LXR linux/fs/xfs/xfs

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4 * All Rights Reserved.
   5 */
   6#ifndef __XFS_MOUNT_H__
   7#define __XFS_MOUNT_H__
   8
   9struct xlog;
  10struct xfs_inode;
  11struct xfs_mru_cache;
  12struct xfs_ail;
  13struct xfs_quotainfo;
  14struct xfs_da_geometry;
  15
  16/* dynamic preallocation free space thresholds, 5% down to 1% */
  17enum {
  18        XFS_LOWSP_1_PCNT = 0,
  19        XFS_LOWSP_2_PCNT,
  20        XFS_LOWSP_3_PCNT,
  21        XFS_LOWSP_4_PCNT,
  22        XFS_LOWSP_5_PCNT,
  23        XFS_LOWSP_MAX,
  24};
  25
  26/*
  27 * Error Configuration
  28 *
  29 * Error classes define the subsystem the configuration belongs to.
  30 * Error numbers define the errors that are configurable.
  31 */
  32enum {
  33        XFS_ERR_METADATA,
  34        XFS_ERR_CLASS_MAX,
  35};
  36enum {
  37        XFS_ERR_DEFAULT,
  38        XFS_ERR_EIO,
  39        XFS_ERR_ENOSPC,
  40        XFS_ERR_ENODEV,
  41        XFS_ERR_ERRNO_MAX,
  42};
  43
  44#define XFS_ERR_RETRY_FOREVER   -1
  45
  46/*
  47 * Although retry_timeout is in jiffies which is normally an unsigned long,
  48 * we limit the retry timeout to 86400 seconds, or one day.  So even a
  49 * signed 32-bit long is sufficient for a HZ value up to 24855.  Making it
  50 * signed lets us store the special "-1" value, meaning retry forever.
  51 */
  52struct xfs_error_cfg {
  53        struct xfs_kobj kobj;
  54        int             max_retries;
  55        long            retry_timeout;  /* in jiffies, -1 = infinite */
  56};
  57
  58typedef struct xfs_mount {
  59        struct super_block      *m_super;
  60
  61        /*
  62         * Bitsets of per-fs metadata that have been checked and/or are sick.
  63         * Callers must hold m_sb_lock to access these two fields.
  64         */
  65        uint8_t                 m_fs_checked;
  66        uint8_t                 m_fs_sick;
  67        /*
  68         * Bitsets of rt metadata that have been checked and/or are sick.
  69         * Callers must hold m_sb_lock to access this field.
  70         */
  71        uint8_t                 m_rt_checked;
  72        uint8_t                 m_rt_sick;
  73
  74        struct xfs_ail          *m_ail;         /* fs active log item list */
  75
  76        struct xfs_sb           m_sb;           /* copy of fs superblock */
  77        spinlock_t              m_sb_lock;      /* sb counter lock */
  78        struct percpu_counter   m_icount;       /* allocated inodes counter */
  79        struct percpu_counter   m_ifree;        /* free inodes counter */
  80        struct percpu_counter   m_fdblocks;     /* free block counter */
  81        /*
  82         * Count of data device blocks reserved for delayed allocations,
  83         * including indlen blocks.  Does not include allocated CoW staging
  84         * extents or anything related to the rt device.
  85         */
  86        struct percpu_counter   m_delalloc_blks;
  87
  88        struct xfs_buf          *m_sb_bp;       /* buffer for superblock */
  89        char                    *m_rtname;      /* realtime device name */
  90        char                    *m_logname;     /* external log device name */
  91        int                     m_bsize;        /* fs logical block size */
  92        xfs_agnumber_t          m_agfrotor;     /* last ag where space found */
  93        xfs_agnumber_t          m_agirotor;     /* last ag dir inode alloced */
  94        spinlock_t              m_agirotor_lock;/* .. and lock protecting it */
  95        xfs_agnumber_t          m_maxagi;       /* highest inode alloc group */
  96        uint                    m_allocsize_log;/* min write size log bytes */
  97        uint                    m_allocsize_blocks; /* min write size blocks */
  98        struct xfs_da_geometry  *m_dir_geo;     /* directory block geometry */
  99        struct xfs_da_geometry  *m_attr_geo;    /* attribute block geometry */
 100        struct xlog             *m_log;         /* log specific stuff */
 101        struct xfs_ino_geometry m_ino_geo;      /* inode geometry */
 102        int                     m_logbufs;      /* number of log buffers */
 103        int                     m_logbsize;     /* size of each log buffer */
 104        uint                    m_rsumlevels;   /* rt summary levels */
 105        uint                    m_rsumsize;     /* size of rt summary, bytes */
 106        /*
 107         * Optional cache of rt summary level per bitmap block with the
 108         * invariant that m_rsum_cache[bbno] <= the minimum i for which
 109         * rsum[i][bbno] != 0. Reads and writes are serialized by the rsumip
 110         * inode lock.
 111         */
 112        uint8_t                 *m_rsum_cache;
 113        struct xfs_inode        *m_rbmip;       /* pointer to bitmap inode */
 114        struct xfs_inode        *m_rsumip;      /* pointer to summary inode */
 115        struct xfs_inode        *m_rootip;      /* pointer to root directory */
 116        struct xfs_quotainfo    *m_quotainfo;   /* disk quota information */
 117        xfs_buftarg_t           *m_ddev_targp;  /* saves taking the address */
 118        xfs_buftarg_t           *m_logdev_targp;/* ptr to log device */
 119        xfs_buftarg_t           *m_rtdev_targp; /* ptr to rt device */
 120        uint8_t                 m_blkbit_log;   /* blocklog + NBBY */
 121        uint8_t                 m_blkbb_log;    /* blocklog - BBSHIFT */
 122        uint8_t                 m_agno_log;     /* log #ag's */
 123        uint                    m_blockmask;    /* sb_blocksize-1 */
 124        uint                    m_blockwsize;   /* sb_blocksize in words */
 125        uint                    m_blockwmask;   /* blockwsize-1 */
 126        uint                    m_alloc_mxr[2]; /* max alloc btree records */
 127        uint                    m_alloc_mnr[2]; /* min alloc btree records */
 128        uint                    m_bmap_dmxr[2]; /* max bmap btree records */
 129        uint                    m_bmap_dmnr[2]; /* min bmap btree records */
 130        uint                    m_rmap_mxr[2];  /* max rmap btree records */
 131        uint                    m_rmap_mnr[2];  /* min rmap btree records */
 132        uint                    m_refc_mxr[2];  /* max refc btree records */
 133        uint                    m_refc_mnr[2];  /* min refc btree records */
 134        uint                    m_ag_maxlevels; /* XFS_AG_MAXLEVELS */
 135        uint                    m_bm_maxlevels[2]; /* XFS_BM_MAXLEVELS */
 136        uint                    m_rmap_maxlevels; /* max rmap btree levels */
 137        uint                    m_refc_maxlevels; /* max refcount btree level */
 138        xfs_extlen_t            m_ag_prealloc_blocks; /* reserved ag blocks */
 139        uint                    m_alloc_set_aside; /* space we can't use */
 140        uint                    m_ag_max_usable; /* max space per AG */
 141        struct radix_tree_root  m_perag_tree;   /* per-ag accounting info */
 142        spinlock_t              m_perag_lock;   /* lock for m_perag_tree */
 143        struct mutex            m_growlock;     /* growfs mutex */
 144        int                     m_fixedfsid[2]; /* unchanged for life of FS */
 145        uint64_t                m_flags;        /* global mount flags */
 146        bool                    m_finobt_nores; /* no per-AG finobt resv. */
 147        uint                    m_qflags;       /* quota status flags */
 148        struct xfs_trans_resv   m_resv;         /* precomputed res values */
 149        uint64_t                m_resblks;      /* total reserved blocks */
 150        uint64_t                m_resblks_avail;/* available reserved blocks */
 151        uint64_t                m_resblks_save; /* reserved blks @ remount,ro */
 152        int                     m_dalign;       /* stripe unit */
 153        int                     m_swidth;       /* stripe width */
 154        uint8_t                 m_sectbb_log;   /* sectlog - BBSHIFT */
 155        atomic_t                m_active_trans; /* number trans frozen */
 156        struct xfs_mru_cache    *m_filestream;  /* per-mount filestream data */
 157        struct delayed_work     m_reclaim_work; /* background inode reclaim */
 158        struct delayed_work     m_eofblocks_work; /* background eof blocks
 159                                                     trimming */
 160        struct delayed_work     m_cowblocks_work; /* background cow blocks
 161                                                     trimming */
 162        bool                    m_update_sb;    /* sb needs update in mount */
 163        int64_t                 m_low_space[XFS_LOWSP_MAX];
 164                                                /* low free space thresholds */
 165        struct xfs_kobj         m_kobj;
 166        struct xfs_kobj         m_error_kobj;
 167        struct xfs_kobj         m_error_meta_kobj;
 168        struct xfs_error_cfg    m_error_cfg[XFS_ERR_CLASS_MAX][XFS_ERR_ERRNO_MAX];
 169        struct xstats           m_stats;        /* per-fs stats */
 170
 171        struct workqueue_struct *m_buf_workqueue;
 172        struct workqueue_struct *m_unwritten_workqueue;
 173        struct workqueue_struct *m_cil_workqueue;
 174        struct workqueue_struct *m_reclaim_workqueue;
 175        struct workqueue_struct *m_eofblocks_workqueue;
 176        struct workqueue_struct *m_sync_workqueue;
 177
 178        /*
 179         * Generation of the filesysyem layout.  This is incremented by each
 180         * growfs, and used by the pNFS server to ensure the client updates
 181         * its view of the block device once it gets a layout that might
 182         * reference the newly added blocks.  Does not need to be persistent
 183         * as long as we only allow file system size increments, but if we
 184         * ever support shrinks it would have to be persisted in addition
 185         * to various other kinds of pain inflicted on the pNFS server.
 186         */
 187        uint32_t                m_generation;
 188
 189        bool                    m_always_cow;
 190        bool                    m_fail_unmount;
 191#ifdef DEBUG
 192        /*
 193         * Frequency with which errors are injected.  Replaces xfs_etest; the
 194         * value stored in here is the inverse of the frequency with which the
 195         * error triggers.  1 = always, 2 = half the time, etc.
 196         */
 197        unsigned int            *m_errortag;
 198        struct xfs_kobj         m_errortag_kobj;
 199#endif
 200} xfs_mount_t;
 201
 202#define M_IGEO(mp)              (&(mp)->m_ino_geo)
 203
 204/*
 205 * Flags for m_flags.
 206 */
 207#define XFS_MOUNT_WSYNC         (1ULL << 0)     /* for nfs - all metadata ops
 208                                                   must be synchronous except
 209                                                   for space allocations */
 210#define XFS_MOUNT_UNMOUNTING    (1ULL << 1)     /* filesystem is unmounting */
 211#define XFS_MOUNT_WAS_CLEAN     (1ULL << 3)
 212#define XFS_MOUNT_FS_SHUTDOWN   (1ULL << 4)     /* atomic stop of all filesystem
 213                                                   operations, typically for
 214                                                   disk errors in metadata */
 215#define XFS_MOUNT_DISCARD       (1ULL << 5)     /* discard unused blocks */
 216#define XFS_MOUNT_NOALIGN       (1ULL << 7)     /* turn off stripe alignment
 217                                                   allocations */
 218#define XFS_MOUNT_ATTR2         (1ULL << 8)     /* allow use of attr2 format */
 219#define XFS_MOUNT_GRPID         (1ULL << 9)     /* group-ID assigned from directory */
 220#define XFS_MOUNT_NORECOVERY    (1ULL << 10)    /* no recovery - dirty fs */
 221#define XFS_MOUNT_ALLOCSIZE     (1ULL << 12)    /* specified allocation size */
 222#define XFS_MOUNT_SMALL_INUMS   (1ULL << 14)    /* user wants 32bit inodes */
 223#define XFS_MOUNT_32BITINODES   (1ULL << 15)    /* inode32 allocator active */
 224#define XFS_MOUNT_NOUUID        (1ULL << 16)    /* ignore uuid during mount */
 225#define XFS_MOUNT_IKEEP         (1ULL << 18)    /* keep empty inode clusters*/
 226#define XFS_MOUNT_SWALLOC       (1ULL << 19)    /* turn on stripe width
 227                                                 * allocation */
 228#define XFS_MOUNT_RDONLY        (1ULL << 20)    /* read-only fs */
 229#define XFS_MOUNT_DIRSYNC       (1ULL << 21)    /* synchronous directory ops */
 230#define XFS_MOUNT_LARGEIO       (1ULL << 22)    /* report large preferred
 231                                                 * I/O size in stat() */
 232#define XFS_MOUNT_FILESTREAMS   (1ULL << 24)    /* enable the filestreams
 233                                                   allocator */
 234#define XFS_MOUNT_NOATTR2       (1ULL << 25)    /* disable use of attr2 format */
 235#define XFS_MOUNT_DAX_ALWAYS    (1ULL << 26)
 236#define XFS_MOUNT_DAX_NEVER     (1ULL << 27)
 237
 238/*
 239 * Max and min values for mount-option defined I/O
 240 * preallocation sizes.
 241 */
 242#define XFS_MAX_IO_LOG          30      /* 1G */
 243#define XFS_MIN_IO_LOG          PAGE_SHIFT
 244
 245#define XFS_LAST_UNMOUNT_WAS_CLEAN(mp)  \
 246                                ((mp)->m_flags & XFS_MOUNT_WAS_CLEAN)
 247#define XFS_FORCED_SHUTDOWN(mp) ((mp)->m_flags & XFS_MOUNT_FS_SHUTDOWN)
 248void xfs_do_force_shutdown(struct xfs_mount *mp, int flags, char *fname,
 249                int lnnum);
 250#define xfs_force_shutdown(m,f) \
 251        xfs_do_force_shutdown(m, f, __FILE__, __LINE__)
 252
 253#define SHUTDOWN_META_IO_ERROR  0x0001  /* write attempt to metadata failed */
 254#define SHUTDOWN_LOG_IO_ERROR   0x0002  /* write attempt to the log failed */
 255#define SHUTDOWN_FORCE_UMOUNT   0x0004  /* shutdown from a forced unmount */
 256#define SHUTDOWN_CORRUPT_INCORE 0x0008  /* corrupt in-memory data structures */
 257#define SHUTDOWN_REMOTE_REQ     0x0010  /* shutdown came from remote cell */
 258#define SHUTDOWN_DEVICE_REQ     0x0020  /* failed all paths to the device */
 259
 260/*
 261 * Flags for xfs_mountfs
 262 */
 263#define XFS_MFSI_QUIET          0x40    /* Be silent if mount errors found */
 264
 265static inline xfs_agnumber_t
 266xfs_daddr_to_agno(struct xfs_mount *mp, xfs_daddr_t d)
 267{
 268        xfs_rfsblock_t ld = XFS_BB_TO_FSBT(mp, d);
 269        do_div(ld, mp->m_sb.sb_agblocks);
 270        return (xfs_agnumber_t) ld;
 271}
 272
 273static inline xfs_agblock_t
 274xfs_daddr_to_agbno(struct xfs_mount *mp, xfs_daddr_t d)
 275{
 276        xfs_rfsblock_t ld = XFS_BB_TO_FSBT(mp, d);
 277        return (xfs_agblock_t) do_div(ld, mp->m_sb.sb_agblocks);
 278}
 279
 280/* per-AG block reservation data structures*/
 281struct xfs_ag_resv {
 282        /* number of blocks originally reserved here */
 283        xfs_extlen_t                    ar_orig_reserved;
 284        /* number of blocks reserved here */
 285        xfs_extlen_t                    ar_reserved;
 286        /* number of blocks originally asked for */
 287        xfs_extlen_t                    ar_asked;
 288};
 289
 290/*
 291 * Per-ag incore structure, copies of information in agf and agi, to improve the
 292 * performance of allocation group selection.
 293 */
 294typedef struct xfs_perag {
 295        struct xfs_mount *pag_mount;    /* owner filesystem */
 296        xfs_agnumber_t  pag_agno;       /* AG this structure belongs to */
 297        atomic_t        pag_ref;        /* perag reference count */
 298        char            pagf_init;      /* this agf's entry is initialized */
 299        char            pagi_init;      /* this agi's entry is initialized */
 300        char            pagf_metadata;  /* the agf is preferred to be metadata */
 301        char            pagi_inodeok;   /* The agi is ok for inodes */
 302        uint8_t         pagf_levels[XFS_BTNUM_AGF];
 303                                        /* # of levels in bno & cnt btree */
 304        bool            pagf_agflreset; /* agfl requires reset before use */
 305        uint32_t        pagf_flcount;   /* count of blocks in freelist */
 306        xfs_extlen_t    pagf_freeblks;  /* total free blocks */
 307        xfs_extlen_t    pagf_longest;   /* longest free space */
 308        uint32_t        pagf_btreeblks; /* # of blocks held in AGF btrees */
 309        xfs_agino_t     pagi_freecount; /* number of free inodes */
 310        xfs_agino_t     pagi_count;     /* number of allocated inodes */
 311
 312        /*
 313         * Inode allocation search lookup optimisation.
 314         * If the pagino matches, the search for new inodes
 315         * doesn't need to search the near ones again straight away
 316         */
 317        xfs_agino_t     pagl_pagino;
 318        xfs_agino_t     pagl_leftrec;
 319        xfs_agino_t     pagl_rightrec;
 320
 321        /*
 322         * Bitsets of per-ag metadata that have been checked and/or are sick.
 323         * Callers should hold pag_state_lock before accessing this field.
 324         */
 325        uint16_t        pag_checked;
 326        uint16_t        pag_sick;
 327        spinlock_t      pag_state_lock;
 328
 329        spinlock_t      pagb_lock;      /* lock for pagb_tree */
 330        struct rb_root  pagb_tree;      /* ordered tree of busy extents */
 331        unsigned int    pagb_gen;       /* generation count for pagb_tree */
 332        wait_queue_head_t pagb_wait;    /* woken when pagb_gen changes */
 333
 334        atomic_t        pagf_fstrms;    /* # of filestreams active in this AG */
 335
 336        spinlock_t      pag_ici_lock;   /* incore inode cache lock */
 337        struct radix_tree_root pag_ici_root;    /* incore inode cache root */
 338        int             pag_ici_reclaimable;    /* reclaimable inodes */
 339        struct mutex    pag_ici_reclaim_lock;   /* serialisation point */
 340        unsigned long   pag_ici_reclaim_cursor; /* reclaim restart point */
 341
 342        /* buffer cache index */
 343        spinlock_t      pag_buf_lock;   /* lock for pag_buf_hash */
 344        struct rhashtable pag_buf_hash;
 345
 346        /* for rcu-safe freeing */
 347        struct rcu_head rcu_head;
 348        int             pagb_count;     /* pagb slots in use */
 349
 350        /* Blocks reserved for all kinds of metadata. */
 351        struct xfs_ag_resv      pag_meta_resv;
 352        /* Blocks reserved for the reverse mapping btree. */
 353        struct xfs_ag_resv      pag_rmapbt_resv;
 354
 355        /* reference count */
 356        uint8_t                 pagf_refcount_level;
 357
 358        /*
 359         * Unlinked inode information.  This incore information reflects
 360         * data stored in the AGI, so callers must hold the AGI buffer lock
 361         * or have some other means to control concurrency.
 362         */
 363        struct rhashtable       pagi_unlinked_hash;
 364} xfs_perag_t;
 365
 366static inline struct xfs_ag_resv *
 367xfs_perag_resv(
 368        struct xfs_perag        *pag,
 369        enum xfs_ag_resv_type   type)
 370{
 371        switch (type) {
 372        case XFS_AG_RESV_METADATA:
 373                return &pag->pag_meta_resv;
 374        case XFS_AG_RESV_RMAPBT:
 375                return &pag->pag_rmapbt_resv;
 376        default:
 377                return NULL;
 378        }
 379}
 380
 381int xfs_buf_hash_init(xfs_perag_t *pag);
 382void xfs_buf_hash_destroy(xfs_perag_t *pag);
 383
 384extern void     xfs_uuid_table_free(void);
 385extern int      xfs_log_sbcount(xfs_mount_t *);
 386extern uint64_t xfs_default_resblks(xfs_mount_t *mp);
 387extern int      xfs_mountfs(xfs_mount_t *mp);
 388extern int      xfs_initialize_perag(xfs_mount_t *mp, xfs_agnumber_t agcount,
 389                                     xfs_agnumber_t *maxagi);
 390extern void     xfs_unmountfs(xfs_mount_t *);
 391
 392extern int      xfs_mod_icount(struct xfs_mount *mp, int64_t delta);
 393extern int      xfs_mod_ifree(struct xfs_mount *mp, int64_t delta);
 394extern int      xfs_mod_fdblocks(struct xfs_mount *mp, int64_t delta,
 395                                 bool reserved);
 396extern int      xfs_mod_frextents(struct xfs_mount *mp, int64_t delta);
 397
 398extern struct xfs_buf *xfs_getsb(xfs_mount_t *);
 399extern int      xfs_readsb(xfs_mount_t *, int);
 400extern void     xfs_freesb(xfs_mount_t *);
 401extern bool     xfs_fs_writable(struct xfs_mount *mp, int level);
 402extern int      xfs_sb_validate_fsb_count(struct xfs_sb *, uint64_t);
 403
 404extern int      xfs_dev_is_read_only(struct xfs_mount *, char *);
 405
 406extern void     xfs_set_low_space_thresholds(struct xfs_mount *);
 407
 408int     xfs_zero_extent(struct xfs_inode *ip, xfs_fsblock_t start_fsb,
 409                        xfs_off_t count_fsb);
 410
 411struct xfs_error_cfg * xfs_error_get_cfg(struct xfs_mount *mp,
 412                int error_class, int error);
 413void xfs_force_summary_recalc(struct xfs_mount *mp);
 414void xfs_mod_delalloc(struct xfs_mount *mp, int64_t delta);
 415
 416#endif  /* __XFS_MOUNT_H__ */
 417