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
  49         /* mds requires start and length rather than start and end */
  50         if (LLONG_MAX == fl->fl_end)
  51                 length = 0;
  52         else
  53                 length = fl->fl_end - fl->fl_start + 1;
  54
  55         if (lock_type == CEPH_LOCK_FCNTL)
  56                 owner = secure_addr(fl->fl_owner);
  57         else
  58                 owner = secure_addr(fl->fl_file);
  59
  60         dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
  61              "start: %llu, length: %llu, wait: %d, type: %d", (int)lock_type,
  62              (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
  63              wait, fl->fl_type);
  64
  65         req->r_args.filelock_change.rule = lock_type;
  66         req->r_args.filelock_change.type = cmd;
  67         req->r_args.filelock_change.owner = cpu_to_le64(owner);
  68         req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
  69         req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
  70         req->r_args.filelock_change.length = cpu_to_le64(length);
  71         req->r_args.filelock_change.wait = wait;
  72
  73         err = ceph_mdsc_do_request(mdsc, inode, req);
  74
  75         if (operation == CEPH_MDS_OP_GETFILELOCK) {
  76                 fl->fl_pid = le64_to_cpu(req->r_reply_info.filelock_reply->pid);
  77                 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
  78                         fl->fl_type = F_RDLCK;
  79                 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
  80                         fl->fl_type = F_WRLCK;
  81                 else
  82                         fl->fl_type = F_UNLCK;
  83
  84                 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
  85                 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
  86                                                  le64_to_cpu(req->r_reply_info.filelock_reply->length);
  87                 if (length >= 1)
  88                         fl->fl_end = length -1;
  89                 else
  90                         fl->fl_end = 0;
  91
  92         }
  93         ceph_mdsc_put_request(req);
  94         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
  95              "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
  96              (int)operation, (u64)fl->fl_pid, fl->fl_start,
  97              length, wait, fl->fl_type, err);
  98         return err;
  99 }
 100
 101 /**
 102  * Attempt to set an fcntl lock.
 103  * For now, this just goes away to the server. Later it may be more awesome.
 104  */
 105 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 106 {
 107         u8 lock_cmd;
 108         int err;
 109         u8 wait = 0;
 110         u16 op = CEPH_MDS_OP_SETFILELOCK;
 111
 112         if (!(fl->fl_flags & FL_POSIX))
 113                 return -ENOLCK;
 114         /* No mandatory locks */
 115         if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
 116                 return -ENOLCK;
 117
 118         dout("ceph_lock, fl_owner: %p", fl->fl_owner);
 119
 120         /* set wait bit as appropriate, then make command as Ceph expects it*/
 121         if (IS_GETLK(cmd))
 122                 op = CEPH_MDS_OP_GETFILELOCK;
 123         else if (IS_SETLKW(cmd))
 124                 wait = 1;
 125
 126         if (F_RDLCK == fl->fl_type)
 127                 lock_cmd = CEPH_LOCK_SHARED;
 128         else if (F_WRLCK == fl->fl_type)
 129                 lock_cmd = CEPH_LOCK_EXCL;
 130         else
 131                 lock_cmd = CEPH_LOCK_UNLOCK;
 132
 133         err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
 134         if (!err) {
 135                 if (op != CEPH_MDS_OP_GETFILELOCK) {
 136                         dout("mds locked, locking locally");
 137                         err = posix_lock_file(file, fl, NULL);
 138                         if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
 139                                 /* undo! This should only happen if
 140                                  * the kernel detects local
 141                                  * deadlock. */
 142                                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 143                                                   CEPH_LOCK_UNLOCK, 0, fl);
 144                                 dout("got %d on posix_lock_file, undid lock",
 145                                      err);
 146                         }
 147                 }
 148
 149         } else if (err == -ERESTARTSYS) {
 150                 dout("undoing lock\n");
 151                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 152                                   CEPH_LOCK_UNLOCK, 0, fl);
 153         }
 154         return err;
 155 }
 156
 157 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 158 {
 159         u8 lock_cmd;
 160         int err;
 161         u8 wait = 0;
 162
 163         if (!(fl->fl_flags & FL_FLOCK))
 164                 return -ENOLCK;
 165         /* No mandatory locks */
 166         if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
 167                 return -ENOLCK;
 168
 169         dout("ceph_flock, fl_file: %p", fl->fl_file);
 170
 171         if (IS_SETLKW(cmd))
 172                 wait = 1;
 173
 174         if (F_RDLCK == fl->fl_type)
 175                 lock_cmd = CEPH_LOCK_SHARED;
 176         else if (F_WRLCK == fl->fl_type)
 177                 lock_cmd = CEPH_LOCK_EXCL;
 178         else
 179                 lock_cmd = CEPH_LOCK_UNLOCK;
 180
 181         err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
 182                                 file, lock_cmd, wait, fl);
 183         if (!err) {
 184                 err = flock_lock_file_wait(file, fl);
 185                 if (err) {
 186                         ceph_lock_message(CEPH_LOCK_FLOCK,
 187                                           CEPH_MDS_OP_SETFILELOCK,
 188                                           file, CEPH_LOCK_UNLOCK, 0, fl);
 189                         dout("got %d on flock_lock_file_wait, undid lock", err);
 190                 }
 191         } else if (err == -ERESTARTSYS) {
 192                 dout("undoing lock\n");
 193                 ceph_lock_message(CEPH_LOCK_FLOCK,
 194                                   CEPH_MDS_OP_SETFILELOCK,
 195                                   file, CEPH_LOCK_UNLOCK, 0, fl);
 196         }
 197         return err;
 198 }
 199
 200 /**
 201  * Must be called with lock_flocks() already held. Fills in the passed
 202  * counter variables, so you can prepare pagelist metadata before calling
 203  * ceph_encode_locks.
 204  */
 205 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
 206 {
 207         struct file_lock *lock;
 208
 209         *fcntl_count = 0;
 210         *flock_count = 0;
 211
 212         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 213                 if (lock->fl_flags & FL_POSIX)
 214                         ++(*fcntl_count);
 215                 else if (lock->fl_flags & FL_FLOCK)
 216                         ++(*flock_count);
 217         }
 218         dout("counted %d flock locks and %d fcntl locks",
 219              *flock_count, *fcntl_count);
 220 }
 221
 222 /**
 223  * Encode the flock and fcntl locks for the given inode into the ceph_filelock
 224  * array. Must be called with inode->i_lock already held.
 225  * If we encounter more of a specific lock type than expected, return -ENOSPC.
 226  */
 227 int ceph_encode_locks_to_buffer(struct inode *inode,
 228                                 struct ceph_filelock *flocks,
 229                                 int num_fcntl_locks, int num_flock_locks)
 230 {
 231         struct file_lock *lock;
 232         int err = 0;
 233         int seen_fcntl = 0;
 234         int seen_flock = 0;
 235         int l = 0;
 236
 237         dout("encoding %d flock and %d fcntl locks", num_flock_locks,
 238              num_fcntl_locks);
 239
 240         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 241                 if (lock->fl_flags & FL_POSIX) {
 242                         ++seen_fcntl;
 243                         if (seen_fcntl > num_fcntl_locks) {
 244                                 err = -ENOSPC;
 245                                 goto fail;
 246                         }
 247                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 248                         if (err)
 249                                 goto fail;
 250                         ++l;
 251                 }
 252         }
 253         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 254                 if (lock->fl_flags & FL_FLOCK) {
 255                         ++seen_flock;
 256                         if (seen_flock > num_flock_locks) {
 257                                 err = -ENOSPC;
 258                                 goto fail;
 259                         }
 260                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 261                         if (err)
 262                                 goto fail;
 263                         ++l;
 264                 }
 265         }
 266 fail:
 267         return err;
 268 }
 269
 270 /**
 271  * Copy the encoded flock and fcntl locks into the pagelist.
 272  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
 273  * sequential flock locks.
 274  * Returns zero on success.
 275  */
 276 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
 277                            struct ceph_pagelist *pagelist,
 278                            int num_fcntl_locks, int num_flock_locks)
 279 {
 280         int err = 0;
 281         __le32 nlocks;
 282
 283         nlocks = cpu_to_le32(num_fcntl_locks);
 284         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 285         if (err)
 286                 goto out_fail;
 287
 288         err = ceph_pagelist_append(pagelist, flocks,
 289                                    num_fcntl_locks * sizeof(*flocks));
 290         if (err)
 291                 goto out_fail;
 292
 293         nlocks = cpu_to_le32(num_flock_locks);
 294         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 295         if (err)
 296                 goto out_fail;
 297
 298         err = ceph_pagelist_append(pagelist,
 299                                    &flocks[num_fcntl_locks],
 300                                    num_flock_locks * sizeof(*flocks));
 301 out_fail:
 302         return err;
 303 }
 304
 305 /*
 306  * Given a pointer to a lock, convert it to a ceph filelock
 307  */
 308 int lock_to_ceph_filelock(struct file_lock *lock,
 309                           struct ceph_filelock *cephlock)
 310 {
 311         int err = 0;
 312         cephlock->start = cpu_to_le64(lock->fl_start);
 313         cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
 314         cephlock->client = cpu_to_le64(0);
 315         cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
 316         if (lock->fl_flags & FL_POSIX)
 317                 cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
 318         else
 319                 cephlock->owner = cpu_to_le64(secure_addr(lock->fl_file));
 320
 321         switch (lock->fl_type) {
 322         case F_RDLCK:
 323                 cephlock->type = CEPH_LOCK_SHARED;
 324                 break;
 325         case F_WRLCK:
 326                 cephlock->type = CEPH_LOCK_EXCL;
 327                 break;
 328         case F_UNLCK:
 329                 cephlock->type = CEPH_LOCK_UNLOCK;
 330                 break;
 331         default:
 332                 dout("Have unknown lock type %d", lock->fl_type);
 333                 err = -EINVAL;
 334         }
 335
 336         return err;
 337 }