpu_to_le64(length);
  69         req->r_args.filelock_change.wait = wait;
  70
  71         err = ceph_mdsc_do_request(mdsc, inode, req);
  72
  73         if (operation == CEPH_MDS_OP_GETFILELOCK) {
  74                 fl->fl_pid = le64_to_cpu(req->r_reply_info.filelock_reply->pid);
  75                 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
  76                         fl->fl_type = F_RDLCK;
  77                 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
  78                         fl->fl_type = F_WRLCK;
  79                 else
  80                         fl->fl_type = F_UNLCK;
  81
  82                 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
  83                 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
  84                                                  le64_to_cpu(req->r_reply_info.filelock_reply->length);
  85                 if (length >= 1)
  86                         fl->fl_end = length -1;
  87                 else
  88                         fl->fl_end = 0;
  89
  90         }
  91         ceph_mdsc_put_request(req);
  92         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
  93              "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
  94              (int)operation, (u64)fl->fl_pid, fl->fl_start,
  95              length, wait, fl->fl_type, err);
  96         return err;
  97 }
  98
  99 /**
 100  * Attempt to set an fcntl lock.
 101  * For now, this just goes away to the server. Later it may be more awesome.
 102  */
 103 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 104 {
 105         u8 lock_cmd;
 106         int err;
 107         u8 wait = 0;
 108         u16 op = CEPH_MDS_OP_SETFILELOCK;
 109
 110         if (!(fl->fl_flags & FL_POSIX))
 111                 return -ENOLCK;
 112         /* No mandatory locks */
 113         if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
 114                 return -ENOLCK;
 115
 116         dout("ceph_lock, fl_owner: %p", fl->fl_owner);
 117
 118         /* set wait bit as appropriate, then make command as Ceph expects it*/
 119         if (IS_GETLK(cmd))
 120                 op = CEPH_MDS_OP_GETFILELOCK;
 121         else if (IS_SETLKW(cmd))
 122                 wait = 1;
 123
 124         if (F_RDLCK == fl->fl_type)
 125                 lock_cmd = CEPH_LOCK_SHARED;
 126         else if (F_WRLCK == fl->fl_type)
 127                 lock_cmd = CEPH_LOCK_EXCL;
 128         else
 129                 lock_cmd = CEPH_LOCK_UNLOCK;
 130
 131         err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
 132         if (!err) {
 133                 if (op != CEPH_MDS_OP_GETFILELOCK) {
 134                         dout("mds locked, locking locally");
 135                         err = posix_lock_file(file, fl, NULL);
 136                         if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
 137                                 /* undo! This should only happen if
 138                                  * the kernel detects local
 139                                  * deadlock. */
 140                                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 141                                                   CEPH_LOCK_UNLOCK, 0, fl);
 142                                 dout("got %d on posix_lock_file, undid lock",
 143                                      err);
 144                         }
 145                 }
 146
 147         } else if (err == -ERESTARTSYS) {
 148                 dout("undoing lock\n");
 149                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 150                                   CEPH_LOCK_UNLOCK, 0, fl);
 151         }
 152         return err;
 153 }
 154
 155 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 156 {
 157         u8 lock_cmd;
 158         int err;
 159         u8 wait = 0;
 160
 161         if (!(fl->fl_flags & FL_FLOCK))
 162                 return -ENOLCK;
 163         /* No mandatory locks */
 164         if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
 165                 return -ENOLCK;
 166
 167         dout("ceph_flock, fl_file: %p", fl->fl_file);
 168
 169         if (IS_SETLKW(cmd))
 170                 wait = 1;
 171
 172         if (F_RDLCK == fl->fl_type)
 173                 lock_cmd = CEPH_LOCK_SHARED;
 174         else if (F_WRLCK == fl->fl_type)
 175                 lock_cmd = CEPH_LOCK_EXCL;
 176         else
 177                 lock_cmd = CEPH_LOCK_UNLOCK;
 178
 179         err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
 180                                 file, lock_cmd, wait, fl);
 181         if (!err) {
 182                 err = flock_lock_file_wait(file, fl);
 183                 if (err) {
 184                         ceph_lock_message(CEPH_LOCK_FLOCK,
 185                                           CEPH_MDS_OP_SETFILELOCK,
 186                                           file, CEPH_LOCK_UNLOCK, 0, fl);
 187                         dout("got %d on flock_lock_file_wait, undid lock", err);
 188                 }
 189         } else if (err == -ERESTARTSYS) {
 190                 dout("undoing lock\n");
 191                 ceph_lock_message(CEPH_LOCK_FLOCK,
 192                                   CEPH_MDS_OP_SETFILELOCK,
 193                                   file, CEPH_LOCK_UNLOCK, 0, fl);
 194         }
 195         return err;
 196 }
 197
 198 /**
 199  * Must be called with lock_flocks() already held. Fills in the passed
 200  * counter variables, so you can prepare pagelist metadata before calling
 201  * ceph_encode_locks.
 202  */
 203 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
 204 {
 205         struct file_lock *lock;
 206
 207         *fcntl_count = 0;
 208         *flock_count = 0;
 209
 210         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 211                 if (lock->fl_flags & FL_POSIX)
 212                         ++(*fcntl_count);
 213                 else if (lock->fl_flags & FL_FLOCK)
 214                         ++(*flock_count);
 215         }
 216         dout("counted %d flock locks and %d fcntl locks",
 217              *flock_count, *fcntl_count);
 218 }
 219
 220 /**
 221  * Encode the flock and fcntl locks for the given inode into the ceph_filelock
 222  * array. Must be called with inode->i_lock already held.
 223  * If we encounter more of a specific lock type than expected, return -ENOSPC.
 224  */
 225 int ceph_encode_locks_to_buffer(struct inode *inode,
 226                                 struct ceph_filelock *flocks,
 227                                 int num_fcntl_locks, int num_flock_locks)
 228 {
 229         struct file_lock *lock;
 230         int err = 0;
 231         int seen_fcntl = 0;
 232         int seen_flock = 0;
 233         int l = 0;
 234
 235         dout("encoding %d flock and %d fcntl locks", num_flock_locks,
 236              num_fcntl_locks);
 237
 238         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 239                 if (lock->fl_flags & FL_POSIX) {
 240                         ++seen_fcntl;
 241                         if (seen_fcntl > num_fcntl_locks) {
 242                                 err = -ENOSPC;
 243                                 goto fail;
 244                         }
 245                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 246                         if (err)
 247                                 goto fail;
 248                         ++l;
 249                 }
 250         }
 251         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 252                 if (lock->fl_flags & FL_FLOCK) {
 253                         ++seen_flock;
 254                         if (seen_flock > num_flock_locks) {
 255                                 err = -ENOSPC;
 256                                 goto fail;
 257                         }
 258                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 259                         if (err)
 260                                 goto fail;
 261                         ++l;
 262                 }
 263         }
 264 fail:
 265         return err;
 266 }
 267
 268 /**
 269  * Copy the encoded flock and fcntl locks into the pagelist.
 270  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
 271  * sequential flock locks.
 272  * Returns zero on success.
 273  */
 274 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
 275                            struct ceph_pagelist *pagelist,
 276                            int num_fcntl_locks, int num_flock_locks)
 277 {
 278         int err = 0;
 279         __le32 nlocks;
 280
 281         nlocks = cpu_to_le32(num_fcntl_locks);
 282         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 283         if (err)
 284                 goto out_fail;
 285
 286         err = ceph_pagelist_append(pagelist, flocks,
 287                                    num_fcntl_locks * sizeof(*flocks));
 288         if (err)
 289                 goto out_fail;
 290
 291         nlocks = cpu_to_le32(num_flock_locks);
 292         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 293         if (err)
 294                 goto out_fail;
 295
 296         err = ceph_pagelist_append(pagelist,
 297                                    &flocks[num_fcntl_locks],
 298                                    num_flock_locks * sizeof(*flocks));
 299 out_fail:
 300         return err;
 301 }
 302
 303 /*
 304  * Given a pointer to a lock, convert it to a ceph filelock
 305  */
 306 int lock_to_ceph_filelock(struct file_lock *lock,
 307                           struct ceph_filelock *cephlock)
 308 {
 309         int err = 0;
 310         cephlock->start = cpu_to_le64(lock->fl_start);
 311         cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
 312         cephlock->client = cpu_to_le64(0);
 313         cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
 314         cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
 315
 316         switch (lock->fl_type) {
 317         case F_RDLCK:
 318                 cephlock->type = CEPH_LOCK_SHARED;
 319                 break;
 320         case F_WRLCK:
 321                 cephlock->type = CEPH_LOCK_EXCL;
 322                 break;
 323         case F_UNLCK:
 324                 cephlock->type = CEPH_LOCK_UNLOCK;
 325                 break;
 326         default:
 327                 dout("Have unknown lock type %d", lock->fl_type);
 328                 err = -EINVAL;
 329         }
 330
 331         return err;
 332 }