Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 3 | * Copyright 2004-2011 Red Hat, Inc. |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 4 | * |
| 5 | * This copyrighted material is made available to anyone wishing to use, |
| 6 | * modify, copy, or redistribute it subject to the terms and conditions |
| 7 | * of the GNU General Public License version 2. |
| 8 | */ |
| 9 | |
| 10 | #include <linux/fs.h> |
| 11 | #include <linux/dlm.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 12 | #include <linux/slab.h> |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 13 | #include <linux/types.h> |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 14 | #include <linux/delay.h> |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 15 | #include <linux/gfs2_ondisk.h> |
| 16 | |
| 17 | #include "incore.h" |
| 18 | #include "glock.h" |
| 19 | #include "util.h" |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 20 | #include "sys.h" |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 21 | |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 22 | extern struct workqueue_struct *gfs2_control_wq; |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 23 | |
| 24 | static void gdlm_ast(void *arg) |
| 25 | { |
| 26 | struct gfs2_glock *gl = arg; |
| 27 | unsigned ret = gl->gl_state; |
| 28 | |
| 29 | BUG_ON(gl->gl_lksb.sb_flags & DLM_SBF_DEMOTED); |
| 30 | |
| 31 | if (gl->gl_lksb.sb_flags & DLM_SBF_VALNOTVALID) |
| 32 | memset(gl->gl_lvb, 0, GDLM_LVB_SIZE); |
| 33 | |
| 34 | switch (gl->gl_lksb.sb_status) { |
| 35 | case -DLM_EUNLOCK: /* Unlocked, so glock can be freed */ |
Steven Whitehouse | fc0e38d | 2011-03-09 10:58:04 +0000 | [diff] [blame] | 36 | gfs2_glock_free(gl); |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 37 | return; |
| 38 | case -DLM_ECANCEL: /* Cancel while getting lock */ |
| 39 | ret |= LM_OUT_CANCELED; |
| 40 | goto out; |
| 41 | case -EAGAIN: /* Try lock fails */ |
Steven Whitehouse | 1fea7c2 | 2010-09-08 10:09:25 +0100 | [diff] [blame] | 42 | case -EDEADLK: /* Deadlock detected */ |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 43 | goto out; |
Steven Whitehouse | 1fea7c2 | 2010-09-08 10:09:25 +0100 | [diff] [blame] | 44 | case -ETIMEDOUT: /* Canceled due to timeout */ |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 45 | ret |= LM_OUT_ERROR; |
| 46 | goto out; |
| 47 | case 0: /* Success */ |
| 48 | break; |
| 49 | default: /* Something unexpected */ |
| 50 | BUG(); |
| 51 | } |
| 52 | |
Benjamin Marzinski | 02ffad08 | 2009-03-06 10:03:20 -0600 | [diff] [blame] | 53 | ret = gl->gl_req; |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 54 | if (gl->gl_lksb.sb_flags & DLM_SBF_ALTMODE) { |
Benjamin Marzinski | 02ffad08 | 2009-03-06 10:03:20 -0600 | [diff] [blame] | 55 | if (gl->gl_req == LM_ST_SHARED) |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 56 | ret = LM_ST_DEFERRED; |
Benjamin Marzinski | 02ffad08 | 2009-03-06 10:03:20 -0600 | [diff] [blame] | 57 | else if (gl->gl_req == LM_ST_DEFERRED) |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 58 | ret = LM_ST_SHARED; |
| 59 | else |
| 60 | BUG(); |
| 61 | } |
| 62 | |
| 63 | set_bit(GLF_INITIAL, &gl->gl_flags); |
| 64 | gfs2_glock_complete(gl, ret); |
| 65 | return; |
| 66 | out: |
| 67 | if (!test_bit(GLF_INITIAL, &gl->gl_flags)) |
| 68 | gl->gl_lksb.sb_lkid = 0; |
| 69 | gfs2_glock_complete(gl, ret); |
| 70 | } |
| 71 | |
| 72 | static void gdlm_bast(void *arg, int mode) |
| 73 | { |
| 74 | struct gfs2_glock *gl = arg; |
| 75 | |
| 76 | switch (mode) { |
| 77 | case DLM_LOCK_EX: |
| 78 | gfs2_glock_cb(gl, LM_ST_UNLOCKED); |
| 79 | break; |
| 80 | case DLM_LOCK_CW: |
| 81 | gfs2_glock_cb(gl, LM_ST_DEFERRED); |
| 82 | break; |
| 83 | case DLM_LOCK_PR: |
| 84 | gfs2_glock_cb(gl, LM_ST_SHARED); |
| 85 | break; |
| 86 | default: |
| 87 | printk(KERN_ERR "unknown bast mode %d", mode); |
| 88 | BUG(); |
| 89 | } |
| 90 | } |
| 91 | |
| 92 | /* convert gfs lock-state to dlm lock-mode */ |
| 93 | |
| 94 | static int make_mode(const unsigned int lmstate) |
| 95 | { |
| 96 | switch (lmstate) { |
| 97 | case LM_ST_UNLOCKED: |
| 98 | return DLM_LOCK_NL; |
| 99 | case LM_ST_EXCLUSIVE: |
| 100 | return DLM_LOCK_EX; |
| 101 | case LM_ST_DEFERRED: |
| 102 | return DLM_LOCK_CW; |
| 103 | case LM_ST_SHARED: |
| 104 | return DLM_LOCK_PR; |
| 105 | } |
| 106 | printk(KERN_ERR "unknown LM state %d", lmstate); |
| 107 | BUG(); |
| 108 | return -1; |
| 109 | } |
| 110 | |
| 111 | static u32 make_flags(const u32 lkid, const unsigned int gfs_flags, |
| 112 | const int req) |
| 113 | { |
| 114 | u32 lkf = 0; |
| 115 | |
| 116 | if (gfs_flags & LM_FLAG_TRY) |
| 117 | lkf |= DLM_LKF_NOQUEUE; |
| 118 | |
| 119 | if (gfs_flags & LM_FLAG_TRY_1CB) { |
| 120 | lkf |= DLM_LKF_NOQUEUE; |
| 121 | lkf |= DLM_LKF_NOQUEUEBAST; |
| 122 | } |
| 123 | |
| 124 | if (gfs_flags & LM_FLAG_PRIORITY) { |
| 125 | lkf |= DLM_LKF_NOORDER; |
| 126 | lkf |= DLM_LKF_HEADQUE; |
| 127 | } |
| 128 | |
| 129 | if (gfs_flags & LM_FLAG_ANY) { |
| 130 | if (req == DLM_LOCK_PR) |
| 131 | lkf |= DLM_LKF_ALTCW; |
| 132 | else if (req == DLM_LOCK_CW) |
| 133 | lkf |= DLM_LKF_ALTPR; |
| 134 | else |
| 135 | BUG(); |
| 136 | } |
| 137 | |
| 138 | if (lkid != 0) |
| 139 | lkf |= DLM_LKF_CONVERT; |
| 140 | |
| 141 | lkf |= DLM_LKF_VALBLK; |
| 142 | |
| 143 | return lkf; |
| 144 | } |
| 145 | |
Steven Whitehouse | 921169c | 2010-11-29 12:50:38 +0000 | [diff] [blame] | 146 | static int gdlm_lock(struct gfs2_glock *gl, unsigned int req_state, |
| 147 | unsigned int flags) |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 148 | { |
| 149 | struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct; |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 150 | int req; |
| 151 | u32 lkf; |
| 152 | |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 153 | req = make_mode(req_state); |
| 154 | lkf = make_flags(gl->gl_lksb.sb_lkid, flags, req); |
| 155 | |
| 156 | /* |
| 157 | * Submit the actual lock request. |
| 158 | */ |
| 159 | |
Steven Whitehouse | 921169c | 2010-11-29 12:50:38 +0000 | [diff] [blame] | 160 | return dlm_lock(ls->ls_dlm, req, &gl->gl_lksb, lkf, gl->gl_strname, |
| 161 | GDLM_STRNAME_BYTES - 1, 0, gdlm_ast, gl, gdlm_bast); |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 162 | } |
| 163 | |
Steven Whitehouse | bc015cb | 2011-01-19 09:30:01 +0000 | [diff] [blame] | 164 | static void gdlm_put_lock(struct gfs2_glock *gl) |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 165 | { |
Steven Whitehouse | e402746 | 2010-01-25 11:20:19 +0000 | [diff] [blame] | 166 | struct gfs2_sbd *sdp = gl->gl_sbd; |
| 167 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 168 | int error; |
| 169 | |
| 170 | if (gl->gl_lksb.sb_lkid == 0) { |
Steven Whitehouse | fc0e38d | 2011-03-09 10:58:04 +0000 | [diff] [blame] | 171 | gfs2_glock_free(gl); |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 172 | return; |
| 173 | } |
| 174 | |
| 175 | error = dlm_unlock(ls->ls_dlm, gl->gl_lksb.sb_lkid, DLM_LKF_VALBLK, |
| 176 | NULL, gl); |
| 177 | if (error) { |
| 178 | printk(KERN_ERR "gdlm_unlock %x,%llx err=%d\n", |
| 179 | gl->gl_name.ln_type, |
| 180 | (unsigned long long)gl->gl_name.ln_number, error); |
| 181 | return; |
| 182 | } |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 183 | } |
| 184 | |
| 185 | static void gdlm_cancel(struct gfs2_glock *gl) |
| 186 | { |
| 187 | struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct; |
| 188 | dlm_unlock(ls->ls_dlm, gl->gl_lksb.sb_lkid, DLM_LKF_CANCEL, NULL, gl); |
| 189 | } |
| 190 | |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 191 | /* |
| 192 | * dlm/gfs2 recovery coordination using dlm_recover callbacks |
| 193 | * |
| 194 | * 1. dlm_controld sees lockspace members change |
| 195 | * 2. dlm_controld blocks dlm-kernel locking activity |
| 196 | * 3. dlm_controld within dlm-kernel notifies gfs2 (recover_prep) |
| 197 | * 4. dlm_controld starts and finishes its own user level recovery |
| 198 | * 5. dlm_controld starts dlm-kernel dlm_recoverd to do kernel recovery |
| 199 | * 6. dlm_recoverd notifies gfs2 of failed nodes (recover_slot) |
| 200 | * 7. dlm_recoverd does its own lock recovery |
| 201 | * 8. dlm_recoverd unblocks dlm-kernel locking activity |
| 202 | * 9. dlm_recoverd notifies gfs2 when done (recover_done with new generation) |
| 203 | * 10. gfs2_control updates control_lock lvb with new generation and jid bits |
| 204 | * 11. gfs2_control enqueues journals for gfs2_recover to recover (maybe none) |
| 205 | * 12. gfs2_recover dequeues and recovers journals of failed nodes |
| 206 | * 13. gfs2_recover provides recovery results to gfs2_control (recovery_result) |
| 207 | * 14. gfs2_control updates control_lock lvb jid bits for recovered journals |
| 208 | * 15. gfs2_control unblocks normal locking when all journals are recovered |
| 209 | * |
| 210 | * - failures during recovery |
| 211 | * |
| 212 | * recover_prep() may set BLOCK_LOCKS (step 3) again before gfs2_control |
| 213 | * clears BLOCK_LOCKS (step 15), e.g. another node fails while still |
| 214 | * recovering for a prior failure. gfs2_control needs a way to detect |
| 215 | * this so it can leave BLOCK_LOCKS set in step 15. This is managed using |
| 216 | * the recover_block and recover_start values. |
| 217 | * |
| 218 | * recover_done() provides a new lockspace generation number each time it |
| 219 | * is called (step 9). This generation number is saved as recover_start. |
| 220 | * When recover_prep() is called, it sets BLOCK_LOCKS and sets |
| 221 | * recover_block = recover_start. So, while recover_block is equal to |
| 222 | * recover_start, BLOCK_LOCKS should remain set. (recover_spin must |
| 223 | * be held around the BLOCK_LOCKS/recover_block/recover_start logic.) |
| 224 | * |
| 225 | * - more specific gfs2 steps in sequence above |
| 226 | * |
| 227 | * 3. recover_prep sets BLOCK_LOCKS and sets recover_block = recover_start |
| 228 | * 6. recover_slot records any failed jids (maybe none) |
| 229 | * 9. recover_done sets recover_start = new generation number |
| 230 | * 10. gfs2_control sets control_lock lvb = new gen + bits for failed jids |
| 231 | * 12. gfs2_recover does journal recoveries for failed jids identified above |
| 232 | * 14. gfs2_control clears control_lock lvb bits for recovered jids |
| 233 | * 15. gfs2_control checks if recover_block == recover_start (step 3 occured |
| 234 | * again) then do nothing, otherwise if recover_start > recover_block |
| 235 | * then clear BLOCK_LOCKS. |
| 236 | * |
| 237 | * - parallel recovery steps across all nodes |
| 238 | * |
| 239 | * All nodes attempt to update the control_lock lvb with the new generation |
| 240 | * number and jid bits, but only the first to get the control_lock EX will |
| 241 | * do so; others will see that it's already done (lvb already contains new |
| 242 | * generation number.) |
| 243 | * |
| 244 | * . All nodes get the same recover_prep/recover_slot/recover_done callbacks |
| 245 | * . All nodes attempt to set control_lock lvb gen + bits for the new gen |
| 246 | * . One node gets control_lock first and writes the lvb, others see it's done |
| 247 | * . All nodes attempt to recover jids for which they see control_lock bits set |
| 248 | * . One node succeeds for a jid, and that one clears the jid bit in the lvb |
| 249 | * . All nodes will eventually see all lvb bits clear and unblock locks |
| 250 | * |
| 251 | * - is there a problem with clearing an lvb bit that should be set |
| 252 | * and missing a journal recovery? |
| 253 | * |
| 254 | * 1. jid fails |
| 255 | * 2. lvb bit set for step 1 |
| 256 | * 3. jid recovered for step 1 |
| 257 | * 4. jid taken again (new mount) |
| 258 | * 5. jid fails (for step 4) |
| 259 | * 6. lvb bit set for step 5 (will already be set) |
| 260 | * 7. lvb bit cleared for step 3 |
| 261 | * |
| 262 | * This is not a problem because the failure in step 5 does not |
| 263 | * require recovery, because the mount in step 4 could not have |
| 264 | * progressed far enough to unblock locks and access the fs. The |
| 265 | * control_mount() function waits for all recoveries to be complete |
| 266 | * for the latest lockspace generation before ever unblocking locks |
| 267 | * and returning. The mount in step 4 waits until the recovery in |
| 268 | * step 1 is done. |
| 269 | * |
| 270 | * - special case of first mounter: first node to mount the fs |
| 271 | * |
| 272 | * The first node to mount a gfs2 fs needs to check all the journals |
| 273 | * and recover any that need recovery before other nodes are allowed |
| 274 | * to mount the fs. (Others may begin mounting, but they must wait |
| 275 | * for the first mounter to be done before taking locks on the fs |
| 276 | * or accessing the fs.) This has two parts: |
| 277 | * |
| 278 | * 1. The mounted_lock tells a node it's the first to mount the fs. |
| 279 | * Each node holds the mounted_lock in PR while it's mounted. |
| 280 | * Each node tries to acquire the mounted_lock in EX when it mounts. |
| 281 | * If a node is granted the mounted_lock EX it means there are no |
| 282 | * other mounted nodes (no PR locks exist), and it is the first mounter. |
| 283 | * The mounted_lock is demoted to PR when first recovery is done, so |
| 284 | * others will fail to get an EX lock, but will get a PR lock. |
| 285 | * |
| 286 | * 2. The control_lock blocks others in control_mount() while the first |
| 287 | * mounter is doing first mount recovery of all journals. |
| 288 | * A mounting node needs to acquire control_lock in EX mode before |
| 289 | * it can proceed. The first mounter holds control_lock in EX while doing |
| 290 | * the first mount recovery, blocking mounts from other nodes, then demotes |
| 291 | * control_lock to NL when it's done (others_may_mount/first_done), |
| 292 | * allowing other nodes to continue mounting. |
| 293 | * |
| 294 | * first mounter: |
| 295 | * control_lock EX/NOQUEUE success |
| 296 | * mounted_lock EX/NOQUEUE success (no other PR, so no other mounters) |
| 297 | * set first=1 |
| 298 | * do first mounter recovery |
| 299 | * mounted_lock EX->PR |
| 300 | * control_lock EX->NL, write lvb generation |
| 301 | * |
| 302 | * other mounter: |
| 303 | * control_lock EX/NOQUEUE success (if fail -EAGAIN, retry) |
| 304 | * mounted_lock EX/NOQUEUE fail -EAGAIN (expected due to other mounters PR) |
| 305 | * mounted_lock PR/NOQUEUE success |
| 306 | * read lvb generation |
| 307 | * control_lock EX->NL |
| 308 | * set first=0 |
| 309 | * |
| 310 | * - mount during recovery |
| 311 | * |
| 312 | * If a node mounts while others are doing recovery (not first mounter), |
| 313 | * the mounting node will get its initial recover_done() callback without |
| 314 | * having seen any previous failures/callbacks. |
| 315 | * |
| 316 | * It must wait for all recoveries preceding its mount to be finished |
| 317 | * before it unblocks locks. It does this by repeating the "other mounter" |
| 318 | * steps above until the lvb generation number is >= its mount generation |
| 319 | * number (from initial recover_done) and all lvb bits are clear. |
| 320 | * |
| 321 | * - control_lock lvb format |
| 322 | * |
| 323 | * 4 bytes generation number: the latest dlm lockspace generation number |
| 324 | * from recover_done callback. Indicates the jid bitmap has been updated |
| 325 | * to reflect all slot failures through that generation. |
| 326 | * 4 bytes unused. |
| 327 | * GDLM_LVB_SIZE-8 bytes of jid bit map. If bit N is set, it indicates |
| 328 | * that jid N needs recovery. |
| 329 | */ |
| 330 | |
| 331 | #define JID_BITMAP_OFFSET 8 /* 4 byte generation number + 4 byte unused */ |
| 332 | |
| 333 | static void control_lvb_read(struct lm_lockstruct *ls, uint32_t *lvb_gen, |
| 334 | char *lvb_bits) |
| 335 | { |
| 336 | uint32_t gen; |
| 337 | memcpy(lvb_bits, ls->ls_control_lvb, GDLM_LVB_SIZE); |
| 338 | memcpy(&gen, lvb_bits, sizeof(uint32_t)); |
| 339 | *lvb_gen = le32_to_cpu(gen); |
| 340 | } |
| 341 | |
| 342 | static void control_lvb_write(struct lm_lockstruct *ls, uint32_t lvb_gen, |
| 343 | char *lvb_bits) |
| 344 | { |
| 345 | uint32_t gen; |
| 346 | memcpy(ls->ls_control_lvb, lvb_bits, GDLM_LVB_SIZE); |
| 347 | gen = cpu_to_le32(lvb_gen); |
| 348 | memcpy(ls->ls_control_lvb, &gen, sizeof(uint32_t)); |
| 349 | } |
| 350 | |
| 351 | static int all_jid_bits_clear(char *lvb) |
| 352 | { |
| 353 | int i; |
| 354 | for (i = JID_BITMAP_OFFSET; i < GDLM_LVB_SIZE; i++) { |
| 355 | if (lvb[i]) |
| 356 | return 0; |
| 357 | } |
| 358 | return 1; |
| 359 | } |
| 360 | |
| 361 | static void sync_wait_cb(void *arg) |
| 362 | { |
| 363 | struct lm_lockstruct *ls = arg; |
| 364 | complete(&ls->ls_sync_wait); |
| 365 | } |
| 366 | |
| 367 | static int sync_unlock(struct gfs2_sbd *sdp, struct dlm_lksb *lksb, char *name) |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 368 | { |
| 369 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 370 | int error; |
| 371 | |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 372 | error = dlm_unlock(ls->ls_dlm, lksb->sb_lkid, 0, lksb, ls); |
| 373 | if (error) { |
| 374 | fs_err(sdp, "%s lkid %x error %d\n", |
| 375 | name, lksb->sb_lkid, error); |
| 376 | return error; |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 377 | } |
| 378 | |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 379 | wait_for_completion(&ls->ls_sync_wait); |
| 380 | |
| 381 | if (lksb->sb_status != -DLM_EUNLOCK) { |
| 382 | fs_err(sdp, "%s lkid %x status %d\n", |
| 383 | name, lksb->sb_lkid, lksb->sb_status); |
| 384 | return -1; |
| 385 | } |
| 386 | return 0; |
| 387 | } |
| 388 | |
| 389 | static int sync_lock(struct gfs2_sbd *sdp, int mode, uint32_t flags, |
| 390 | unsigned int num, struct dlm_lksb *lksb, char *name) |
| 391 | { |
| 392 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 393 | char strname[GDLM_STRNAME_BYTES]; |
| 394 | int error, status; |
| 395 | |
| 396 | memset(strname, 0, GDLM_STRNAME_BYTES); |
| 397 | snprintf(strname, GDLM_STRNAME_BYTES, "%8x%16x", LM_TYPE_NONDISK, num); |
| 398 | |
| 399 | error = dlm_lock(ls->ls_dlm, mode, lksb, flags, |
| 400 | strname, GDLM_STRNAME_BYTES - 1, |
| 401 | 0, sync_wait_cb, ls, NULL); |
| 402 | if (error) { |
| 403 | fs_err(sdp, "%s lkid %x flags %x mode %d error %d\n", |
| 404 | name, lksb->sb_lkid, flags, mode, error); |
| 405 | return error; |
| 406 | } |
| 407 | |
| 408 | wait_for_completion(&ls->ls_sync_wait); |
| 409 | |
| 410 | status = lksb->sb_status; |
| 411 | |
| 412 | if (status && status != -EAGAIN) { |
| 413 | fs_err(sdp, "%s lkid %x flags %x mode %d status %d\n", |
| 414 | name, lksb->sb_lkid, flags, mode, status); |
| 415 | } |
| 416 | |
| 417 | return status; |
| 418 | } |
| 419 | |
| 420 | static int mounted_unlock(struct gfs2_sbd *sdp) |
| 421 | { |
| 422 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 423 | return sync_unlock(sdp, &ls->ls_mounted_lksb, "mounted_lock"); |
| 424 | } |
| 425 | |
| 426 | static int mounted_lock(struct gfs2_sbd *sdp, int mode, uint32_t flags) |
| 427 | { |
| 428 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 429 | return sync_lock(sdp, mode, flags, GFS2_MOUNTED_LOCK, |
| 430 | &ls->ls_mounted_lksb, "mounted_lock"); |
| 431 | } |
| 432 | |
| 433 | static int control_unlock(struct gfs2_sbd *sdp) |
| 434 | { |
| 435 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 436 | return sync_unlock(sdp, &ls->ls_control_lksb, "control_lock"); |
| 437 | } |
| 438 | |
| 439 | static int control_lock(struct gfs2_sbd *sdp, int mode, uint32_t flags) |
| 440 | { |
| 441 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 442 | return sync_lock(sdp, mode, flags, GFS2_CONTROL_LOCK, |
| 443 | &ls->ls_control_lksb, "control_lock"); |
| 444 | } |
| 445 | |
| 446 | static void gfs2_control_func(struct work_struct *work) |
| 447 | { |
| 448 | struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_control_work.work); |
| 449 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 450 | char lvb_bits[GDLM_LVB_SIZE]; |
| 451 | uint32_t block_gen, start_gen, lvb_gen, flags; |
| 452 | int recover_set = 0; |
| 453 | int write_lvb = 0; |
| 454 | int recover_size; |
| 455 | int i, error; |
| 456 | |
| 457 | spin_lock(&ls->ls_recover_spin); |
| 458 | /* |
| 459 | * No MOUNT_DONE means we're still mounting; control_mount() |
| 460 | * will set this flag, after which this thread will take over |
| 461 | * all further clearing of BLOCK_LOCKS. |
| 462 | * |
| 463 | * FIRST_MOUNT means this node is doing first mounter recovery, |
| 464 | * for which recovery control is handled by |
| 465 | * control_mount()/control_first_done(), not this thread. |
| 466 | */ |
| 467 | if (!test_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags) || |
| 468 | test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags)) { |
| 469 | spin_unlock(&ls->ls_recover_spin); |
| 470 | return; |
| 471 | } |
| 472 | block_gen = ls->ls_recover_block; |
| 473 | start_gen = ls->ls_recover_start; |
| 474 | spin_unlock(&ls->ls_recover_spin); |
| 475 | |
| 476 | /* |
| 477 | * Equal block_gen and start_gen implies we are between |
| 478 | * recover_prep and recover_done callbacks, which means |
| 479 | * dlm recovery is in progress and dlm locking is blocked. |
| 480 | * There's no point trying to do any work until recover_done. |
| 481 | */ |
| 482 | |
| 483 | if (block_gen == start_gen) |
| 484 | return; |
| 485 | |
| 486 | /* |
| 487 | * Propagate recover_submit[] and recover_result[] to lvb: |
| 488 | * dlm_recoverd adds to recover_submit[] jids needing recovery |
| 489 | * gfs2_recover adds to recover_result[] journal recovery results |
| 490 | * |
| 491 | * set lvb bit for jids in recover_submit[] if the lvb has not |
| 492 | * yet been updated for the generation of the failure |
| 493 | * |
| 494 | * clear lvb bit for jids in recover_result[] if the result of |
| 495 | * the journal recovery is SUCCESS |
| 496 | */ |
| 497 | |
| 498 | error = control_lock(sdp, DLM_LOCK_EX, DLM_LKF_CONVERT|DLM_LKF_VALBLK); |
| 499 | if (error) { |
| 500 | fs_err(sdp, "control lock EX error %d\n", error); |
| 501 | return; |
| 502 | } |
| 503 | |
| 504 | control_lvb_read(ls, &lvb_gen, lvb_bits); |
| 505 | |
| 506 | spin_lock(&ls->ls_recover_spin); |
| 507 | if (block_gen != ls->ls_recover_block || |
| 508 | start_gen != ls->ls_recover_start) { |
| 509 | fs_info(sdp, "recover generation %u block1 %u %u\n", |
| 510 | start_gen, block_gen, ls->ls_recover_block); |
| 511 | spin_unlock(&ls->ls_recover_spin); |
| 512 | control_lock(sdp, DLM_LOCK_NL, DLM_LKF_CONVERT); |
| 513 | return; |
| 514 | } |
| 515 | |
| 516 | recover_size = ls->ls_recover_size; |
| 517 | |
| 518 | if (lvb_gen <= start_gen) { |
| 519 | /* |
| 520 | * Clear lvb bits for jids we've successfully recovered. |
| 521 | * Because all nodes attempt to recover failed journals, |
| 522 | * a journal can be recovered multiple times successfully |
| 523 | * in succession. Only the first will really do recovery, |
| 524 | * the others find it clean, but still report a successful |
| 525 | * recovery. So, another node may have already recovered |
| 526 | * the jid and cleared the lvb bit for it. |
| 527 | */ |
| 528 | for (i = 0; i < recover_size; i++) { |
| 529 | if (ls->ls_recover_result[i] != LM_RD_SUCCESS) |
| 530 | continue; |
| 531 | |
| 532 | ls->ls_recover_result[i] = 0; |
| 533 | |
| 534 | if (!test_bit_le(i, lvb_bits + JID_BITMAP_OFFSET)) |
| 535 | continue; |
| 536 | |
| 537 | __clear_bit_le(i, lvb_bits + JID_BITMAP_OFFSET); |
| 538 | write_lvb = 1; |
| 539 | } |
| 540 | } |
| 541 | |
| 542 | if (lvb_gen == start_gen) { |
| 543 | /* |
| 544 | * Failed slots before start_gen are already set in lvb. |
| 545 | */ |
| 546 | for (i = 0; i < recover_size; i++) { |
| 547 | if (!ls->ls_recover_submit[i]) |
| 548 | continue; |
| 549 | if (ls->ls_recover_submit[i] < lvb_gen) |
| 550 | ls->ls_recover_submit[i] = 0; |
| 551 | } |
| 552 | } else if (lvb_gen < start_gen) { |
| 553 | /* |
| 554 | * Failed slots before start_gen are not yet set in lvb. |
| 555 | */ |
| 556 | for (i = 0; i < recover_size; i++) { |
| 557 | if (!ls->ls_recover_submit[i]) |
| 558 | continue; |
| 559 | if (ls->ls_recover_submit[i] < start_gen) { |
| 560 | ls->ls_recover_submit[i] = 0; |
| 561 | __set_bit_le(i, lvb_bits + JID_BITMAP_OFFSET); |
| 562 | } |
| 563 | } |
| 564 | /* even if there are no bits to set, we need to write the |
| 565 | latest generation to the lvb */ |
| 566 | write_lvb = 1; |
| 567 | } else { |
| 568 | /* |
| 569 | * we should be getting a recover_done() for lvb_gen soon |
| 570 | */ |
| 571 | } |
| 572 | spin_unlock(&ls->ls_recover_spin); |
| 573 | |
| 574 | if (write_lvb) { |
| 575 | control_lvb_write(ls, start_gen, lvb_bits); |
| 576 | flags = DLM_LKF_CONVERT | DLM_LKF_VALBLK; |
| 577 | } else { |
| 578 | flags = DLM_LKF_CONVERT; |
| 579 | } |
| 580 | |
| 581 | error = control_lock(sdp, DLM_LOCK_NL, flags); |
| 582 | if (error) { |
| 583 | fs_err(sdp, "control lock NL error %d\n", error); |
| 584 | return; |
| 585 | } |
| 586 | |
| 587 | /* |
| 588 | * Everyone will see jid bits set in the lvb, run gfs2_recover_set(), |
| 589 | * and clear a jid bit in the lvb if the recovery is a success. |
| 590 | * Eventually all journals will be recovered, all jid bits will |
| 591 | * be cleared in the lvb, and everyone will clear BLOCK_LOCKS. |
| 592 | */ |
| 593 | |
| 594 | for (i = 0; i < recover_size; i++) { |
| 595 | if (test_bit_le(i, lvb_bits + JID_BITMAP_OFFSET)) { |
| 596 | fs_info(sdp, "recover generation %u jid %d\n", |
| 597 | start_gen, i); |
| 598 | gfs2_recover_set(sdp, i); |
| 599 | recover_set++; |
| 600 | } |
| 601 | } |
| 602 | if (recover_set) |
| 603 | return; |
| 604 | |
| 605 | /* |
| 606 | * No more jid bits set in lvb, all recovery is done, unblock locks |
| 607 | * (unless a new recover_prep callback has occured blocking locks |
| 608 | * again while working above) |
| 609 | */ |
| 610 | |
| 611 | spin_lock(&ls->ls_recover_spin); |
| 612 | if (ls->ls_recover_block == block_gen && |
| 613 | ls->ls_recover_start == start_gen) { |
| 614 | clear_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags); |
| 615 | spin_unlock(&ls->ls_recover_spin); |
| 616 | fs_info(sdp, "recover generation %u done\n", start_gen); |
| 617 | gfs2_glock_thaw(sdp); |
| 618 | } else { |
| 619 | fs_info(sdp, "recover generation %u block2 %u %u\n", |
| 620 | start_gen, block_gen, ls->ls_recover_block); |
| 621 | spin_unlock(&ls->ls_recover_spin); |
| 622 | } |
| 623 | } |
| 624 | |
| 625 | static int control_mount(struct gfs2_sbd *sdp) |
| 626 | { |
| 627 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 628 | char lvb_bits[GDLM_LVB_SIZE]; |
| 629 | uint32_t start_gen, block_gen, mount_gen, lvb_gen; |
| 630 | int mounted_mode; |
| 631 | int retries = 0; |
| 632 | int error; |
| 633 | |
| 634 | memset(&ls->ls_mounted_lksb, 0, sizeof(struct dlm_lksb)); |
| 635 | memset(&ls->ls_control_lksb, 0, sizeof(struct dlm_lksb)); |
| 636 | memset(&ls->ls_control_lvb, 0, GDLM_LVB_SIZE); |
| 637 | ls->ls_control_lksb.sb_lvbptr = ls->ls_control_lvb; |
| 638 | init_completion(&ls->ls_sync_wait); |
| 639 | |
| 640 | set_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags); |
| 641 | |
| 642 | error = control_lock(sdp, DLM_LOCK_NL, DLM_LKF_VALBLK); |
| 643 | if (error) { |
| 644 | fs_err(sdp, "control_mount control_lock NL error %d\n", error); |
| 645 | return error; |
| 646 | } |
| 647 | |
| 648 | error = mounted_lock(sdp, DLM_LOCK_NL, 0); |
| 649 | if (error) { |
| 650 | fs_err(sdp, "control_mount mounted_lock NL error %d\n", error); |
| 651 | control_unlock(sdp); |
| 652 | return error; |
| 653 | } |
| 654 | mounted_mode = DLM_LOCK_NL; |
| 655 | |
| 656 | restart: |
| 657 | if (retries++ && signal_pending(current)) { |
| 658 | error = -EINTR; |
| 659 | goto fail; |
| 660 | } |
| 661 | |
| 662 | /* |
| 663 | * We always start with both locks in NL. control_lock is |
| 664 | * demoted to NL below so we don't need to do it here. |
| 665 | */ |
| 666 | |
| 667 | if (mounted_mode != DLM_LOCK_NL) { |
| 668 | error = mounted_lock(sdp, DLM_LOCK_NL, DLM_LKF_CONVERT); |
| 669 | if (error) |
| 670 | goto fail; |
| 671 | mounted_mode = DLM_LOCK_NL; |
| 672 | } |
| 673 | |
| 674 | /* |
| 675 | * Other nodes need to do some work in dlm recovery and gfs2_control |
| 676 | * before the recover_done and control_lock will be ready for us below. |
| 677 | * A delay here is not required but often avoids having to retry. |
| 678 | */ |
| 679 | |
| 680 | msleep_interruptible(500); |
| 681 | |
| 682 | /* |
| 683 | * Acquire control_lock in EX and mounted_lock in either EX or PR. |
| 684 | * control_lock lvb keeps track of any pending journal recoveries. |
| 685 | * mounted_lock indicates if any other nodes have the fs mounted. |
| 686 | */ |
| 687 | |
| 688 | error = control_lock(sdp, DLM_LOCK_EX, DLM_LKF_CONVERT|DLM_LKF_NOQUEUE|DLM_LKF_VALBLK); |
| 689 | if (error == -EAGAIN) { |
| 690 | goto restart; |
| 691 | } else if (error) { |
| 692 | fs_err(sdp, "control_mount control_lock EX error %d\n", error); |
| 693 | goto fail; |
| 694 | } |
| 695 | |
| 696 | error = mounted_lock(sdp, DLM_LOCK_EX, DLM_LKF_CONVERT|DLM_LKF_NOQUEUE); |
| 697 | if (!error) { |
| 698 | mounted_mode = DLM_LOCK_EX; |
| 699 | goto locks_done; |
| 700 | } else if (error != -EAGAIN) { |
| 701 | fs_err(sdp, "control_mount mounted_lock EX error %d\n", error); |
| 702 | goto fail; |
| 703 | } |
| 704 | |
| 705 | error = mounted_lock(sdp, DLM_LOCK_PR, DLM_LKF_CONVERT|DLM_LKF_NOQUEUE); |
| 706 | if (!error) { |
| 707 | mounted_mode = DLM_LOCK_PR; |
| 708 | goto locks_done; |
| 709 | } else { |
| 710 | /* not even -EAGAIN should happen here */ |
| 711 | fs_err(sdp, "control_mount mounted_lock PR error %d\n", error); |
| 712 | goto fail; |
| 713 | } |
| 714 | |
| 715 | locks_done: |
| 716 | /* |
| 717 | * If we got both locks above in EX, then we're the first mounter. |
| 718 | * If not, then we need to wait for the control_lock lvb to be |
| 719 | * updated by other mounted nodes to reflect our mount generation. |
| 720 | * |
| 721 | * In simple first mounter cases, first mounter will see zero lvb_gen, |
| 722 | * but in cases where all existing nodes leave/fail before mounting |
| 723 | * nodes finish control_mount, then all nodes will be mounting and |
| 724 | * lvb_gen will be non-zero. |
| 725 | */ |
| 726 | |
| 727 | control_lvb_read(ls, &lvb_gen, lvb_bits); |
| 728 | |
| 729 | if (lvb_gen == 0xFFFFFFFF) { |
| 730 | /* special value to force mount attempts to fail */ |
| 731 | fs_err(sdp, "control_mount control_lock disabled\n"); |
| 732 | error = -EINVAL; |
| 733 | goto fail; |
| 734 | } |
| 735 | |
| 736 | if (mounted_mode == DLM_LOCK_EX) { |
| 737 | /* first mounter, keep both EX while doing first recovery */ |
| 738 | spin_lock(&ls->ls_recover_spin); |
| 739 | clear_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags); |
| 740 | set_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags); |
| 741 | set_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags); |
| 742 | spin_unlock(&ls->ls_recover_spin); |
| 743 | fs_info(sdp, "first mounter control generation %u\n", lvb_gen); |
| 744 | return 0; |
| 745 | } |
| 746 | |
| 747 | error = control_lock(sdp, DLM_LOCK_NL, DLM_LKF_CONVERT); |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 748 | if (error) |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 749 | goto fail; |
| 750 | |
| 751 | /* |
| 752 | * We are not first mounter, now we need to wait for the control_lock |
| 753 | * lvb generation to be >= the generation from our first recover_done |
| 754 | * and all lvb bits to be clear (no pending journal recoveries.) |
| 755 | */ |
| 756 | |
| 757 | if (!all_jid_bits_clear(lvb_bits)) { |
| 758 | /* journals need recovery, wait until all are clear */ |
| 759 | fs_info(sdp, "control_mount wait for journal recovery\n"); |
| 760 | goto restart; |
| 761 | } |
| 762 | |
| 763 | spin_lock(&ls->ls_recover_spin); |
| 764 | block_gen = ls->ls_recover_block; |
| 765 | start_gen = ls->ls_recover_start; |
| 766 | mount_gen = ls->ls_recover_mount; |
| 767 | |
| 768 | if (lvb_gen < mount_gen) { |
| 769 | /* wait for mounted nodes to update control_lock lvb to our |
| 770 | generation, which might include new recovery bits set */ |
| 771 | fs_info(sdp, "control_mount wait1 block %u start %u mount %u " |
| 772 | "lvb %u flags %lx\n", block_gen, start_gen, mount_gen, |
| 773 | lvb_gen, ls->ls_recover_flags); |
| 774 | spin_unlock(&ls->ls_recover_spin); |
| 775 | goto restart; |
| 776 | } |
| 777 | |
| 778 | if (lvb_gen != start_gen) { |
| 779 | /* wait for mounted nodes to update control_lock lvb to the |
| 780 | latest recovery generation */ |
| 781 | fs_info(sdp, "control_mount wait2 block %u start %u mount %u " |
| 782 | "lvb %u flags %lx\n", block_gen, start_gen, mount_gen, |
| 783 | lvb_gen, ls->ls_recover_flags); |
| 784 | spin_unlock(&ls->ls_recover_spin); |
| 785 | goto restart; |
| 786 | } |
| 787 | |
| 788 | if (block_gen == start_gen) { |
| 789 | /* dlm recovery in progress, wait for it to finish */ |
| 790 | fs_info(sdp, "control_mount wait3 block %u start %u mount %u " |
| 791 | "lvb %u flags %lx\n", block_gen, start_gen, mount_gen, |
| 792 | lvb_gen, ls->ls_recover_flags); |
| 793 | spin_unlock(&ls->ls_recover_spin); |
| 794 | goto restart; |
| 795 | } |
| 796 | |
| 797 | clear_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags); |
| 798 | set_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags); |
| 799 | memset(ls->ls_recover_submit, 0, ls->ls_recover_size*sizeof(uint32_t)); |
| 800 | memset(ls->ls_recover_result, 0, ls->ls_recover_size*sizeof(uint32_t)); |
| 801 | spin_unlock(&ls->ls_recover_spin); |
| 802 | return 0; |
| 803 | |
| 804 | fail: |
| 805 | mounted_unlock(sdp); |
| 806 | control_unlock(sdp); |
| 807 | return error; |
| 808 | } |
| 809 | |
| 810 | static int dlm_recovery_wait(void *word) |
| 811 | { |
| 812 | schedule(); |
| 813 | return 0; |
| 814 | } |
| 815 | |
| 816 | static int control_first_done(struct gfs2_sbd *sdp) |
| 817 | { |
| 818 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 819 | char lvb_bits[GDLM_LVB_SIZE]; |
| 820 | uint32_t start_gen, block_gen; |
| 821 | int error; |
| 822 | |
| 823 | restart: |
| 824 | spin_lock(&ls->ls_recover_spin); |
| 825 | start_gen = ls->ls_recover_start; |
| 826 | block_gen = ls->ls_recover_block; |
| 827 | |
| 828 | if (test_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags) || |
| 829 | !test_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags) || |
| 830 | !test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags)) { |
| 831 | /* sanity check, should not happen */ |
| 832 | fs_err(sdp, "control_first_done start %u block %u flags %lx\n", |
| 833 | start_gen, block_gen, ls->ls_recover_flags); |
| 834 | spin_unlock(&ls->ls_recover_spin); |
| 835 | control_unlock(sdp); |
| 836 | return -1; |
| 837 | } |
| 838 | |
| 839 | if (start_gen == block_gen) { |
| 840 | /* |
| 841 | * Wait for the end of a dlm recovery cycle to switch from |
| 842 | * first mounter recovery. We can ignore any recover_slot |
| 843 | * callbacks between the recover_prep and next recover_done |
| 844 | * because we are still the first mounter and any failed nodes |
| 845 | * have not fully mounted, so they don't need recovery. |
| 846 | */ |
| 847 | spin_unlock(&ls->ls_recover_spin); |
| 848 | fs_info(sdp, "control_first_done wait gen %u\n", start_gen); |
| 849 | |
| 850 | wait_on_bit(&ls->ls_recover_flags, DFL_DLM_RECOVERY, |
| 851 | dlm_recovery_wait, TASK_UNINTERRUPTIBLE); |
| 852 | goto restart; |
| 853 | } |
| 854 | |
| 855 | clear_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags); |
| 856 | set_bit(DFL_FIRST_MOUNT_DONE, &ls->ls_recover_flags); |
| 857 | memset(ls->ls_recover_submit, 0, ls->ls_recover_size*sizeof(uint32_t)); |
| 858 | memset(ls->ls_recover_result, 0, ls->ls_recover_size*sizeof(uint32_t)); |
| 859 | spin_unlock(&ls->ls_recover_spin); |
| 860 | |
| 861 | memset(lvb_bits, 0, sizeof(lvb_bits)); |
| 862 | control_lvb_write(ls, start_gen, lvb_bits); |
| 863 | |
| 864 | error = mounted_lock(sdp, DLM_LOCK_PR, DLM_LKF_CONVERT); |
| 865 | if (error) |
| 866 | fs_err(sdp, "control_first_done mounted PR error %d\n", error); |
| 867 | |
| 868 | error = control_lock(sdp, DLM_LOCK_NL, DLM_LKF_CONVERT|DLM_LKF_VALBLK); |
| 869 | if (error) |
| 870 | fs_err(sdp, "control_first_done control NL error %d\n", error); |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 871 | |
| 872 | return error; |
| 873 | } |
| 874 | |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 875 | /* |
| 876 | * Expand static jid arrays if necessary (by increments of RECOVER_SIZE_INC) |
| 877 | * to accomodate the largest slot number. (NB dlm slot numbers start at 1, |
| 878 | * gfs2 jids start at 0, so jid = slot - 1) |
| 879 | */ |
| 880 | |
| 881 | #define RECOVER_SIZE_INC 16 |
| 882 | |
| 883 | static int set_recover_size(struct gfs2_sbd *sdp, struct dlm_slot *slots, |
| 884 | int num_slots) |
| 885 | { |
| 886 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 887 | uint32_t *submit = NULL; |
| 888 | uint32_t *result = NULL; |
| 889 | uint32_t old_size, new_size; |
| 890 | int i, max_jid; |
| 891 | |
| 892 | max_jid = 0; |
| 893 | for (i = 0; i < num_slots; i++) { |
| 894 | if (max_jid < slots[i].slot - 1) |
| 895 | max_jid = slots[i].slot - 1; |
| 896 | } |
| 897 | |
| 898 | old_size = ls->ls_recover_size; |
| 899 | |
| 900 | if (old_size >= max_jid + 1) |
| 901 | return 0; |
| 902 | |
| 903 | new_size = old_size + RECOVER_SIZE_INC; |
| 904 | |
| 905 | submit = kzalloc(new_size * sizeof(uint32_t), GFP_NOFS); |
| 906 | result = kzalloc(new_size * sizeof(uint32_t), GFP_NOFS); |
| 907 | if (!submit || !result) { |
| 908 | kfree(submit); |
| 909 | kfree(result); |
| 910 | return -ENOMEM; |
| 911 | } |
| 912 | |
| 913 | spin_lock(&ls->ls_recover_spin); |
| 914 | memcpy(submit, ls->ls_recover_submit, old_size * sizeof(uint32_t)); |
| 915 | memcpy(result, ls->ls_recover_result, old_size * sizeof(uint32_t)); |
| 916 | kfree(ls->ls_recover_submit); |
| 917 | kfree(ls->ls_recover_result); |
| 918 | ls->ls_recover_submit = submit; |
| 919 | ls->ls_recover_result = result; |
| 920 | ls->ls_recover_size = new_size; |
| 921 | spin_unlock(&ls->ls_recover_spin); |
| 922 | return 0; |
| 923 | } |
| 924 | |
| 925 | static void free_recover_size(struct lm_lockstruct *ls) |
| 926 | { |
| 927 | kfree(ls->ls_recover_submit); |
| 928 | kfree(ls->ls_recover_result); |
| 929 | ls->ls_recover_submit = NULL; |
| 930 | ls->ls_recover_result = NULL; |
| 931 | ls->ls_recover_size = 0; |
| 932 | } |
| 933 | |
| 934 | /* dlm calls before it does lock recovery */ |
| 935 | |
| 936 | static void gdlm_recover_prep(void *arg) |
| 937 | { |
| 938 | struct gfs2_sbd *sdp = arg; |
| 939 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 940 | |
| 941 | spin_lock(&ls->ls_recover_spin); |
| 942 | ls->ls_recover_block = ls->ls_recover_start; |
| 943 | set_bit(DFL_DLM_RECOVERY, &ls->ls_recover_flags); |
| 944 | |
| 945 | if (!test_bit(DFL_MOUNT_DONE, &ls->ls_recover_flags) || |
| 946 | test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags)) { |
| 947 | spin_unlock(&ls->ls_recover_spin); |
| 948 | return; |
| 949 | } |
| 950 | set_bit(DFL_BLOCK_LOCKS, &ls->ls_recover_flags); |
| 951 | spin_unlock(&ls->ls_recover_spin); |
| 952 | } |
| 953 | |
| 954 | /* dlm calls after recover_prep has been completed on all lockspace members; |
| 955 | identifies slot/jid of failed member */ |
| 956 | |
| 957 | static void gdlm_recover_slot(void *arg, struct dlm_slot *slot) |
| 958 | { |
| 959 | struct gfs2_sbd *sdp = arg; |
| 960 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 961 | int jid = slot->slot - 1; |
| 962 | |
| 963 | spin_lock(&ls->ls_recover_spin); |
| 964 | if (ls->ls_recover_size < jid + 1) { |
| 965 | fs_err(sdp, "recover_slot jid %d gen %u short size %d", |
| 966 | jid, ls->ls_recover_block, ls->ls_recover_size); |
| 967 | spin_unlock(&ls->ls_recover_spin); |
| 968 | return; |
| 969 | } |
| 970 | |
| 971 | if (ls->ls_recover_submit[jid]) { |
| 972 | fs_info(sdp, "recover_slot jid %d gen %u prev %u", |
| 973 | jid, ls->ls_recover_block, ls->ls_recover_submit[jid]); |
| 974 | } |
| 975 | ls->ls_recover_submit[jid] = ls->ls_recover_block; |
| 976 | spin_unlock(&ls->ls_recover_spin); |
| 977 | } |
| 978 | |
| 979 | /* dlm calls after recover_slot and after it completes lock recovery */ |
| 980 | |
| 981 | static void gdlm_recover_done(void *arg, struct dlm_slot *slots, int num_slots, |
| 982 | int our_slot, uint32_t generation) |
| 983 | { |
| 984 | struct gfs2_sbd *sdp = arg; |
| 985 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 986 | |
| 987 | /* ensure the ls jid arrays are large enough */ |
| 988 | set_recover_size(sdp, slots, num_slots); |
| 989 | |
| 990 | spin_lock(&ls->ls_recover_spin); |
| 991 | ls->ls_recover_start = generation; |
| 992 | |
| 993 | if (!ls->ls_recover_mount) { |
| 994 | ls->ls_recover_mount = generation; |
| 995 | ls->ls_jid = our_slot - 1; |
| 996 | } |
| 997 | |
| 998 | if (!test_bit(DFL_UNMOUNT, &ls->ls_recover_flags)) |
| 999 | queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, 0); |
| 1000 | |
| 1001 | clear_bit(DFL_DLM_RECOVERY, &ls->ls_recover_flags); |
| 1002 | smp_mb__after_clear_bit(); |
| 1003 | wake_up_bit(&ls->ls_recover_flags, DFL_DLM_RECOVERY); |
| 1004 | spin_unlock(&ls->ls_recover_spin); |
| 1005 | } |
| 1006 | |
| 1007 | /* gfs2_recover thread has a journal recovery result */ |
| 1008 | |
| 1009 | static void gdlm_recovery_result(struct gfs2_sbd *sdp, unsigned int jid, |
| 1010 | unsigned int result) |
| 1011 | { |
| 1012 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 1013 | |
| 1014 | if (test_bit(DFL_NO_DLM_OPS, &ls->ls_recover_flags)) |
| 1015 | return; |
| 1016 | |
| 1017 | /* don't care about the recovery of own journal during mount */ |
| 1018 | if (jid == ls->ls_jid) |
| 1019 | return; |
| 1020 | |
| 1021 | spin_lock(&ls->ls_recover_spin); |
| 1022 | if (test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags)) { |
| 1023 | spin_unlock(&ls->ls_recover_spin); |
| 1024 | return; |
| 1025 | } |
| 1026 | if (ls->ls_recover_size < jid + 1) { |
| 1027 | fs_err(sdp, "recovery_result jid %d short size %d", |
| 1028 | jid, ls->ls_recover_size); |
| 1029 | spin_unlock(&ls->ls_recover_spin); |
| 1030 | return; |
| 1031 | } |
| 1032 | |
| 1033 | fs_info(sdp, "recover jid %d result %s\n", jid, |
| 1034 | result == LM_RD_GAVEUP ? "busy" : "success"); |
| 1035 | |
| 1036 | ls->ls_recover_result[jid] = result; |
| 1037 | |
| 1038 | /* GAVEUP means another node is recovering the journal; delay our |
| 1039 | next attempt to recover it, to give the other node a chance to |
| 1040 | finish before trying again */ |
| 1041 | |
| 1042 | if (!test_bit(DFL_UNMOUNT, &ls->ls_recover_flags)) |
| 1043 | queue_delayed_work(gfs2_control_wq, &sdp->sd_control_work, |
| 1044 | result == LM_RD_GAVEUP ? HZ : 0); |
| 1045 | spin_unlock(&ls->ls_recover_spin); |
| 1046 | } |
| 1047 | |
| 1048 | const struct dlm_lockspace_ops gdlm_lockspace_ops = { |
| 1049 | .recover_prep = gdlm_recover_prep, |
| 1050 | .recover_slot = gdlm_recover_slot, |
| 1051 | .recover_done = gdlm_recover_done, |
| 1052 | }; |
| 1053 | |
| 1054 | static int gdlm_mount(struct gfs2_sbd *sdp, const char *table) |
| 1055 | { |
| 1056 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 1057 | char cluster[GFS2_LOCKNAME_LEN]; |
| 1058 | const char *fsname; |
| 1059 | uint32_t flags; |
| 1060 | int error, ops_result; |
| 1061 | |
| 1062 | /* |
| 1063 | * initialize everything |
| 1064 | */ |
| 1065 | |
| 1066 | INIT_DELAYED_WORK(&sdp->sd_control_work, gfs2_control_func); |
| 1067 | spin_lock_init(&ls->ls_recover_spin); |
| 1068 | ls->ls_recover_flags = 0; |
| 1069 | ls->ls_recover_mount = 0; |
| 1070 | ls->ls_recover_start = 0; |
| 1071 | ls->ls_recover_block = 0; |
| 1072 | ls->ls_recover_size = 0; |
| 1073 | ls->ls_recover_submit = NULL; |
| 1074 | ls->ls_recover_result = NULL; |
| 1075 | |
| 1076 | error = set_recover_size(sdp, NULL, 0); |
| 1077 | if (error) |
| 1078 | goto fail; |
| 1079 | |
| 1080 | /* |
| 1081 | * prepare dlm_new_lockspace args |
| 1082 | */ |
| 1083 | |
| 1084 | fsname = strchr(table, ':'); |
| 1085 | if (!fsname) { |
| 1086 | fs_info(sdp, "no fsname found\n"); |
| 1087 | error = -EINVAL; |
| 1088 | goto fail_free; |
| 1089 | } |
| 1090 | memset(cluster, 0, sizeof(cluster)); |
| 1091 | memcpy(cluster, table, strlen(table) - strlen(fsname)); |
| 1092 | fsname++; |
| 1093 | |
| 1094 | flags = DLM_LSFL_FS | DLM_LSFL_NEWEXCL; |
| 1095 | if (ls->ls_nodir) |
| 1096 | flags |= DLM_LSFL_NODIR; |
| 1097 | |
| 1098 | /* |
| 1099 | * create/join lockspace |
| 1100 | */ |
| 1101 | |
| 1102 | error = dlm_new_lockspace(fsname, cluster, flags, GDLM_LVB_SIZE, |
| 1103 | &gdlm_lockspace_ops, sdp, &ops_result, |
| 1104 | &ls->ls_dlm); |
| 1105 | if (error) { |
| 1106 | fs_err(sdp, "dlm_new_lockspace error %d\n", error); |
| 1107 | goto fail_free; |
| 1108 | } |
| 1109 | |
| 1110 | if (ops_result < 0) { |
| 1111 | /* |
| 1112 | * dlm does not support ops callbacks, |
| 1113 | * old dlm_controld/gfs_controld are used, try without ops. |
| 1114 | */ |
| 1115 | fs_info(sdp, "dlm lockspace ops not used\n"); |
| 1116 | free_recover_size(ls); |
| 1117 | set_bit(DFL_NO_DLM_OPS, &ls->ls_recover_flags); |
| 1118 | return 0; |
| 1119 | } |
| 1120 | |
| 1121 | if (!test_bit(SDF_NOJOURNALID, &sdp->sd_flags)) { |
| 1122 | fs_err(sdp, "dlm lockspace ops disallow jid preset\n"); |
| 1123 | error = -EINVAL; |
| 1124 | goto fail_release; |
| 1125 | } |
| 1126 | |
| 1127 | /* |
| 1128 | * control_mount() uses control_lock to determine first mounter, |
| 1129 | * and for later mounts, waits for any recoveries to be cleared. |
| 1130 | */ |
| 1131 | |
| 1132 | error = control_mount(sdp); |
| 1133 | if (error) { |
| 1134 | fs_err(sdp, "mount control error %d\n", error); |
| 1135 | goto fail_release; |
| 1136 | } |
| 1137 | |
| 1138 | ls->ls_first = !!test_bit(DFL_FIRST_MOUNT, &ls->ls_recover_flags); |
| 1139 | clear_bit(SDF_NOJOURNALID, &sdp->sd_flags); |
| 1140 | smp_mb__after_clear_bit(); |
| 1141 | wake_up_bit(&sdp->sd_flags, SDF_NOJOURNALID); |
| 1142 | return 0; |
| 1143 | |
| 1144 | fail_release: |
| 1145 | dlm_release_lockspace(ls->ls_dlm, 2); |
| 1146 | fail_free: |
| 1147 | free_recover_size(ls); |
| 1148 | fail: |
| 1149 | return error; |
| 1150 | } |
| 1151 | |
| 1152 | static void gdlm_first_done(struct gfs2_sbd *sdp) |
| 1153 | { |
| 1154 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 1155 | int error; |
| 1156 | |
| 1157 | if (test_bit(DFL_NO_DLM_OPS, &ls->ls_recover_flags)) |
| 1158 | return; |
| 1159 | |
| 1160 | error = control_first_done(sdp); |
| 1161 | if (error) |
| 1162 | fs_err(sdp, "mount first_done error %d\n", error); |
| 1163 | } |
| 1164 | |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 1165 | static void gdlm_unmount(struct gfs2_sbd *sdp) |
| 1166 | { |
| 1167 | struct lm_lockstruct *ls = &sdp->sd_lockstruct; |
| 1168 | |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 1169 | if (test_bit(DFL_NO_DLM_OPS, &ls->ls_recover_flags)) |
| 1170 | goto release; |
| 1171 | |
| 1172 | /* wait for gfs2_control_wq to be done with this mount */ |
| 1173 | |
| 1174 | spin_lock(&ls->ls_recover_spin); |
| 1175 | set_bit(DFL_UNMOUNT, &ls->ls_recover_flags); |
| 1176 | spin_unlock(&ls->ls_recover_spin); |
| 1177 | flush_delayed_work_sync(&sdp->sd_control_work); |
| 1178 | |
| 1179 | /* mounted_lock and control_lock will be purged in dlm recovery */ |
| 1180 | release: |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 1181 | if (ls->ls_dlm) { |
| 1182 | dlm_release_lockspace(ls->ls_dlm, 2); |
| 1183 | ls->ls_dlm = NULL; |
| 1184 | } |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 1185 | |
| 1186 | free_recover_size(ls); |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 1187 | } |
| 1188 | |
| 1189 | static const match_table_t dlm_tokens = { |
| 1190 | { Opt_jid, "jid=%d"}, |
| 1191 | { Opt_id, "id=%d"}, |
| 1192 | { Opt_first, "first=%d"}, |
| 1193 | { Opt_nodir, "nodir=%d"}, |
| 1194 | { Opt_err, NULL }, |
| 1195 | }; |
| 1196 | |
| 1197 | const struct lm_lockops gfs2_dlm_ops = { |
| 1198 | .lm_proto_name = "lock_dlm", |
| 1199 | .lm_mount = gdlm_mount, |
David Teigland | e0c2a9a | 2012-01-09 17:18:05 -0500 | [diff] [blame] | 1200 | .lm_first_done = gdlm_first_done, |
| 1201 | .lm_recovery_result = gdlm_recovery_result, |
Steven Whitehouse | f057f6c | 2009-01-12 10:43:39 +0000 | [diff] [blame] | 1202 | .lm_unmount = gdlm_unmount, |
| 1203 | .lm_put_lock = gdlm_put_lock, |
| 1204 | .lm_lock = gdlm_lock, |
| 1205 | .lm_cancel = gdlm_cancel, |
| 1206 | .lm_tokens = &dlm_tokens, |
| 1207 | }; |
| 1208 | |