fs/xfs/xfs_mount.h

   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
   9 struct xlog;
  10 struct xfs_inode;
  11 struct xfs_mru_cache;
  12 struct xfs_ail;
  13 struct xfs_quotainfo;
  14 struct xfs_da_geometry;
  15
  16 /* dynamic preallocation free space thresholds, 5% down to 1% */
  17 enum {
  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  */
  32 enum {
  33         XFS_ERR_METADATA,
  34         XFS_ERR_CLASS_MAX,
  35 };
  36 enum {
  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  */
  52 struct xfs_error_cfg {
  53         struct xfs_kobj kobj;
  54         int             max_retries;
  55         long            retry_timeout;  /* in jiffies, -1 = infinite */
  56 };
  57
  58 /*
  59  * The struct xfsmount layout is optimised to separate read-mostly variables
  60  * from variables that are frequently modified. We put the read-mostly variables
  61  * first, then place all the other variables at the end.
  62  *
  63  * Typically, read-mostly variables are those that are set at mount time and
  64  * never changed again, or only change rarely as a result of things like sysfs
  65  * knobs being tweaked.
  66  */
  67 typedef struct xfs_mount {
  68         struct xfs_sb           m_sb;           /* copy of fs superblock */
  69         struct super_block      *m_super;
  70         struct xfs_ail          *m_ail;         /* fs active log item list */
  71         struct xfs_buf          *m_sb_bp;       /* buffer for superblock */
  72         char                    *m_rtname;      /* realtime device name */
  73         char                    *m_logname;     /* external log device name */
  74         struct xfs_da_geometry  *m_dir_geo;     /* directory block geometry */
  75         struct xfs_da_geometry  *m_attr_geo;    /* attribute block geometry */
  76         struct xlog             *m_log;         /* log specific stuff */
  77         struct xfs_inode        *m_rbmip;       /* pointer to bitmap inode */
  78         struct xfs_inode        *m_rsumip;      /* pointer to summary inode */
  79         struct xfs_inode        *m_rootip;      /* pointer to root directory */
  80         struct xfs_quotainfo    *m_quotainfo;   /* disk quota information */
  81         xfs_buftarg_t           *m_ddev_targp;  /* saves taking the address */
  82         xfs_buftarg_t           *m_logdev_targp;/* ptr to log device */
  83         xfs_buftarg_t           *m_rtdev_targp; /* ptr to rt device */
  84         /*
  85          * Optional cache of rt summary level per bitmap block with the
  86          * invariant that m_rsum_cache[bbno] <= the minimum i for which
  87          * rsum[i][bbno] != 0. Reads and writes are serialized by the rsumip
  88          * inode lock.
  89          */
  90         uint8_t                 *m_rsum_cache;
  91         struct xfs_mru_cache    *m_filestream;  /* per-mount filestream data */
  92         struct workqueue_struct *m_buf_workqueue;
  93         struct workqueue_struct *m_unwritten_workqueue;
  94         struct workqueue_struct *m_cil_workqueue;
  95         struct workqueue_struct *m_reclaim_workqueue;
  96         struct workqueue_struct *m_eofblocks_workqueue;
  97         struct workqueue_struct *m_sync_workqueue;
  98
  99         int                     m_bsize;        /* fs logical block size */
 100         uint8_t                 m_blkbit_log;   /* blocklog + NBBY */
 101         uint8_t                 m_blkbb_log;    /* blocklog - BBSHIFT */
 102         uint8_t                 m_agno_log;     /* log #ag's */
 103         uint8_t                 m_sectbb_log;   /* sectlog - BBSHIFT */
 104         uint                    m_blockmask;    /* sb_blocksize-1 */
 105         uint                    m_blockwsize;   /* sb_blocksize in words */
 106         uint                    m_blockwmask;   /* blockwsize-1 */
 107         uint                    m_alloc_mxr[2]; /* max alloc btree records */
 108         uint                    m_alloc_mnr[2]; /* min alloc btree records */
 109         uint                    m_bmap_dmxr[2]; /* max bmap btree records */
 110         uint                    m_bmap_dmnr[2]; /* min bmap btree records */
 111         uint                    m_rmap_mxr[2];  /* max rmap btree records */
 112         uint                    m_rmap_mnr[2];  /* min rmap btree records */
 113         uint                    m_refc_mxr[2];  /* max refc btree records */
 114         uint                    m_refc_mnr[2];  /* min refc btree records */
 115         uint                    m_ag_maxlevels; /* XFS_AG_MAXLEVELS */
 116         uint                    m_bm_maxlevels[2]; /* XFS_BM_MAXLEVELS */
 117         uint                    m_rmap_maxlevels; /* max rmap btree levels */
 118         uint                    m_refc_maxlevels; /* max refcount btree level */
 119         xfs_extlen_t            m_ag_prealloc_blocks; /* reserved ag blocks */
 120         uint                    m_alloc_set_aside; /* space we can't use */
 121         uint                    m_ag_max_usable; /* max space per AG */
 122         int                     m_dalign;       /* stripe unit */
 123         int                     m_swidth;       /* stripe width */
 124         xfs_agnumber_t          m_maxagi;       /* highest inode alloc group */
 125         uint                    m_allocsize_log;/* min write size log bytes */
 126         uint                    m_allocsize_blocks; /* min write size blocks */
 127         int                     m_logbufs;      /* number of log buffers */
 128         int                     m_logbsize;     /* size of each log buffer */
 129         uint                    m_rsumlevels;   /* rt summary levels */
 130         uint                    m_rsumsize;     /* size of rt summary, bytes */
 131         int                     m_fixedfsid[2]; /* unchanged for life of FS */
 132         uint                    m_qflags;       /* quota status flags */
 133         uint64_t                m_flags;        /* global mount flags */
 134         int64_t                 m_low_space[XFS_LOWSP_MAX];
 135         struct xfs_ino_geometry m_ino_geo;      /* inode geometry */
 136         struct xfs_trans_resv   m_resv;         /* precomputed res values */
 137                                                 /* low free space thresholds */
 138         bool                    m_always_cow;
 139         bool                    m_fail_unmount;
 140         bool                    m_finobt_nores; /* no per-AG finobt resv. */
 141         bool                    m_update_sb;    /* sb needs update in mount */
 142
 143         /*
 144          * Bitsets of per-fs metadata that have been checked and/or are sick.
 145          * Callers must hold m_sb_lock to access these two fields.
 146          */
 147         uint8_t                 m_fs_checked;
 148         uint8_t                 m_fs_sick;
 149         /*
 150          * Bitsets of rt metadata that have been checked and/or are sick.
 151          * Callers must hold m_sb_lock to access this field.
 152          */
 153         uint8_t                 m_rt_checked;
 154         uint8_t                 m_rt_sick;
 155
 156         /*
 157          * End of read-mostly variables. Frequently written variables and locks
 158          * should be placed below this comment from now on. The first variable
 159          * here is marked as cacheline aligned so they it is separated from
 160          * the read-mostly variables.
 161          */
 162
 163         spinlock_t ____cacheline_aligned m_sb_lock; /* sb counter lock */
 164         struct percpu_counter   m_icount;       /* allocated inodes counter */
 165         struct percpu_counter   m_ifree;        /* free inodes counter */
 166         struct percpu_counter   m_fdblocks;     /* free block counter */
 167         /*
 168          * Count of data device blocks reserved for delayed allocations,
 169          * including indlen blocks.  Does not include allocated CoW staging
 170          * extents or anything related to the rt device.
 171          */
 172         struct percpu_counter   m_delalloc_blks;
 173
 174         struct radix_tree_root  m_perag_tree;   /* per-ag accounting info */
 175         spinlock_t              m_perag_lock;   /* lock for m_perag_tree */
 176         uint64_t                m_resblks;      /* total reserved blocks */
 177         uint64_t                m_resblks_avail;/* available reserved blocks */
 178         uint64_t                m_resblks_save; /* reserved blks @ remount,ro */
 179         struct delayed_work     m_reclaim_work; /* background inode reclaim */
 180         struct delayed_work     m_eofblocks_work; /* background eof blocks
 181                                                      trimming */
 182         struct delayed_work     m_cowblocks_work; /* background cow blocks
 183                                                      trimming */
 184         struct xfs_kobj         m_kobj;
 185         struct xfs_kobj         m_error_kobj;
 186         struct xfs_kobj         m_error_meta_kobj;
 187         struct xfs_error_cfg    m_error_cfg[XFS_ERR_CLASS_MAX][XFS_ERR_ERRNO_MAX];
 188         struct xstats           m_stats;        /* per-fs stats */
 189         xfs_agnumber_t          m_agfrotor;     /* last ag where space found */
 190         xfs_agnumber_t          m_agirotor;     /* last ag dir inode alloced */
 191         spinlock_t              m_agirotor_lock;/* .. and lock protecting it */
 192
 193         /*
 194          * Workqueue item so that we can coalesce multiple inode flush attempts
 195          * into a single flush.
 196          */
 197         struct work_struct      m_flush_inodes_work;
 198
 199         /*
 200          * Generation of the filesysyem layout.  This is incremented by each
 201          * growfs, and used by the pNFS server to ensure the client updates
 202          * its view of the block device once it gets a layout that might
 203          * reference the newly added blocks.  Does not need to be persistent
 204          * as long as we only allow file system size increments, but if we
 205          * ever support shrinks it would have to be persisted in addition
 206          * to various other kinds of pain inflicted on the pNFS server.
 207          */
 208         uint32_t                m_generation;
 209         struct mutex            m_growlock;     /* growfs mutex */
 210
 211 #ifdef DEBUG
 212         /*
 213          * Frequency with which errors are injected.  Replaces xfs_etest; the
 214          * value stored in here is the inverse of the frequency with which the
 215          * error triggers.  1 = always, 2 = half the time, etc.
 216          */
 217         unsigned int            *m_errortag;
 218         struct xfs_kobj         m_errortag_kobj;
 219 #endif
 220 } xfs_mount_t;
 221
 222 #define M_IGEO(mp)              (&(mp)->m_ino_geo)
 223
 224 /*
 225  * Flags for m_flags.
 226  */
 227 #define XFS_MOUNT_WSYNC         (1ULL << 0)     /* for nfs - all metadata ops
 228                                                    must be synchronous except
 229                                                    for space allocations */
 230 #define XFS_MOUNT_UNMOUNTING    (1ULL << 1)     /* filesystem is unmounting */
 231 #define XFS_MOUNT_WAS_CLEAN     (1ULL << 3)
 232 #define XFS_MOUNT_FS_SHUTDOWN   (1ULL << 4)     /* atomic stop of all filesystem
 233                                                    operations, typically for
 234                                                    disk errors in metadata */
 235 #define XFS_MOUNT_DISCARD       (1ULL << 5)     /* discard unused blocks */
 236 #define XFS_MOUNT_NOALIGN       (1ULL << 7)     /* turn off stripe alignment
 237                                                    allocations */
 238 #define XFS_MOUNT_ATTR2         (1ULL << 8)     /* allow use of attr2 format */
 239 #define XFS_MOUNT_GRPID         (1ULL << 9)     /* group-ID assigned from directory */
 240 #define XFS_MOUNT_NORECOVERY    (1ULL << 10)    /* no recovery - dirty fs */
 241 #define XFS_MOUNT_ALLOCSIZE     (1ULL << 12)    /* specified allocation size */
 242 #define XFS_MOUNT_SMALL_INUMS   (1ULL << 14)    /* user wants 32bit inodes */
 243 #define XFS_MOUNT_32BITINODES   (1ULL << 15)    /* inode32 allocator active */
 244 #define XFS_MOUNT_NOUUID        (1ULL << 16)    /* ignore uuid during mount */
 245 #define XFS_MOUNT_IKEEP         (1ULL << 18)    /* keep empty inode clusters*/
 246 #define XFS_MOUNT_SWALLOC       (1ULL << 19)    /* turn on stripe width
 247                                                  * allocation */
 248 #define XFS_MOUNT_RDONLY        (1ULL << 20)    /* read-only fs */
 249 #define XFS_MOUNT_DIRSYNC       (1ULL << 21)    /* synchronous directory ops */
 250 #define XFS_MOUNT_LARGEIO       (1ULL << 22)    /* report large preferred
 251                                                  * I/O size in stat() */
 252 #define XFS_MOUNT_FILESTREAMS   (1ULL << 24)    /* enable the filestreams
 253                                                    allocator */
 254 #define XFS_MOUNT_NOATTR2       (1ULL << 25)    /* disable use of attr2 format */
 255 #define XFS_MOUNT_DAX_ALWAYS    (1ULL << 26)
 256 #define XFS_MOUNT_DAX_NEVER     (1ULL << 27)
 257
 258 /*
 259  * Max and min values for mount-option defined I/O
 260  * preallocation sizes.
 261  */
 262 #define XFS_MAX_IO_LOG          30      /* 1G */
 263 #define XFS_MIN_IO_LOG          PAGE_SHIFT
 264
 265 #define XFS_LAST_UNMOUNT_WAS_CLEAN(mp)  \
 266                                 ((mp)->m_flags & XFS_MOUNT_WAS_CLEAN)
 267 #define XFS_FORCED_SHUTDOWN(mp) ((mp)->m_flags & XFS_MOUNT_FS_SHUTDOWN)
 268 void xfs_do_force_shutdown(struct xfs_mount *mp, int flags, char *fname,
 269                 int lnnum);
 270 #define xfs_force_shutdown(m,f) \
 271         xfs_do_force_shutdown(m, f, __FILE__, __LINE__)
 272
 273 #define SHUTDOWN_META_IO_ERROR  0x0001  /* write attempt to metadata failed */
 274 #define SHUTDOWN_LOG_IO_ERROR   0x0002  /* write attempt to the log failed */
 275 #define SHUTDOWN_FORCE_UMOUNT   0x0004  /* shutdown from a forced unmount */
 276 #define SHUTDOWN_CORRUPT_INCORE 0x0008  /* corrupt in-memory data structures */
 277
 278 /*
 279  * Flags for xfs_mountfs
 280  */
 281 #define XFS_MFSI_QUIET          0x40    /* Be silent if mount errors found */
 282
 283 static inline xfs_agnumber_t
 284 xfs_daddr_to_agno(struct xfs_mount *mp, xfs_daddr_t d)
 285 {
 286         xfs_rfsblock_t ld = XFS_BB_TO_FSBT(mp, d);
 287         do_div(ld, mp->m_sb.sb_agblocks);
 288         return (xfs_agnumber_t) ld;
 289 }
 290
 291 static inline xfs_agblock_t
 292 xfs_daddr_to_agbno(struct xfs_mount *mp, xfs_daddr_t d)
 293 {
 294         xfs_rfsblock_t ld = XFS_BB_TO_FSBT(mp, d);
 295         return (xfs_agblock_t) do_div(ld, mp->m_sb.sb_agblocks);
 296 }
 297
 298 /* per-AG block reservation data structures*/
 299 struct xfs_ag_resv {
 300         /* number of blocks originally reserved here */
 301         xfs_extlen_t                    ar_orig_reserved;
 302         /* number of blocks reserved here */
 303         xfs_extlen_t                    ar_reserved;
 304         /* number of blocks originally asked for */
 305         xfs_extlen_t                    ar_asked;
 306 };
 307
 308 /*
 309  * Per-ag incore structure, copies of information in agf and agi, to improve the
 310  * performance of allocation group selection.
 311  */
 312 typedef struct xfs_perag {
 313         struct xfs_mount *pag_mount;    /* owner filesystem */
 314         xfs_agnumber_t  pag_agno;       /* AG this structure belongs to */
 315         atomic_t        pag_ref;        /* perag reference count */
 316         char            pagf_init;      /* this agf's entry is initialized */
 317         char            pagi_init;      /* this agi's entry is initialized */
 318         char            pagf_metadata;  /* the agf is preferred to be metadata */
 319         char            pagi_inodeok;   /* The agi is ok for inodes */
 320         uint8_t         pagf_levels[XFS_BTNUM_AGF];
 321                                         /* # of levels in bno & cnt btree */
 322         bool            pagf_agflreset; /* agfl requires reset before use */
 323         uint32_t        pagf_flcount;   /* count of blocks in freelist */
 324         xfs_extlen_t    pagf_freeblks;  /* total free blocks */
 325         xfs_extlen_t    pagf_longest;   /* longest free space */
 326         uint32_t        pagf_btreeblks; /* # of blocks held in AGF btrees */
 327         xfs_agino_t     pagi_freecount; /* number of free inodes */
 328         xfs_agino_t     pagi_count;     /* number of allocated inodes */
 329
 330         /*
 331          * Inode allocation search lookup optimisation.
 332          * If the pagino matches, the search for new inodes
 333          * doesn't need to search the near ones again straight away
 334          */
 335         xfs_agino_t     pagl_pagino;
 336         xfs_agino_t     pagl_leftrec;
 337         xfs_agino_t     pagl_rightrec;
 338
 339         /*
 340          * Bitsets of per-ag metadata that have been checked and/or are sick.
 341          * Callers should hold pag_state_lock before accessing this field.
 342          */
 343         uint16_t        pag_checked;
 344         uint16_t        pag_sick;
 345         spinlock_t      pag_state_lock;
 346
 347         spinlock_t      pagb_lock;      /* lock for pagb_tree */
 348         struct rb_root  pagb_tree;      /* ordered tree of busy extents */
 349         unsigned int    pagb_gen;       /* generation count for pagb_tree */
 350         wait_queue_head_t pagb_wait;    /* woken when pagb_gen changes */
 351
 352         atomic_t        pagf_fstrms;    /* # of filestreams active in this AG */
 353
 354         spinlock_t      pag_ici_lock;   /* incore inode cache lock */
 355         struct radix_tree_root pag_ici_root;    /* incore inode cache root */
 356         int             pag_ici_reclaimable;    /* reclaimable inodes */
 357         unsigned long   pag_ici_reclaim_cursor; /* reclaim restart point */
 358
 359         /* buffer cache index */
 360         spinlock_t      pag_buf_lock;   /* lock for pag_buf_hash */
 361         struct rhashtable pag_buf_hash;
 362
 363         /* for rcu-safe freeing */
 364         struct rcu_head rcu_head;
 365         int             pagb_count;     /* pagb slots in use */
 366
 367         /* Blocks reserved for all kinds of metadata. */
 368         struct xfs_ag_resv      pag_meta_resv;
 369         /* Blocks reserved for the reverse mapping btree. */
 370         struct xfs_ag_resv      pag_rmapbt_resv;
 371
 372         /* reference count */
 373         uint8_t                 pagf_refcount_level;
 374
 375         /*
 376          * Unlinked inode information.  This incore information reflects
 377          * data stored in the AGI, so callers must hold the AGI buffer lock
 378          * or have some other means to control concurrency.
 379          */
 380         struct rhashtable       pagi_unlinked_hash;
 381 } xfs_perag_t;
 382
 383 static inline struct xfs_ag_resv *
 384 xfs_perag_resv(
 385         struct xfs_perag        *pag,
 386         enum xfs_ag_resv_type   type)
 387 {
 388         switch (type) {
 389         case XFS_AG_RESV_METADATA:
 390                 return &pag->pag_meta_resv;
 391         case XFS_AG_RESV_RMAPBT:
 392                 return &pag->pag_rmapbt_resv;
 393         default:
 394                 return NULL;
 395         }
 396 }
 397
 398 int xfs_buf_hash_init(xfs_perag_t *pag);
 399 void xfs_buf_hash_destroy(xfs_perag_t *pag);
 400
 401 extern void     xfs_uuid_table_free(void);
 402 extern int      xfs_log_sbcount(xfs_mount_t *);
 403 extern uint64_t xfs_default_resblks(xfs_mount_t *mp);
 404 extern int      xfs_mountfs(xfs_mount_t *mp);
 405 extern int      xfs_initialize_perag(xfs_mount_t *mp, xfs_agnumber_t agcount,
 406                                      xfs_agnumber_t *maxagi);
 407 extern void     xfs_unmountfs(xfs_mount_t *);
 408
 409 extern int      xfs_mod_fdblocks(struct xfs_mount *mp, int64_t delta,
 410                                  bool reserved);
 411 extern int      xfs_mod_frextents(struct xfs_mount *mp, int64_t delta);
 412
 413 extern int      xfs_readsb(xfs_mount_t *, int);
 414 extern void     xfs_freesb(xfs_mount_t *);
 415 extern bool     xfs_fs_writable(struct xfs_mount *mp, int level);
 416 extern int      xfs_sb_validate_fsb_count(struct xfs_sb *, uint64_t);
 417
 418 extern int      xfs_dev_is_read_only(struct xfs_mount *, char *);
 419
 420 extern void     xfs_set_low_space_thresholds(struct xfs_mount *);
 421
 422 int     xfs_zero_extent(struct xfs_inode *ip, xfs_fsblock_t start_fsb,
 423                         xfs_off_t count_fsb);
 424
 425 struct xfs_error_cfg * xfs_error_get_cfg(struct xfs_mount *mp,
 426                 int error_class, int error);
 427 void xfs_force_summary_recalc(struct xfs_mount *mp);
 428 void xfs_mod_delalloc(struct xfs_mount *mp, int64_t delta);
 429
 430 #endif  /* __XFS_MOUNT_H__ */