server/change.c

   1 /*
   2  * Server-side change notification management
   3  *
   4  * Copyright (C) 1998 Alexandre Julliard
   5  * Copyright (C) 2006 Mike McCormack
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with this library; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
  20  */
  21
  22 #include "config.h"
  23 #include "wine/port.h"
  24
  25 #include <assert.h>
  26 #include <fcntl.h>
  27 #include <stdio.h>
  28 #include <stdlib.h>
  29 #include <signal.h>
  30 #include <sys/stat.h>
  31 #include <sys/types.h>
  32 #include <limits.h>
  33 #include <dirent.h>
  34 #include <errno.h>
  35
  36 #include "ntstatus.h"
  37 #define WIN32_NO_STATUS
  38 #include "windef.h"
  39
  40 #include "file.h"
  41 #include "handle.h"
  42 #include "thread.h"
  43 #include "request.h"
  44 #include "process.h"
  45 #include "security.h"
  46 #include "winternl.h"
  47
  48 /* dnotify support */
  49
  50 #ifdef linux
  51 #ifndef F_NOTIFY
  52 #define F_NOTIFY 1026
  53 #define DN_ACCESS       0x00000001      /* File accessed */
  54 #define DN_MODIFY       0x00000002      /* File modified */
  55 #define DN_CREATE       0x00000004      /* File created */
  56 #define DN_DELETE       0x00000008      /* File removed */
  57 #define DN_RENAME       0x00000010      /* File renamed */
  58 #define DN_ATTRIB       0x00000020      /* File changed attributes */
  59 #define DN_MULTISHOT    0x80000000      /* Don't remove notifier */
  60 #endif
  61 #endif
  62
  63 /* inotify support */
  64
  65 #ifdef HAVE_SYS_INOTIFY_H
  66 #include <sys/inotify.h>
  67 #define USE_INOTIFY
  68 #elif defined(__linux__) && defined(__i386__)
  69
  70 #define SYS_inotify_init        291
  71 #define SYS_inotify_add_watch   292
  72 #define SYS_inotify_rm_watch    293
  73
  74 struct inotify_event {
  75     int           wd;
  76     unsigned int  mask;
  77     unsigned int  cookie;
  78     unsigned int  len;
  79     char          name[1];
  80 };
  81
  82 #define IN_ACCESS        0x00000001
  83 #define IN_MODIFY        0x00000002
  84 #define IN_ATTRIB        0x00000004
  85 #define IN_CLOSE_WRITE   0x00000008
  86 #define IN_CLOSE_NOWRITE 0x00000010
  87 #define IN_OPEN          0x00000020
  88 #define IN_MOVED_FROM    0x00000040
  89 #define IN_MOVED_TO      0x00000080
  90 #define IN_CREATE        0x00000100
  91 #define IN_DELETE        0x00000200
  92 #define IN_DELETE_SELF   0x00000400
  93
  94 #define IN_ISDIR         0x40000000
  95
  96 static inline int inotify_init( void )
  97 {
  98     return syscall( SYS_inotify_init );
  99 }
 100
 101 static inline int inotify_add_watch( int fd, const char *name, unsigned int mask )
 102 {
 103     return syscall( SYS_inotify_add_watch, fd, name, mask );
 104 }
 105
 106 static inline int inotify_rm_watch( int fd, int wd )
 107 {
 108     return syscall( SYS_inotify_rm_watch, fd, wd );
 109 }
 110
 111 #define USE_INOTIFY
 112
 113 #endif
 114
 115 struct inode;
 116
 117 static void free_inode( struct inode *inode );
 118
 119 static struct fd *inotify_fd;
 120
 121 struct change_record {
 122     struct list entry;
 123     unsigned int cookie;
 124     struct filesystem_event event;
 125 };
 126
 127 struct dir
 128 {
 129     struct object  obj;      /* object header */
 130     struct fd     *fd;       /* file descriptor to the directory */
 131     mode_t         mode;     /* file stat.st_mode */
 132     uid_t          uid;      /* file stat.st_uid */
 133     struct list    entry;    /* entry in global change notifications list */
 134     unsigned int   filter;   /* notification filter */
 135     int            notified; /* SIGIO counter */
 136     int            want_data; /* return change data */
 137     int            subtree;  /* do we want to watch subdirectories? */
 138     struct list    change_records;   /* data for the change */
 139     struct list    in_entry; /* entry in the inode dirs list */
 140     struct inode  *inode;    /* inode of the associated directory */
 141 };
 142
 143 static struct fd *dir_get_fd( struct object *obj );
 144 static struct security_descriptor *dir_get_sd( struct object *obj );
 145 static int dir_set_sd( struct object *obj, const struct security_descriptor *sd,
 146                        unsigned int set_info );
 147 static void dir_dump( struct object *obj, int verbose );
 148 static void dir_destroy( struct object *obj );
 149
 150 static const struct object_ops dir_ops =
 151 {
 152     sizeof(struct dir),       /* size */
 153     dir_dump,                 /* dump */
 154     no_get_type,              /* get_type */
 155     add_queue,                /* add_queue */
 156     remove_queue,             /* remove_queue */
 157     default_fd_signaled,      /* signaled */
 158     no_satisfied,             /* satisfied */
 159     no_signal,                /* signal */
 160     dir_get_fd,               /* get_fd */
 161     default_fd_map_access,    /* map_access */
 162     dir_get_sd,               /* get_sd */
 163     dir_set_sd,               /* set_sd */
 164     no_lookup_name,           /* lookup_name */
 165     no_open_file,             /* open_file */
 166     fd_close_handle,          /* close_handle */
 167     dir_destroy               /* destroy */
 168 };
 169
 170 static int dir_get_poll_events( struct fd *fd );
 171 static enum server_fd_type dir_get_fd_type( struct fd *fd );
 172
 173 static const struct fd_ops dir_fd_ops =
 174 {
 175     dir_get_poll_events,         /* get_poll_events */
 176     default_poll_event,          /* poll_event */
 177     no_flush,                    /* flush */
 178     dir_get_fd_type,             /* get_fd_type */
 179     default_fd_ioctl,            /* ioctl */
 180     default_fd_queue_async,      /* queue_async */
 181     default_fd_reselect_async,   /* reselect_async */
 182     default_fd_cancel_async      /* cancel_async */
 183 };
 184
 185 static struct list change_list = LIST_INIT(change_list);
 186
 187 static void dnotify_adjust_changes( struct dir *dir )
 188 {
 189 #if defined(F_SETSIG) && defined(F_NOTIFY)
 190     int fd = get_unix_fd( dir->fd );
 191     unsigned int filter = dir->filter;
 192     unsigned int val;
 193     if ( 0 > fcntl( fd, F_SETSIG, SIGIO) )
 194         return;
 195
 196     val = DN_MULTISHOT;
 197     if (filter & FILE_NOTIFY_CHANGE_FILE_NAME)
 198         val |= DN_RENAME | DN_DELETE | DN_CREATE;
 199     if (filter & FILE_NOTIFY_CHANGE_DIR_NAME)
 200         val |= DN_RENAME | DN_DELETE | DN_CREATE;
 201     if (filter & FILE_NOTIFY_CHANGE_ATTRIBUTES)
 202         val |= DN_ATTRIB;
 203     if (filter & FILE_NOTIFY_CHANGE_SIZE)
 204         val |= DN_MODIFY;
 205     if (filter & FILE_NOTIFY_CHANGE_LAST_WRITE)
 206         val |= DN_MODIFY;
 207     if (filter & FILE_NOTIFY_CHANGE_LAST_ACCESS)
 208         val |= DN_ACCESS;
 209     if (filter & FILE_NOTIFY_CHANGE_CREATION)
 210         val |= DN_CREATE;
 211     if (filter & FILE_NOTIFY_CHANGE_SECURITY)
 212         val |= DN_ATTRIB;
 213     fcntl( fd, F_NOTIFY, val );
 214 #endif
 215 }
 216
 217 /* insert change in the global list */
 218 static inline void insert_change( struct dir *dir )
 219 {
 220     sigset_t sigset;
 221
 222     sigemptyset( &sigset );
 223     sigaddset( &sigset, SIGIO );
 224     sigprocmask( SIG_BLOCK, &sigset, NULL );
 225     list_add_head( &change_list, &dir->entry );
 226     sigprocmask( SIG_UNBLOCK, &sigset, NULL );
 227 }
 228
 229 /* remove change from the global list */
 230 static inline void remove_change( struct dir *dir )
 231 {
 232     sigset_t sigset;
 233
 234     sigemptyset( &sigset );
 235     sigaddset( &sigset, SIGIO );
 236     sigprocmask( SIG_BLOCK, &sigset, NULL );
 237     list_remove( &dir->entry );
 238     sigprocmask( SIG_UNBLOCK, &sigset, NULL );
 239 }
 240
 241 static void dir_dump( struct object *obj, int verbose )
 242 {
 243     struct dir *dir = (struct dir *)obj;
 244     assert( obj->ops == &dir_ops );
 245     fprintf( stderr, "Dirfile fd=%p filter=%08x\n", dir->fd, dir->filter );
 246 }
 247
 248 /* enter here directly from SIGIO signal handler */
 249 void do_change_notify( int unix_fd )
 250 {
 251     struct dir *dir;
 252
 253     /* FIXME: this is O(n) ... probably can be improved */
 254     LIST_FOR_EACH_ENTRY( dir, &change_list, struct dir, entry )
 255     {
 256         if (get_unix_fd( dir->fd ) != unix_fd) continue;
 257         interlocked_xchg_add( &dir->notified, 1 );
 258         break;
 259     }
 260 }
 261
 262 /* SIGIO callback, called synchronously with the poll loop */
 263 void sigio_callback(void)
 264 {
 265     struct dir *dir;
 266
 267     LIST_FOR_EACH_ENTRY( dir, &change_list, struct dir, entry )
 268     {
 269         if (interlocked_xchg( &dir->notified, 0 ))
 270             fd_async_wake_up( dir->fd, ASYNC_TYPE_WAIT, STATUS_NOTIFY_ENUM_DIR );
 271     }
 272 }
 273
 274 static struct fd *dir_get_fd( struct object *obj )
 275 {
 276     struct dir *dir = (struct dir *)obj;
 277     assert( obj->ops == &dir_ops );
 278     return (struct fd *)grab_object( dir->fd );
 279 }
 280
 281 static int get_dir_unix_fd( struct dir *dir )
 282 {
 283     return get_unix_fd( dir->fd );
 284 }
 285
 286 static struct security_descriptor *dir_get_sd( struct object *obj )
 287 {
 288     struct dir *dir = (struct dir *)obj;
 289     int unix_fd;
 290     struct stat st;
 291     struct security_descriptor *sd;
 292     assert( obj->ops == &dir_ops );
 293
 294     unix_fd = get_dir_unix_fd( dir );
 295
 296     if (unix_fd == -1 || fstat( unix_fd, &st ) == -1)
 297         return obj->sd;
 298
 299     /* mode and uid the same? if so, no need to re-generate security descriptor */
 300     if (obj->sd &&
 301         (st.st_mode & (S_IRWXU|S_IRWXO)) == (dir->mode & (S_IRWXU|S_IRWXO)) &&
 302         (st.st_uid == dir->uid))
 303         return obj->sd;
 304
 305     sd = mode_to_sd( st.st_mode,
 306                      security_unix_uid_to_sid( st.st_uid ),
 307                      token_get_primary_group( current->process->token ));
 308     if (!sd) return obj->sd;
 309
 310     dir->mode = st.st_mode;
 311     dir->uid = st.st_uid;
 312     free( obj->sd );
 313     obj->sd = sd;
 314     return sd;
 315 }
 316
 317 static int dir_set_sd( struct object *obj, const struct security_descriptor *sd,
 318                        unsigned int set_info )
 319 {
 320     struct dir *dir = (struct dir *)obj;
 321     const SID *owner;
 322     struct stat st;
 323     mode_t mode;
 324     int unix_fd;
 325
 326     assert( obj->ops == &dir_ops );
 327
 328     unix_fd = get_dir_unix_fd( dir );
 329
 330     if (unix_fd == -1 || fstat( unix_fd, &st ) == -1) return 1;
 331
 332     if (set_info & OWNER_SECURITY_INFORMATION)
 333     {
 334         owner = sd_get_owner( sd );
 335         if (!owner)
 336         {
 337             set_error( STATUS_INVALID_SECURITY_DESCR );
 338             return 0;
 339         }
 340         if (!obj->sd || !security_equal_sid( owner, sd_get_owner( obj->sd ) ))
 341         {
 342             /* FIXME: get Unix uid and call fchown */
 343         }
 344     }
 345     else if (obj->sd)
 346         owner = sd_get_owner( obj->sd );
 347     else
 348         owner = token_get_user( current->process->token );
 349
 350     if (set_info & DACL_SECURITY_INFORMATION)
 351     {
 352         /* keep the bits that we don't map to access rights in the ACL */
 353         mode = st.st_mode & (S_ISUID|S_ISGID|S_ISVTX);
 354         mode |= sd_to_mode( sd, owner );
 355
 356         if (((st.st_mode ^ mode) & (S_IRWXU|S_IRWXG|S_IRWXO)) && fchmod( unix_fd, mode ) == -1)
 357         {
 358             file_set_error();
 359             return 0;
 360         }
 361     }
 362     return 1;
 363 }
 364
 365 static struct change_record *get_first_change_record( struct dir *dir )
 366 {
 367     struct list *ptr = list_head( &dir->change_records );
 368     if (!ptr) return NULL;
 369     list_remove( ptr );
 370     return LIST_ENTRY( ptr, struct change_record, entry );
 371 }
 372
 373 static void dir_destroy( struct object *obj )
 374 {
 375     struct change_record *record;
 376     struct dir *dir = (struct dir *)obj;
 377     assert (obj->ops == &dir_ops);
 378
 379     if (dir->filter)
 380         remove_change( dir );
 381
 382     if (dir->inode)
 383     {
 384         list_remove( &dir->in_entry );
 385         free_inode( dir->inode );
 386     }
 387
 388     while ((record = get_first_change_record( dir ))) free( record );
 389
 390     release_object( dir->fd );
 391
 392     if (inotify_fd && list_empty( &change_list ))
 393     {
 394         release_object( inotify_fd );
 395         inotify_fd = NULL;
 396     }
 397 }
 398
 399 struct dir *get_dir_obj( struct process *process, obj_handle_t handle, unsigned int access )
 400 {
 401     return (struct dir *)get_handle_obj( process, handle, access, &dir_ops );
 402 }
 403
 404 static int dir_get_poll_events( struct fd *fd )
 405 {
 406     return 0;
 407 }
 408
 409 static enum server_fd_type dir_get_fd_type( struct fd *fd )
 410 {
 411     return FD_TYPE_DIR;
 412 }
 413
 414 #ifdef USE_INOTIFY
 415
 416 #define HASH_SIZE 31
 417
 418 struct inode {
 419     struct list ch_entry;    /* entry in the children list */
 420     struct list children;    /* children of this inode */
 421     struct inode *parent;    /* parent of this inode */
 422     struct list dirs;        /* directory handles watching this inode */
 423     struct list ino_entry;   /* entry in the inode hash */
 424     struct list wd_entry;    /* entry in the watch descriptor hash */
 425     dev_t dev;               /* device number */
 426     ino_t ino;               /* device's inode number */
 427     int wd;                  /* inotify's watch descriptor */
 428     char *name;              /* basename name of the inode */
 429 };
 430
 431 static struct list inode_hash[ HASH_SIZE ];
 432 static struct list wd_hash[ HASH_SIZE ];
 433
 434 static int inotify_add_dir( char *path, unsigned int filter );
 435
 436 static struct inode *inode_from_wd( int wd )
 437 {
 438     struct list *bucket = &wd_hash[ wd % HASH_SIZE ];
 439     struct inode *inode;
 440
 441     LIST_FOR_EACH_ENTRY( inode, bucket, struct inode, wd_entry )
 442         if (inode->wd == wd)
 443             return inode;
 444
 445     return NULL;
 446 }
 447
 448 static inline struct list *get_hash_list( dev_t dev, ino_t ino )
 449 {
 450     return &inode_hash[ (ino ^ dev) % HASH_SIZE ];
 451 }
 452
 453 static struct inode *find_inode( dev_t dev, ino_t ino )
 454 {
 455     struct list *bucket = get_hash_list( dev, ino );
 456     struct inode *inode;
 457
 458     LIST_FOR_EACH_ENTRY( inode, bucket, struct inode, ino_entry )
 459         if (inode->ino == ino && inode->dev == dev)
 460              return inode;
 461
 462     return NULL;
 463 }
 464
 465 static struct inode *create_inode( dev_t dev, ino_t ino )
 466 {
 467     struct inode *inode;
 468
 469     inode = malloc( sizeof *inode );
 470     if (inode)
 471     {
 472         list_init( &inode->children );
 473         list_init( &inode->dirs );
 474         inode->ino = ino;
 475         inode->dev = dev;
 476         inode->wd = -1;
 477         inode->parent = NULL;
 478         inode->name = NULL;
 479         list_add_tail( get_hash_list( dev, ino ), &inode->ino_entry );
 480     }
 481     return inode;
 482 }
 483
 484 static struct inode *get_inode( dev_t dev, ino_t ino )
 485 {
 486     struct inode *inode;
 487
 488     inode = find_inode( dev, ino );
 489     if (inode)
 490         return inode;
 491     return create_inode( dev, ino );
 492 }
 493
 494 static void inode_set_wd( struct inode *inode, int wd )
 495 {
 496     if (inode->wd != -1)
 497         list_remove( &inode->wd_entry );
 498     inode->wd = wd;
 499     list_add_tail( &wd_hash[ wd % HASH_SIZE ], &inode->wd_entry );
 500 }
 501
 502 static void inode_set_name( struct inode *inode, const char *name )
 503 {
 504     free (inode->name);
 505     inode->name = name ? strdup( name ) : NULL;
 506 }
 507
 508 static void free_inode( struct inode *inode )
 509 {
 510     int subtree = 0, watches = 0;
 511     struct inode *tmp, *next;
 512     struct dir *dir;
 513
 514     LIST_FOR_EACH_ENTRY( dir, &inode->dirs, struct dir, in_entry )
 515     {
 516         subtree |= dir->subtree;
 517         watches++;
 518     }
 519
 520     if (!subtree && !inode->parent)
 521     {
 522         LIST_FOR_EACH_ENTRY_SAFE( tmp, next, &inode->children,
 523                                   struct inode, ch_entry )
 524         {
 525             assert( tmp != inode );
 526             assert( tmp->parent == inode );
 527             free_inode( tmp );
 528         }
 529     }
 530
 531     if (watches)
 532         return;
 533
 534     if (inode->parent)
 535         list_remove( &inode->ch_entry );
 536
 537     /* disconnect remaining children from the parent */
 538     LIST_FOR_EACH_ENTRY_SAFE( tmp, next, &inode->children, struct inode, ch_entry )
 539     {
 540         list_remove( &tmp->ch_entry );
 541         tmp->parent = NULL;
 542     }
 543
 544     if (inode->wd != -1)
 545     {
 546         inotify_rm_watch( get_unix_fd( inotify_fd ), inode->wd );
 547         list_remove( &inode->wd_entry );
 548     }
 549     list_remove( &inode->ino_entry );
 550
 551     free( inode->name );
 552     free( inode );
 553 }
 554
 555 static struct inode *inode_add( struct inode *parent,
 556                                 dev_t dev, ino_t ino, const char *name )
 557 {
 558     struct inode *inode;
 559
 560     inode = get_inode( dev, ino );
 561     if (!inode)
 562         return NULL;
 563
 564     if (!inode->parent)
 565     {
 566         list_add_tail( &parent->children, &inode->ch_entry );
 567         inode->parent = parent;
 568         assert( inode != parent );
 569     }
 570     inode_set_name( inode, name );
 571
 572     return inode;
 573 }
 574
 575 static struct inode *inode_from_name( struct inode *inode, const char *name )
 576 {
 577     struct inode *i;
 578
 579     LIST_FOR_EACH_ENTRY( i, &inode->children, struct inode, ch_entry )
 580         if (i->name && !strcmp( i->name, name ))
 581             return i;
 582     return NULL;
 583 }
 584
 585 static int inotify_get_poll_events( struct fd *fd );
 586 static void inotify_poll_event( struct fd *fd, int event );
 587
 588 static const struct fd_ops inotify_fd_ops =
 589 {
 590     inotify_get_poll_events,     /* get_poll_events */
 591     inotify_poll_event,          /* poll_event */
 592     NULL,                        /* flush */
 593     NULL,                        /* get_fd_type */
 594     NULL,                        /* ioctl */
 595     NULL,                        /* queue_async */
 596     NULL,                        /* reselect_async */
 597     NULL,                        /* cancel_async */
 598 };
 599
 600 static int inotify_get_poll_events( struct fd *fd )
 601 {
 602     return POLLIN;
 603 }
 604
 605 static void inotify_do_change_notify( struct dir *dir, unsigned int action,
 606                                       unsigned int cookie, const char *relpath )
 607 {
 608     struct change_record *record;
 609
 610     assert( dir->obj.ops == &dir_ops );
 611
 612     if (dir->want_data)
 613     {
 614         size_t len = strlen(relpath);
 615         record = malloc( offsetof(struct change_record, event.name[len]) );
 616         if (!record)
 617             return;
 618
 619         record->cookie = cookie;
 620         record->event.action = action;
 621         memcpy( record->event.name, relpath, len );
 622         record->event.len = len;
 623
 624         list_add_tail( &dir->change_records, &record->entry );
 625     }
 626
 627     fd_async_wake_up( dir->fd, ASYNC_TYPE_WAIT, STATUS_ALERTED );
 628 }
 629
 630 static unsigned int filter_from_event( struct inotify_event *ie )
 631 {
 632     unsigned int filter = 0;
 633
 634     if (ie->mask & (IN_MOVED_FROM | IN_MOVED_TO | IN_DELETE | IN_CREATE))
 635         filter |= FILE_NOTIFY_CHANGE_FILE_NAME | FILE_NOTIFY_CHANGE_DIR_NAME;
 636     if (ie->mask & IN_MODIFY)
 637         filter |= FILE_NOTIFY_CHANGE_SIZE | FILE_NOTIFY_CHANGE_LAST_WRITE;
 638     if (ie->mask & IN_ATTRIB)
 639         filter |= FILE_NOTIFY_CHANGE_ATTRIBUTES | FILE_NOTIFY_CHANGE_SECURITY;
 640     if (ie->mask & IN_ACCESS)
 641         filter |= FILE_NOTIFY_CHANGE_LAST_ACCESS;
 642     if (ie->mask & IN_CREATE)
 643         filter |= FILE_NOTIFY_CHANGE_CREATION;
 644
 645     if (ie->mask & IN_ISDIR)
 646         filter &= ~FILE_NOTIFY_CHANGE_FILE_NAME;
 647     else
 648         filter &= ~FILE_NOTIFY_CHANGE_DIR_NAME;
 649
 650     return filter;
 651 }
 652
 653 /* scan up the parent directories for watches */
 654 static unsigned int filter_from_inode( struct inode *inode, int is_parent )
 655 {
 656     unsigned int filter = 0;
 657     struct dir *dir;
 658
 659     /* combine filters from parents watching subtrees */
 660     while (inode)
 661     {
 662         LIST_FOR_EACH_ENTRY( dir, &inode->dirs, struct dir, in_entry )
 663             if (dir->subtree || !is_parent)
 664                 filter |= dir->filter;
 665         is_parent = 1;
 666         inode = inode->parent;
 667     }
 668
 669     return filter;
 670 }
 671
 672 static char *inode_get_path( struct inode *inode, int sz )
 673 {
 674     struct list *head;
 675     char *path;
 676     int len;
 677
 678     if (!inode)
 679         return NULL;
 680
 681     head = list_head( &inode->dirs );
 682     if (head)
 683     {
 684         int unix_fd = get_unix_fd( LIST_ENTRY( head, struct dir, in_entry )->fd );
 685         path = malloc ( 32 + sz );
 686         if (path)
 687             sprintf( path, "/proc/self/fd/%u/", unix_fd );
 688         return path;
 689     }
 690
 691     if (!inode->name)
 692         return NULL;
 693
 694     len = strlen( inode->name );
 695     path = inode_get_path( inode->parent, sz + len + 1 );
 696     if (!path)
 697         return NULL;
 698
 699     strcat( path, inode->name );
 700     strcat( path, "/" );
 701
 702     return path;
 703 }
 704
 705 static void inode_check_dir( struct inode *parent, const char *name )
 706 {
 707     char *path;
 708     unsigned int filter;
 709     struct inode *inode;
 710     struct stat st;
 711     int wd = -1;
 712
 713     path = inode_get_path( parent, strlen(name) );
 714     if (!path)
 715         return;
 716
 717     strcat( path, name );
 718
 719     if (stat( path, &st ) < 0)
 720         goto end;
 721
 722     filter = filter_from_inode( parent, 1 );
 723     if (!filter)
 724         goto end;
 725
 726     inode = inode_add( parent, st.st_dev, st.st_ino, name );
 727     if (!inode || inode->wd != -1)
 728         goto end;
 729
 730     wd = inotify_add_dir( path, filter );
 731     if (wd != -1)
 732         inode_set_wd( inode, wd );
 733     else
 734         free_inode( inode );
 735
 736 end:
 737     free( path );
 738 }
 739
 740 static int prepend( char **path, const char *segment )
 741 {
 742     int extra;
 743     char *p;
 744
 745     extra = strlen( segment ) + 1;
 746     if (*path)
 747     {
 748         int len = strlen( *path ) + 1;
 749         p = realloc( *path, len + extra );
 750         if (!p) return 0;
 751         memmove( &p[ extra ], p, len );
 752         p[ extra - 1 ] = '/';
 753         memcpy( p, segment, extra - 1 );
 754     }
 755     else
 756     {
 757         p = malloc( extra );
 758         if (!p) return 0;
 759         memcpy( p, segment, extra );
 760     }
 761
 762     *path = p;
 763
 764     return 1;
 765 }
 766
 767 static void inotify_notify_all( struct inotify_event *ie )
 768 {
 769     unsigned int filter, action;
 770     struct inode *inode, *i;
 771     char *path = NULL;
 772     struct dir *dir;
 773
 774     inode = inode_from_wd( ie->wd );
 775     if (!inode)
 776     {
 777         fprintf( stderr, "no inode matches %d\n", ie->wd);
 778         return;
 779     }
 780
 781     filter = filter_from_event( ie );
 782
 783     if (ie->mask & IN_CREATE)
 784     {
 785         if (ie->mask & IN_ISDIR)
 786             inode_check_dir( inode, ie->name );
 787
 788         action = FILE_ACTION_ADDED;
 789     }
 790     else if (ie->mask & IN_DELETE)
 791         action = FILE_ACTION_REMOVED;
 792     else if (ie->mask & IN_MOVED_FROM)
 793         action = FILE_ACTION_RENAMED_OLD_NAME;
 794     else if (ie->mask & IN_MOVED_TO)
 795         action = FILE_ACTION_RENAMED_NEW_NAME;
 796     else
 797         action = FILE_ACTION_MODIFIED;
 798
 799     /*
 800      * Work our way up the inode hierarchy
 801      *  extending the relative path as we go
 802      *  and notifying all recursive watches.
 803      */
 804     if (!prepend( &path, ie->name ))
 805         return;
 806
 807     for (i = inode; i; i = i->parent)
 808     {
 809         LIST_FOR_EACH_ENTRY( dir, &i->dirs, struct dir, in_entry )
 810             if ((filter & dir->filter) && (i==inode || dir->subtree))
 811                 inotify_do_change_notify( dir, action, ie->cookie, path );
 812
 813         if (!i->name || !prepend( &path, i->name ))
 814             break;
 815     }
 816
 817     free( path );
 818
 819     if (ie->mask & IN_DELETE)
 820     {
 821         i = inode_from_name( inode, ie->name );
 822         if (i)
 823             free_inode( i );
 824     }
 825 }
 826
 827 static void inotify_poll_event( struct fd *fd, int event )
 828 {
 829     int r, ofs, unix_fd;
 830     char buffer[0x1000];
 831     struct inotify_event *ie;
 832
 833     unix_fd = get_unix_fd( fd );
 834     r = read( unix_fd, buffer, sizeof buffer );
 835     if (r < 0)
 836     {
 837         fprintf(stderr,"inotify_poll_event(): inotify read failed!\n");
 838         return;
 839     }
 840
 841     for( ofs = 0; ofs < r - offsetof(struct inotify_event, name); )
 842     {
 843         ie = (struct inotify_event*) &buffer[ofs];
 844         if (!ie->len)
 845             break;
 846         ofs += offsetof( struct inotify_event, name[ie->len] );
 847         if (ofs > r) break;
 848         inotify_notify_all( ie );
 849     }
 850 }
 851
 852 static inline struct fd *create_inotify_fd( void )
 853 {
 854     int unix_fd;
 855
 856     unix_fd = inotify_init();
 857     if (unix_fd<0)
 858         return NULL;
 859     return create_anonymous_fd( &inotify_fd_ops, unix_fd, NULL, 0 );
 860 }
 861
 862 static int map_flags( unsigned int filter )
 863 {
 864     unsigned int mask;
 865
 866     /* always watch these so we can track subdirectories in recursive watches */
 867     mask = (IN_MOVED_FROM | IN_MOVED_TO | IN_DELETE | IN_CREATE | IN_DELETE_SELF);
 868
 869     if (filter & FILE_NOTIFY_CHANGE_ATTRIBUTES)
 870         mask |= IN_ATTRIB;
 871     if (filter & FILE_NOTIFY_CHANGE_SIZE)
 872         mask |= IN_MODIFY;
 873     if (filter & FILE_NOTIFY_CHANGE_LAST_WRITE)
 874         mask |= IN_MODIFY;
 875     if (filter & FILE_NOTIFY_CHANGE_LAST_ACCESS)
 876         mask |= IN_ACCESS;
 877     if (filter & FILE_NOTIFY_CHANGE_SECURITY)
 878         mask |= IN_ATTRIB;
 879
 880     return mask;
 881 }
 882
 883 static int inotify_add_dir( char *path, unsigned int filter )
 884 {
 885     int wd = inotify_add_watch( get_unix_fd( inotify_fd ),
 886                                 path, map_flags( filter ) );
 887     if (wd != -1)
 888         set_fd_events( inotify_fd, POLLIN );
 889     return wd;
 890 }
 891
 892 static int init_inotify( void )
 893 {
 894     int i;
 895
 896     if (inotify_fd)
 897         return 1;
 898
 899     inotify_fd = create_inotify_fd();
 900     if (!inotify_fd)
 901         return 0;
 902
 903     for (i=0; i<HASH_SIZE; i++)
 904     {
 905         list_init( &inode_hash[i] );
 906         list_init( &wd_hash[i] );
 907     }
 908
 909     return 1;
 910 }
 911
 912 static int inotify_adjust_changes( struct dir *dir )
 913 {
 914     unsigned int filter;
 915     struct inode *inode;
 916     struct stat st;
 917     char path[32];
 918     int wd, unix_fd;
 919
 920     if (!inotify_fd)
 921         return 0;
 922
 923     unix_fd = get_unix_fd( dir->fd );
 924
 925     inode = dir->inode;
 926     if (!inode)
 927     {
 928         /* check if this fd is already being watched */
 929         if (-1 == fstat( unix_fd, &st ))
 930             return 0;
 931
 932         inode = get_inode( st.st_dev, st.st_ino );
 933         if (!inode)
 934             inode = create_inode( st.st_dev, st.st_ino );
 935         if (!inode)
 936             return 0;
 937         list_add_tail( &inode->dirs, &dir->in_entry );
 938         dir->inode = inode;
 939     }
 940
 941     filter = filter_from_inode( inode, 0 );
 942
 943     sprintf( path, "/proc/self/fd/%u", unix_fd );
 944     wd = inotify_add_dir( path, filter );
 945     if (wd == -1) return 0;
 946
 947     inode_set_wd( inode, wd );
 948
 949     return 1;
 950 }
 951
 952 static char *get_basename( const char *link )
 953 {
 954     char *buffer, *name = NULL;
 955     int r, n = 0x100;
 956
 957     while (1)
 958     {
 959         buffer = malloc( n );
 960         if (!buffer) return NULL;
 961
 962         r = readlink( link, buffer, n );
 963         if (r < 0)
 964             break;
 965
 966         if (r < n)
 967         {
 968             name = buffer;
 969             break;
 970         }
 971         free( buffer );
 972         n *= 2;
 973     }
 974
 975     if (name)
 976     {
 977         while (r > 0 && name[ r - 1 ] == '/' )
 978             r--;
 979         name[ r ] = 0;
 980
 981         name = strrchr( name, '/' );
 982         if (name)
 983             name = strdup( &name[1] );
 984     }
 985
 986     free( buffer );
 987     return name;
 988 }
 989
 990 static int dir_add_to_existing_notify( struct dir *dir )
 991 {
 992     struct inode *inode, *parent;
 993     unsigned int filter = 0;
 994     struct stat st, st_new;
 995     char link[35], *name;
 996     int wd, unix_fd;
 997
 998     if (!inotify_fd)
 999         return 0;
1000
1001     unix_fd = get_unix_fd( dir->fd );
1002
1003     /* check if it's in the list of inodes we want to watch */
1004     if (-1 == fstat( unix_fd, &st_new ))
1005         return 0;
1006     inode = find_inode( st_new.st_dev, st_new.st_ino );
1007     if (inode)
1008         return 0;
1009
1010     /* lookup the parent */
1011     sprintf( link, "/proc/self/fd/%u/..", unix_fd );
1012     if (-1 == stat( link, &st ))
1013         return 0;
1014
1015     /*
1016      * If there's no parent, stop.  We could keep going adding
1017      *  ../ to the path until we hit the root of the tree or
1018      *  find a recursively watched ancestor.
1019      * Assume it's too expensive to search up the tree for now.
1020      */
1021     parent = find_inode( st.st_dev, st.st_ino );
1022     if (!parent)
1023         return 0;
1024
1025     if (parent->wd == -1)
1026         return 0;
1027
1028     filter = filter_from_inode( parent, 1 );
1029     if (!filter)
1030         return 0;
1031
1032     sprintf( link, "/proc/self/fd/%u", unix_fd );
1033     name = get_basename( link );
1034     if (!name)
1035         return 0;
1036     inode = inode_add( parent, st_new.st_dev, st_new.st_ino, name );
1037     free( name );
1038     if (!inode)
1039         return 0;
1040
1041     /* Couldn't find this inode at the start of the function, must be new */
1042     assert( inode->wd == -1 );
1043
1044     wd = inotify_add_dir( link, filter );
1045     if (wd != -1)
1046         inode_set_wd( inode, wd );
1047
1048     return 1;
1049 }
1050
1051 #else
1052
1053 static int init_inotify( void )
1054 {
1055     return 0;
1056 }
1057
1058 static int inotify_adjust_changes( struct dir *dir )
1059 {
1060     return 0;
1061 }
1062
1063 static void free_inode( struct inode *inode )
1064 {
1065     assert( 0 );
1066 }
1067
1068 static int dir_add_to_existing_notify( struct dir *dir )
1069 {
1070     return 0;
1071 }
1072
1073 #endif  /* USE_INOTIFY */
1074
1075 struct object *create_dir_obj( struct fd *fd, unsigned int access, mode_t mode )
1076 {
1077     struct dir *dir;
1078
1079     dir = alloc_object( &dir_ops );
1080     if (!dir)
1081         return NULL;
1082
1083     list_init( &dir->change_records );
1084     dir->filter = 0;
1085     dir->notified = 0;
1086     dir->want_data = 0;
1087     dir->inode = NULL;
1088     grab_object( fd );
1089     dir->fd = fd;
1090     dir->mode = mode;
1091     dir->uid  = ~(uid_t)0;
1092     set_fd_user( fd, &dir_fd_ops, &dir->obj );
1093
1094     dir_add_to_existing_notify( dir );
1095
1096     return &dir->obj;
1097 }
1098
1099 /* enable change notifications for a directory */
1100 DECL_HANDLER(read_directory_changes)
1101 {
1102     struct dir *dir;
1103     struct async *async;
1104
1105     if (!req->filter)
1106     {
1107         set_error(STATUS_INVALID_PARAMETER);
1108         return;
1109     }
1110
1111     dir = get_dir_obj( current->process, req->async.handle, 0 );
1112     if (!dir)
1113         return;
1114
1115     /* requests don't timeout */
1116     if (!(async = fd_queue_async( dir->fd, &req->async, ASYNC_TYPE_WAIT ))) goto end;
1117
1118     /* assign it once */
1119     if (!dir->filter)
1120     {
1121         init_inotify();
1122         insert_change( dir );
1123         dir->filter = req->filter;
1124         dir->subtree = req->subtree;
1125         dir->want_data = req->want_data;
1126     }
1127
1128     /* if there's already a change in the queue, send it */
1129     if (!list_empty( &dir->change_records ))
1130         fd_async_wake_up( dir->fd, ASYNC_TYPE_WAIT, STATUS_ALERTED );
1131
1132     /* setup the real notification */
1133     if (!inotify_adjust_changes( dir ))
1134         dnotify_adjust_changes( dir );
1135
1136     release_object( async );
1137     set_error(STATUS_PENDING);
1138
1139 end:
1140     release_object( dir );
1141 }
1142
1143 DECL_HANDLER(read_change)
1144 {
1145     struct change_record *record, *next;
1146     struct dir *dir;
1147     struct list events;
1148     char *data, *event;
1149     int size = 0;
1150
1151     dir = get_dir_obj( current->process, req->handle, 0 );
1152     if (!dir)
1153         return;
1154
1155     list_init( &events );
1156     list_move_tail( &events, &dir->change_records );
1157     release_object( dir );
1158
1159     if (list_empty( &events ))
1160     {
1161         set_error( STATUS_NO_DATA_DETECTED );
1162         return;
1163     }
1164
1165     LIST_FOR_EACH_ENTRY( record, &events, struct change_record, entry )
1166     {
1167         size += (offsetof(struct filesystem_event, name[record->event.len])
1168                 + sizeof(int)-1) / sizeof(int) * sizeof(int);
1169     }
1170
1171     if (size > get_reply_max_size())
1172         set_error( STATUS_BUFFER_TOO_SMALL );
1173     else if ((data = mem_alloc( size )) != NULL)
1174     {
1175         event = data;
1176         LIST_FOR_EACH_ENTRY( record, &events, struct change_record, entry )
1177         {
1178             data_size_t len = offsetof( struct filesystem_event, name[record->event.len] );
1179
1180             /* FIXME: rename events are sometimes reported as delete/create */
1181             if (record->event.action == FILE_ACTION_RENAMED_OLD_NAME)
1182             {
1183                 struct list *elem = list_next( &events, &record->entry );
1184                 if (elem)
1185                     next = LIST_ENTRY(elem, struct change_record, entry);
1186
1187                 if (elem && next->cookie == record->cookie)
1188                     next->cookie = 0;
1189                 else
1190                     record->event.action = FILE_ACTION_REMOVED;
1191             }
1192             else if (record->event.action == FILE_ACTION_RENAMED_NEW_NAME && record->cookie)
1193                 record->event.action = FILE_ACTION_ADDED;
1194
1195             memcpy( event, &record->event, len );
1196             event += len;
1197             if (len % sizeof(int))
1198             {
1199                 memset( event, 0, sizeof(int) - len % sizeof(int) );
1200                 event += sizeof(int) - len % sizeof(int);
1201             }
1202         }
1203         set_reply_data_ptr( data, size );
1204     }
1205
1206     LIST_FOR_EACH_ENTRY_SAFE( record, next, &events, struct change_record, entry )
1207     {
1208         list_remove( &record->entry );
1209         free( record );
1210     }
1211 }