fs/ceph/locks.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/file.h>
   4 #include <linux/namei.h>
   5
   6 #include "super.h"
   7 #include "mds_client.h"
   8 #include <linux/ceph/pagelist.h>
   9
  10 /**
  11  * Implement fcntl and flock locking functions.
  12  */
  13 static int ceph_lock_message(u8 lock_type, u16 operation, struct file *file,
  14                              int cmd, u8 wait, struct file_lock *fl)
  15 {
  16         struct inode *inode = file_inode(file);
  17         struct ceph_mds_client *mdsc =
  18                 ceph_sb_to_client(inode->i_sb)->mdsc;
  19         struct ceph_mds_request *req;
  20         int err;
  21         u64 length = 0;
  22
  23         req = ceph_mdsc_create_request(mdsc, operation, USE_AUTH_MDS);
  24         if (IS_ERR(req))
  25                 return PTR_ERR(req);
  26         req->r_inode = inode;
  27         ihold(inode);
  28
  29         /* mds requires start and length rather than start and end */
  30         if (LLONG_MAX == fl->fl_end)
  31                 length = 0;
  32         else
  33                 length = fl->fl_end - fl->fl_start + 1;
  34
  35         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
  36              "length: %llu, wait: %d, type: %d", (int)lock_type,
  37              (int)operation, (u64)fl->fl_pid, fl->fl_start,
  38              length, wait, fl->fl_type);
  39
  40         req->r_args.filelock_change.rule = lock_type;
  41         req->r_args.filelock_change.type = cmd;
  42         req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
  43         /* This should be adjusted, but I'm not sure if
  44            namespaces actually get id numbers*/
  45         req->r_args.filelock_change.pid_namespace =
  46                 cpu_to_le64((u64)(unsigned long)fl->fl_nspid);
  47         req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
  48         req->r_args.filelock_change.length = cpu_to_le64(length);
  49         req->r_args.filelock_change.wait = wait;
  50
  51         err = ceph_mdsc_do_request(mdsc, inode, req);
  52
  53         if ( operation == CEPH_MDS_OP_GETFILELOCK){
  54                 fl->fl_pid = le64_to_cpu(req->r_reply_info.filelock_reply->pid);
  55                 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
  56                         fl->fl_type = F_RDLCK;
  57                 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
  58                         fl->fl_type = F_WRLCK;
  59                 else
  60                         fl->fl_type = F_UNLCK;
  61
  62                 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
  63                 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
  64                                                  le64_to_cpu(req->r_reply_info.filelock_reply->length);
  65                 if (length >= 1)
  66                         fl->fl_end = length -1;
  67                 else
  68                         fl->fl_end = 0;
  69
  70         }
  71         ceph_mdsc_put_request(req);
  72         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
  73              "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
  74              (int)operation, (u64)fl->fl_pid, fl->fl_start,
  75              length, wait, fl->fl_type, err);
  76         return err;
  77 }
  78
  79 /**
  80  * Attempt to set an fcntl lock.
  81  * For now, this just goes away to the server. Later it may be more awesome.
  82  */
  83 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
  84 {
  85         u8 lock_cmd;
  86         int err;
  87         u8 wait = 0;
  88         u16 op = CEPH_MDS_OP_SETFILELOCK;
  89
  90         fl->fl_nspid = get_pid(task_tgid(current));
  91         dout("ceph_lock, fl_pid:%d", fl->fl_pid);
  92
  93         /* set wait bit as appropriate, then make command as Ceph expects it*/
  94         if (IS_GETLK(cmd))
  95                 op = CEPH_MDS_OP_GETFILELOCK;
  96         else if (IS_SETLKW(cmd))
  97                 wait = 1;
  98
  99         if (F_RDLCK == fl->fl_type)
 100                 lock_cmd = CEPH_LOCK_SHARED;
 101         else if (F_WRLCK == fl->fl_type)
 102                 lock_cmd = CEPH_LOCK_EXCL;
 103         else
 104                 lock_cmd = CEPH_LOCK_UNLOCK;
 105
 106         err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
 107         if (!err) {
 108                 if ( op != CEPH_MDS_OP_GETFILELOCK ){
 109                         dout("mds locked, locking locally");
 110                         err = posix_lock_file(file, fl, NULL);
 111                         if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
 112                                 /* undo! This should only happen if
 113                                  * the kernel detects local
 114                                  * deadlock. */
 115                                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 116                                                   CEPH_LOCK_UNLOCK, 0, fl);
 117                                 dout("got %d on posix_lock_file, undid lock",
 118                                      err);
 119                         }
 120                 }
 121
 122         } else if (err == -ERESTARTSYS) {
 123                 dout("undoing lock\n");
 124                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 125                                   CEPH_LOCK_UNLOCK, 0, fl);
 126         }
 127         return err;
 128 }
 129
 130 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 131 {
 132         u8 lock_cmd;
 133         int err;
 134         u8 wait = 0;
 135
 136         fl->fl_nspid = get_pid(task_tgid(current));
 137         dout("ceph_flock, fl_pid:%d", fl->fl_pid);
 138
 139         if (IS_SETLKW(cmd))
 140                 wait = 1;
 141
 142         if (F_RDLCK == fl->fl_type)
 143                 lock_cmd = CEPH_LOCK_SHARED;
 144         else if (F_WRLCK == fl->fl_type)
 145                 lock_cmd = CEPH_LOCK_EXCL;
 146         else
 147                 lock_cmd = CEPH_LOCK_UNLOCK;
 148
 149         err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
 150                                 file, lock_cmd, wait, fl);
 151         if (!err) {
 152                 err = flock_lock_file_wait(file, fl);
 153                 if (err) {
 154                         ceph_lock_message(CEPH_LOCK_FLOCK,
 155                                           CEPH_MDS_OP_SETFILELOCK,
 156                                           file, CEPH_LOCK_UNLOCK, 0, fl);
 157                         dout("got %d on flock_lock_file_wait, undid lock", err);
 158                 }
 159         } else if (err == -ERESTARTSYS) {
 160                 dout("undoing lock\n");
 161                 ceph_lock_message(CEPH_LOCK_FLOCK,
 162                                   CEPH_MDS_OP_SETFILELOCK,
 163                                   file, CEPH_LOCK_UNLOCK, 0, fl);
 164         }
 165         return err;
 166 }
 167
 168 /**
 169  * Must be called with lock_flocks() already held. Fills in the passed
 170  * counter variables, so you can prepare pagelist metadata before calling
 171  * ceph_encode_locks.
 172  */
 173 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
 174 {
 175         struct file_lock *lock;
 176
 177         *fcntl_count = 0;
 178         *flock_count = 0;
 179
 180         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 181                 if (lock->fl_flags & FL_POSIX)
 182                         ++(*fcntl_count);
 183                 else if (lock->fl_flags & FL_FLOCK)
 184                         ++(*flock_count);
 185         }
 186         dout("counted %d flock locks and %d fcntl locks",
 187              *flock_count, *fcntl_count);
 188 }
 189
 190 /**
 191  * Encode the flock and fcntl locks for the given inode into the ceph_filelock
 192  * array. Must be called with inode->i_lock already held.
 193  * If we encounter more of a specific lock type than expected, return -ENOSPC.
 194  */
 195 int ceph_encode_locks_to_buffer(struct inode *inode,
 196                                 struct ceph_filelock *flocks,
 197                                 int num_fcntl_locks, int num_flock_locks)
 198 {
 199         struct file_lock *lock;
 200         int err = 0;
 201         int seen_fcntl = 0;
 202         int seen_flock = 0;
 203         int l = 0;
 204
 205         dout("encoding %d flock and %d fcntl locks", num_flock_locks,
 206              num_fcntl_locks);
 207
 208         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 209                 if (lock->fl_flags & FL_POSIX) {
 210                         ++seen_fcntl;
 211                         if (seen_fcntl > num_fcntl_locks) {
 212                                 err = -ENOSPC;
 213                                 goto fail;
 214                         }
 215                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 216                         if (err)
 217                                 goto fail;
 218                         ++l;
 219                 }
 220         }
 221         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 222                 if (lock->fl_flags & FL_FLOCK) {
 223                         ++seen_flock;
 224                         if (seen_flock > num_flock_locks) {
 225                                 err = -ENOSPC;
 226                                 goto fail;
 227                         }
 228                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 229                         if (err)
 230                                 goto fail;
 231                         ++l;
 232                 }
 233         }
 234 fail:
 235         return err;
 236 }
 237
 238 /**
 239  * Copy the encoded flock and fcntl locks into the pagelist.
 240  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
 241  * sequential flock locks.
 242  * Returns zero on success.
 243  */
 244 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
 245                            struct ceph_pagelist *pagelist,
 246                            int num_fcntl_locks, int num_flock_locks)
 247 {
 248         int err = 0;
 249         __le32 nlocks;
 250
 251         nlocks = cpu_to_le32(num_fcntl_locks);
 252         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 253         if (err)
 254                 goto out_fail;
 255
 256         err = ceph_pagelist_append(pagelist, flocks,
 257                                    num_fcntl_locks * sizeof(*flocks));
 258         if (err)
 259                 goto out_fail;
 260
 261         nlocks = cpu_to_le32(num_flock_locks);
 262         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 263         if (err)
 264                 goto out_fail;
 265
 266         err = ceph_pagelist_append(pagelist,
 267                                    &flocks[num_fcntl_locks],
 268                                    num_flock_locks * sizeof(*flocks));
 269 out_fail:
 270         return err;
 271 }
 272
 273 /*
 274  * Given a pointer to a lock, convert it to a ceph filelock
 275  */
 276 int lock_to_ceph_filelock(struct file_lock *lock,
 277                           struct ceph_filelock *cephlock)
 278 {
 279         int err = 0;
 280
 281         cephlock->start = cpu_to_le64(lock->fl_start);
 282         cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
 283         cephlock->client = cpu_to_le64(0);
 284         cephlock->pid = cpu_to_le64(lock->fl_pid);
 285         cephlock->pid_namespace =
 286                 cpu_to_le64((u64)(unsigned long)lock->fl_nspid);
 287
 288         switch (lock->fl_type) {
 289         case F_RDLCK:
 290                 cephlock->type = CEPH_LOCK_SHARED;
 291                 break;
 292         case F_WRLCK:
 293                 cephlock->type = CEPH_LOCK_EXCL;
 294                 break;
 295         case F_UNLCK:
 296                 cephlock->type = CEPH_LOCK_UNLOCK;
 297                 break;
 298         default:
 299                 dout("Have unknown lock type %d", lock->fl_type);
 300                 err = -EINVAL;
 301         }
 302
 303         return err;
 304 }