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         if (lock_type == CEPH_LOCK_FCNTL)
  57                 owner = secure_addr(fl->fl_owner);
  58         else
  59                 owner = secure_addr(fl->fl_file);
  60
  61         dout("ceph_lock_message: rule: %d, op: %d, owner: %llx, pid: %llu, "
  62              "start: %llu, length: %llu, wait: %d, type: %d", (int)lock_type,
  63              (int)operation, owner, (u64)fl->fl_pid, fl->fl_start, length,
  64              wait, fl->fl_type);
  65
  66         req->r_args.filelock_change.rule = lock_type;
  67         req->r_args.filelock_change.type = cmd;
  68         req->r_args.filelock_change.owner = cpu_to_le64(owner);
  69         req->r_args.filelock_change.pid = cpu_to_le64((u64)fl->fl_pid);
  70         req->r_args.filelock_change.start = cpu_to_le64(fl->fl_start);
  71         req->r_args.filelock_change.length = cpu_to_le64(length);
  72         req->r_args.filelock_change.wait = wait;
  73
  74         err = ceph_mdsc_do_request(mdsc, inode, req);
  75
  76         if (operation == CEPH_MDS_OP_GETFILELOCK) {
  77                 fl->fl_pid = le64_to_cpu(req->r_reply_info.filelock_reply->pid);
  78                 if (CEPH_LOCK_SHARED == req->r_reply_info.filelock_reply->type)
  79                         fl->fl_type = F_RDLCK;
  80                 else if (CEPH_LOCK_EXCL == req->r_reply_info.filelock_reply->type)
  81                         fl->fl_type = F_WRLCK;
  82                 else
  83                         fl->fl_type = F_UNLCK;
  84
  85                 fl->fl_start = le64_to_cpu(req->r_reply_info.filelock_reply->start);
  86                 length = le64_to_cpu(req->r_reply_info.filelock_reply->start) +
  87                                                  le64_to_cpu(req->r_reply_info.filelock_reply->length);
  88                 if (length >= 1)
  89                         fl->fl_end = length -1;
  90                 else
  91                         fl->fl_end = 0;
  92
  93         }
  94         ceph_mdsc_put_request(req);
  95         dout("ceph_lock_message: rule: %d, op: %d, pid: %llu, start: %llu, "
  96              "length: %llu, wait: %d, type: %d, err code %d", (int)lock_type,
  97              (int)operation, (u64)fl->fl_pid, fl->fl_start,
  98              length, wait, fl->fl_type, err);
  99         return err;
 100 }
 101
 102 /**
 103  * Attempt to set an fcntl lock.
 104  * For now, this just goes away to the server. Later it may be more awesome.
 105  */
 106 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 107 {
 108         u8 lock_cmd;
 109         int err;
 110         u8 wait = 0;
 111         u16 op = CEPH_MDS_OP_SETFILELOCK;
 112
 113         if (!(fl->fl_flags & FL_POSIX))
 114                 return -ENOLCK;
 115         /* No mandatory locks */
 116         if (__mandatory_lock(file->f_mapping->host) && fl->fl_type != F_UNLCK)
 117                 return -ENOLCK;
 118
 119         dout("ceph_lock, fl_owner: %p", fl->fl_owner);
 120
 121         /* set wait bit as appropriate, then make command as Ceph expects it*/
 122         if (IS_GETLK(cmd))
 123                 op = CEPH_MDS_OP_GETFILELOCK;
 124         else if (IS_SETLKW(cmd))
 125                 wait = 1;
 126
 127         if (F_RDLCK == fl->fl_type)
 128                 lock_cmd = CEPH_LOCK_SHARED;
 129         else if (F_WRLCK == fl->fl_type)
 130                 lock_cmd = CEPH_LOCK_EXCL;
 131         else
 132                 lock_cmd = CEPH_LOCK_UNLOCK;
 133
 134         err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
 135         if (!err) {
 136                 if (op != CEPH_MDS_OP_GETFILELOCK) {
 137                         dout("mds locked, locking locally");
 138                         err = posix_lock_file(file, fl, NULL);
 139                         if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
 140                                 /* undo! This should only happen if
 141                                  * the kernel detects local
 142                                  * deadlock. */
 143                                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 144                                                   CEPH_LOCK_UNLOCK, 0, fl);
 145                                 dout("got %d on posix_lock_file, undid lock",
 146                                      err);
 147                         }
 148                 }
 149
 150         } else if (err == -ERESTARTSYS) {
 151                 dout("undoing lock\n");
 152                 ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
 153                                   CEPH_LOCK_UNLOCK, 0, fl);
 154         }
 155         return err;
 156 }
 157
 158 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 159 {
 160         u8 lock_cmd;
 161         int err;
 162         u8 wait = 0;
 163
 164         if (!(fl->fl_flags & FL_FLOCK))
 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_flock, fl_file: %p", fl->fl_file);
 171
 172         if (IS_SETLKW(cmd))
 173                 wait = 1;
 174
 175         if (F_RDLCK == fl->fl_type)
 176                 lock_cmd = CEPH_LOCK_SHARED;
 177         else if (F_WRLCK == fl->fl_type)
 178                 lock_cmd = CEPH_LOCK_EXCL;
 179         else
 180                 lock_cmd = CEPH_LOCK_UNLOCK;
 181
 182         err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
 183                                 file, lock_cmd, wait, fl);
 184         if (!err) {
 185                 err = flock_lock_file_wait(file, fl);
 186                 if (err) {
 187                         ceph_lock_message(CEPH_LOCK_FLOCK,
 188                                           CEPH_MDS_OP_SETFILELOCK,
 189                                           file, CEPH_LOCK_UNLOCK, 0, fl);
 190                         dout("got %d on flock_lock_file_wait, undid lock", err);
 191                 }
 192         } else if (err == -ERESTARTSYS) {
 193                 dout("undoing lock\n");
 194                 ceph_lock_message(CEPH_LOCK_FLOCK,
 195                                   CEPH_MDS_OP_SETFILELOCK,
 196                                   file, CEPH_LOCK_UNLOCK, 0, fl);
 197         }
 198         return err;
 199 }
 200
 201 /**
 202  * Must be called with lock_flocks() already held. Fills in the passed
 203  * counter variables, so you can prepare pagelist metadata before calling
 204  * ceph_encode_locks.
 205  */
 206 void ceph_count_locks(struct inode *inode, int *fcntl_count, int *flock_count)
 207 {
 208         struct file_lock *lock;
 209
 210         *fcntl_count = 0;
 211         *flock_count = 0;
 212
 213         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 214                 if (lock->fl_flags & FL_POSIX)
 215                         ++(*fcntl_count);
 216                 else if (lock->fl_flags & FL_FLOCK)
 217                         ++(*flock_count);
 218         }
 219         dout("counted %d flock locks and %d fcntl locks",
 220              *flock_count, *fcntl_count);
 221 }
 222
 223 /**
 224  * Encode the flock and fcntl locks for the given inode into the ceph_filelock
 225  * array. Must be called with inode->i_lock already held.
 226  * If we encounter more of a specific lock type than expected, return -ENOSPC.
 227  */
 228 int ceph_encode_locks_to_buffer(struct inode *inode,
 229                                 struct ceph_filelock *flocks,
 230                                 int num_fcntl_locks, int num_flock_locks)
 231 {
 232         struct file_lock *lock;
 233         int err = 0;
 234         int seen_fcntl = 0;
 235         int seen_flock = 0;
 236         int l = 0;
 237
 238         dout("encoding %d flock and %d fcntl locks", num_flock_locks,
 239              num_fcntl_locks);
 240
 241         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 242                 if (lock->fl_flags & FL_POSIX) {
 243                         ++seen_fcntl;
 244                         if (seen_fcntl > num_fcntl_locks) {
 245                                 err = -ENOSPC;
 246                                 goto fail;
 247                         }
 248                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 249                         if (err)
 250                                 goto fail;
 251                         ++l;
 252                 }
 253         }
 254         for (lock = inode->i_flock; lock != NULL; lock = lock->fl_next) {
 255                 if (lock->fl_flags & FL_FLOCK) {
 256                         ++seen_flock;
 257                         if (seen_flock > num_flock_locks) {
 258                                 err = -ENOSPC;
 259                                 goto fail;
 260                         }
 261                         err = lock_to_ceph_filelock(lock, &flocks[l]);
 262                         if (err)
 263                                 goto fail;
 264                         ++l;
 265                 }
 266         }
 267 fail:
 268         return err;
 269 }
 270
 271 /**
 272  * Copy the encoded flock and fcntl locks into the pagelist.
 273  * Format is: #fcntl locks, sequential fcntl locks, #flock locks,
 274  * sequential flock locks.
 275  * Returns zero on success.
 276  */
 277 int ceph_locks_to_pagelist(struct ceph_filelock *flocks,
 278                            struct ceph_pagelist *pagelist,
 279                            int num_fcntl_locks, int num_flock_locks)
 280 {
 281         int err = 0;
 282         __le32 nlocks;
 283
 284         nlocks = cpu_to_le32(num_fcntl_locks);
 285         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 286         if (err)
 287                 goto out_fail;
 288
 289         err = ceph_pagelist_append(pagelist, flocks,
 290                                    num_fcntl_locks * sizeof(*flocks));
 291         if (err)
 292                 goto out_fail;
 293
 294         nlocks = cpu_to_le32(num_flock_locks);
 295         err = ceph_pagelist_append(pagelist, &nlocks, sizeof(nlocks));
 296         if (err)
 297                 goto out_fail;
 298
 299         err = ceph_pagelist_append(pagelist,
 300                                    &flocks[num_fcntl_locks],
 301                                    num_flock_locks * sizeof(*flocks));
 302 out_fail:
 303         return err;
 304 }
 305
 306 /*
 307  * Given a pointer to a lock, convert it to a ceph filelock
 308  */
 309 int lock_to_ceph_filelock(struct file_lock *lock,
 310                           struct ceph_filelock *cephlock)
 311 {
 312         int err = 0;
 313         cephlock->start = cpu_to_le64(lock->fl_start);
 314         cephlock->length = cpu_to_le64(lock->fl_end - lock->fl_start + 1);
 315         cephlock->client = cpu_to_le64(0);
 316         cephlock->pid = cpu_to_le64((u64)lock->fl_pid);
 317         if (lock->fl_flags & FL_POSIX)
 318                 cephlock->owner = cpu_to_le64(secure_addr(lock->fl_owner));
 319         else
 320                 cephlock->owner = cpu_to_le64(secure_addr(lock->fl_file));
 321
 322         switch (lock->fl_type) {
 323         case F_RDLCK:
 324                 cephlock->type = CEPH_LOCK_SHARED;
 325                 break;
 326         case F_WRLCK:
 327                 cephlock->type = CEPH_LOCK_EXCL;
 328                 break;
 329         case F_UNLCK:
 330                 cephlock->type = CEPH_LOCK_UNLOCK;
 331                 break;
 332         default:
 333                 dout("Have unknown lock type %d", lock->fl_type);
 334                 err = -EINVAL;
 335         }
 336
 337         return err;
 338 }