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->f_dentry->d_inode;
  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
  41         req->r_args.filelock_change.rule = lock_type;
  42         req->r_args.filelock_change.type = cmd;
  43         req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
  44         /* This should be adjusted, but I'm not sure if
  45            namespaces actually get id numbers*/
  46         req->r_args.filelock_change.pid_namespace =
  47                 cpu_to_le64((u64)(unsigned long)fl->fl_nspid);
  48         req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
  49         req->r_args.filelock_change.length = cpu_to_le64(length);
  50         req->r_args.filelock_change.wait = wait;
  51
  52         err = ceph_mdsc_do_request(mdsc, inode, req);
  53
  54         if ( operation == CEPH_MDS_OP_GETFILELOCK){
  55                 fl->fl_pid = le64_to_cpu(req->r_reply_info.filelock_reply->pid);
  56                 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
  57                         fl->fl_type = F_RDLCK;
  58                 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
  59                         fl->fl_type = F_WRLCK;
  60                 else
  61                         fl->fl_type = F_UNLCK;
  62
  63                 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
  64                 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
  65                                                  le64_to_cpu(req->r_reply_info.filelock_reply->length);
  66                 if (length >= 1)
  67                         fl->fl_end = length -1;
  68                 else
  69                         fl->fl_end = 0;
  70
  71         }
  72         ceph_mdsc_put_request(req);
  73         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
  74              "length: %llu, wait: %d, type`: %d, err code %d", (int)lock_type,
  75              (int)operation, (u64)fl->fl_pid, fl->fl_start,
  76              length, wait, fl->fl_type, err);
  77         return err;
  78 }
  79
  80 /**
  81  * Attempt to set an fcntl lock.
  82  * For now, this just goes away to the server. Later it may be more awesome.
  83  */
  84 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
  85 {
  86         u8 lock_cmd;
  87         int err;
  88         u8 wait = 0;
  89         u16 op = CEPH_MDS_OP_SETFILELOCK;
  90
  91         fl->fl_nspid = get_pid(task_tgid(current));
  92         dout("ceph_lock, fl_pid:%d", fl->fl_pid);
  93
  94         /* set wait bit as appropriate, then make command as Ceph expects it*/
  95         if (F_SETLKW == cmd)
  96                 wait = 1;
  97         if (F_GETLK == cmd)
  98                 op = CEPH_MDS_OP_GETFILELOCK;
  99
 100         if (F_RDLCK == fl->fl_type)
 101                 lock_cmd = CEPH_LOCK_SHARED;
 102         else if (F_WRLCK == fl->fl_type)
 103                 lock_cmd = CEPH_LOCK_EXCL;
 104         else
 105                 lock_cmd = CEPH_LOCK_UNLOCK;
 106
 107         err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
 108         if (!err) {
 109                 if ( op != CEPH_MDS_OP_GETFILELOCK ){
 110                         dout("mds locked, locking locally");
 111                         err = posix_lock_file(file, fl, NULL);
 112                         if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
 113                                 /* undo! This should only happen if the kernel detects
 114                                  * local 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", err);
 118                         }
 119                 }
 120
 121         } else {
 122                 dout("mds returned error code %d", err);
 123         }
 124         return err;
 125 }
 126
 127 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 128 {
 129         u8 lock_cmd;
 130         int err;
 131         u8 wait = 1;
 132
 133         fl->fl_nspid = get_pid(task_tgid(current));
 134         dout("ceph_flock, fl_pid:%d", fl->fl_pid);
 135
 136         /* set wait bit, then clear it out of cmd*/
 137         if (cmd & LOCK_NB)
 138                 wait = 0;
 139         cmd = cmd & (LOCK_SH | LOCK_EX | LOCK_UN);
 140         /* set command sequence that Ceph wants to see:
 141            shared lock, exclusive lock, or unlock */
 142         if (LOCK_SH == cmd)
 143                 lock_cmd = CEPH_LOCK_SHARED;
 144         else if (LOCK_EX == cmd)
 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 {
 160                 dout("mds error code %d", err);
 161         }
 162         return err;
 163 }
 164
 165 /**
 166  * Must be called with BKL already held. Fills in the passed
 167  * counter variables, so you can prepare pagelist metadata before calling
 168  * ceph_encode_locks.
 169  */
 170 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
 171 {
 172         struct file_lock *lock;
 173
 174         *fcntl_count = 0;
 175         *flock_count = 0;
 176
 177         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 178                 if (lock->fl_flags & FL_POSIX)
 179                         ++(*fcntl_count);
 180                 else if (lock->fl_flags & FL_FLOCK)
 181                         ++(*flock_count);
 182         }
 183         dout("counted %d flock locks and %d fcntl locks",
 184              *flock_count, *fcntl_count);
 185 }
 186
 187 /**
 188  * Encode the flock and fcntl locks for the given inode into the pagelist.
 189  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
 190  * sequential flock locks.
 191  * Must be called with lock_flocks() already held.
 192  * If we encounter more of a specific lock type than expected,
 193  * we return the value 1.
 194  */
 195 int ceph_encode_locks(struct inode *inode, struct ceph_pagelist *pagelist,
 196                       int num_fcntl_locks, int num_flock_locks)
 197 {
 198         struct file_lock *lock;
 199         struct ceph_filelock cephlock;
 200         int err = 0;
 201         int seen_fcntl = 0;
 202         int seen_flock = 0;
 203
 204         dout("encoding %d flock and %d fcntl locks", num_flock_locks,
 205              num_fcntl_locks);
 206         err = ceph_pagelist_append(pagelist, &num_fcntl_locks, sizeof(u32));
 207         if (err)
 208                 goto fail;
 209         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 210                 if (lock->fl_flags & FL_POSIX) {
 211                         ++seen_fcntl;
 212                         if (seen_fcntl > num_fcntl_locks) {
 213                                 err = -ENOSPC;
 214                                 goto fail;
 215                         }
 216                         err = lock_to_ceph_filelock(lock, &cephlock);
 217                         if (err)
 218                                 goto fail;
 219                         err = ceph_pagelist_append(pagelist, &cephlock,
 220                                            sizeof(struct ceph_filelock));
 221                 }
 222                 if (err)
 223                         goto fail;
 224         }
 225
 226         err = ceph_pagelist_append(pagelist, &num_flock_locks, sizeof(u32));
 227         if (err)
 228                 goto fail;
 229         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 230                 if (lock->fl_flags & FL_FLOCK) {
 231                         ++seen_flock;
 232                         if (seen_flock > num_flock_locks) {
 233                                 err = -ENOSPC;
 234                                 goto fail;
 235                         }
 236                         err = lock_to_ceph_filelock(lock, &cephlock);
 237                         if (err)
 238                                 goto fail;
 239                         err = ceph_pagelist_append(pagelist, &cephlock,
 240                                            sizeof(struct ceph_filelock));
 241                 }
 242                 if (err)
 243                         goto fail;
 244         }
 245 fail:
 246         return err;
 247 }
 248
 249 /*
 250  * Given a pointer to a lock, convert it to a ceph filelock
 251  */
 252 int lock_to_ceph_filelock(struct file_lock *lock,
 253                           struct ceph_filelock *cephlock)
 254 {
 255         int err = 0;
 256
 257         cephlock->start = cpu_to_le64(lock->fl_start);
 258         cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
 259         cephlock->client = cpu_to_le64(0);
 260         cephlock->pid = cpu_to_le64(lock->fl_pid);
 261         cephlock->pid_namespace =
 262                 cpu_to_le64((u64)(unsigned long)lock->fl_nspid);
 263
 264         switch (lock->fl_type) {
 265         case F_RDLCK:
 266                 cephlock->type = CEPH_LOCK_SHARED;
 267                 break;
 268         case F_WRLCK:
 269                 cephlock->type = CEPH_LOCK_EXCL;
 270                 break;
 271         case F_UNLCK:
 272                 cephlock->type = CEPH_LOCK_UNLOCK;
 273                 break;
 274         default:
 275                 dout("Have unknown lock type %d", lock->fl_type);
 276                 err = -EINVAL;
 277         }
 278
 279         return err;
 280 }